Research ArticleLocating Station of One-Way Carsharing Based on SpatialDemand Characteristics
Xiaohong Chen1 Jiaqi Cheng 1 Jianhong Ye 1 Yong Jin2 Xi Li2 and Fei Zhang2
1Key Laboratory of Road and Traffic Engineering of the Ministry of Education Tongji University 4800 Caorsquoan RoadShanghai 201804 China2Global Carsharing amp Rental Co Ltd 888 South Moyu Road Shanghai 201805 China
Correspondence should be addressed to Jianhong Ye yjh1875hotmailcom
Received 16 December 2017 Revised 26 March 2018 Accepted 15 April 2018 Published 23 May 2018
Academic Editor Emanuele Crisostomi
Copyright copy 2018 Xiaohong Chen et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
This study aims to explore factors that affect carsharing demand characteristics in different time periods based on EVCARDtransaction data which is the largest station-based one-way carsharing program in Shanghai China Monthly usage intensityand degree of usage imbalance are used as proxies of demand This study uses three groups of independent variables carsharingstation attributes built environment (density diversity design and destination accessibility) and transportation facilities Theadaptive elastic net regression is developed to identify factors that influence carsharing usage intensity and degree of usageimbalance after factor selection using extra-randomized-tree algorithm Finally a station layout is proposed according to bothusage intensity and degree of imbalance The main results of this study are presented as follows (1) different effects of builtenvironment and transportation factors cause dynamic demand across different time periods (2) factors with positive and negativeeffect on carsharing demand are divided clearly for guidance of the carsharing station layout (3) public parking space leads to morepersonal vehicle trip compared to a carsharing trip and (4) as public transportation the relationship of the metro and carsharingis complementary However the bus stop and carsharing have a competitive relationship This study provides a carsharing layoutmethod based on both usage intensity and degree of imbalance Furthermore several policies concerning carsharing are proposed
1 Introduction
Carsharing service allows users to avoid the money and timecost of car ownership such as maintenance insurance andfinding parking space It offers more flexible car travel [1]and has been promptly developed in recent years Manyprevious studies have shown that carsharing service caneffectively reduce vehicle ownership [2 3] reduce the vehiclemiles traveled [4 5] and control the greenhouse effect [67] particularly when some carsharing uses electric vehicles[8] Carsharing can improve social equality and provide aconvenient trip option to people who cannot afford a car[9] These benefits for the society rely on the scale of thecarsharing programand the program sustainability which aredependent on the profit and cost situation of carsharing
The mode of operation is primarily considered basedon profit and cost from the demand feature situations
Carsharing has two operation categories namely station-based carsharing (SBC) and free-floating carsharing (FFC)Moreover there are two types of SBC The first type requiresreturning the rented vehicle to the same station which iscalled round-trip carsharing (RTC) The second type allowsusers to drop off the vehicle at any station after finishing theirtrip which is called one-way carsharing (OWC) HoweverFFC allows users to pick up and drop off the vehicle at anylocation within a specified service area
In RTC users eventually drop off vehicles to the pick-upstation which does not require dispatch in normal operationOWC improves the flexibility of using the vehicle comparedto RTC while bringing extra operational burden Given thatdropping off at another station is allowed there is a highprobability of a spatial imbalance over the stations Theoperator needs to arrange the vehicle relocation to maintainregular operation which results in increased operation cost
HindawiJournal of Advanced TransportationVolume 2018 Article ID 5493632 16 pageshttpsdoiorg10115520185493632
2 Journal of Advanced Transportation
OWC and FFC allow one-way trip which creates theproblem of spatiotemporal imbalance of demand This prob-lem can be handled using different approaches FFC canonly maintain system balance by dispatching techniquesHowever OWC is a station-based system therefore it canmitigate the problem of demand imbalance by appropriatelocation of stations Supposing that there is a more demand-balanced area the pick-up and drop-off at a station can becounteracted during a given temporal intervalTherefore thelayout of stations in the more balanced areas can strategicallyreduce vehicle relocation In the long term it can reducedispatching cost Therefore it is essential to find a small-scale area with more balanced demand in OWC Howeverthe authors found that studies focusing on this issue are rareThus this research will address this gap in literature
Given that station usage intensity is related to the opera-torrsquos operational profits and station usage imbalance is relatedto operational cost finding areas with higher carsharingusage intensity but lower degree of station usage imbalance issignificantly important to the station-based OWC programTherefore this research uses station usage intensity anddegree of station usage imbalance as proxies to representthe demand characteristics to explore how factors affect thecarsharing demand characteristics in different time period ofthe day
Section 2 reviews the literature related to the identifica-tion of factors affecting the demand characteristics in the car-sharing field Section 3 describes the data profile and developsthe monthly usage intensity model and usage imbalancemodel Section 4 presents the result of the models and thendiscusses the factors affecting the demand characteristicsof carsharing and provides result of station layout FinallySection 5 provides a summary and the limitations of thisstudy
2 Literature Review
The research on carsharing has increased rapidly since 2010[10] The current document mainly contains seven subfieldsthe usersrsquo attitude and behavior demand analysis depotlocation vehicle relocation benefit evaluation businessmod-els and policy Two issues are included in the demandanalysis (1) user characteristics or potential demand and (2)identification of factors that influence the current demandThis work aims to address the second issue
Table 1 shows the current references based on analysisunits methods source of variable data dependent variablesand the division stage according to the operating model
The current references mainly focus on RTC [11ndash16] andFFC [17ndash22] but the study on OWC is insufficient In basicanalysis unit the user is always selected for the RTC model[11ndash13]The spatial scales of a single station the station clusterto land parcel (including grid census tracts transportationanalysis zone and district) are in increasing order [14ndash24]The smaller the spatial scale of the analysis unit is the greaterthe model can reflect the demand characteristics of the localspace However the negative effect is the larger proportionof unexplained variance [23] When the amount of datacan ensure that the proportion of random change cannot
obviously affect the model fitting the analysis should beconducted using smaller space scale The study using vehicleas an analysis unit mainly focuses on analyzing the usage rateof different types of vehicles [9]
Given the privacy protection and business sensitivityearly researchers cannot obtain operational data but discon-tinuously collect some sparse data [14 15] These researchstudies assume that operators can dispatch a vehicle fullyaccording to demand where the vehicle number within thespace unit can be used as proxy of the demand in that spatialunit [23] Studies based on transaction data and operationrecord log increased However given that the demand isrestrained by insufficient supply the study on the demand canonly be conducted using actual usage as proxy of the demandwhich is more reliable than studies using supply as proxy fordemand
Dependent variables can be divided into usage frequencyand usage duration which are common dimensions ofdemand characteristics Most studies take usage frequencyand its various variants (such as vehicle usage rate) asdependent variables These studies take booking quantity asusage frequency [9 11ndash13 15 16 19 21] but do not considerpick-up and drop-off mainly because the pick-up and drop-off of RTC are in the same station On the contrary there isno station in FFC In OWC the station not only generatescarsharing trip but also is attracting it as well Completinga carsharing trip based on two stations requires a separatepick-up and drop-off In addition vehicles of RTC belongto one station therefore the effects of supply on demandcan be eliminated by controlling the variable of the vehiclenumber The vehicle usage rate can be used as dependentvariable [14] Without station in FFC one method to obtainusage frequency is by calculating the bookings in area unitwhich makes booking density the most common variable inFFC [17 20ndash22] including usage duration However no studyhas been found on station usage imbalance
Only few research studies use a simple method ofdescriptive statistics [11] similar to the method used inthis study Most studies directly use multiple linear models[12 14ndash17 19 23 24] while other studies exchange naturallogarithms for dependent variables to make them in normaldistribution [23] Given that dependent variables meet withother exponential family distribution the generalized linearmodel is used [12] Moreover given that an independentvariable has large zero value the zero-inflated model [13 20]is used The Dynamic Ordered Probit Model is used becausedecisions of multiple frequencies of usage are cumulativelyrelated to each other [13]
In summary the research on station-based OWC is rareand this work will focus on this operational mode Moreoverfor the dependent variable the station usage intensity usedin this study comprehensively considers the demands forpick-up and drop-off However the degree of station usageimbalance has not yet been studied Finally the analysis ofdemand characteristics across different time period is notsufficient in the current literature In this study time isdivided into several parts and each time section of the dayis modeled individually
Journal of Advanced Transportation 3
Table 1 Summary of document on identification of factors affecting demand characteristics
Item Category Literature
Operational modeRTC [11ndash16]FFC [17ndash22]
NEM (not explicitly mentioned) [9 23 24]
Basic analysis unit
User (RTC) [11ndash13]Single station (RTC) [14]Station cluster (NEM) [23]
Data sources of dependent variable Collected by authors [14 15]Transaction data or operation log [9 11ndash13 16ndash24]
Dependent variable
Usage frequency [RTC and FFC] [11ndash13 16 19 21]Total vehicle hour travel [RTC] [12]Monthly hours per vehicle [RTC] [14]Vehicle usage rate [RTC and NEM] [9 15]
Booking density [FFC] [17 20ndash22]Number of vehicles [NEM] [24]
Vehicle unused duration [FFC] [18]Usage hours of station [NEM] [23]
Method
Descriptive statistics [11]Generalized linear model [9 12 20 21]Multiple linear model [12 14ndash17 19 23 24]
Dynamic ordered probit model [13]Duration model technique [18]
Hot spots by ArcGIS [22]Develop individually models for several time periods of the day [15 21 22]
3 Data and Methodology
31 Data Preparation The four data sources used in thisstudy are shown in Figure 3 The first data source is providedby the Shanghai EVCARD which includes station infor-mation and transaction data starting from January 2015 toDecember 2016 EVCARD is the largest carsharing programin China It is a station-based one-way system which allowsusers to drop off car at any station after a trip All vehiclesin operation are Pure Battery Electric Vehicles (PBEVs)and the price is only based on the trip duration The usageprocedure is completely self-service Station informationincludes station coordinates station age (number of monthsfrom the start of operation until December 2016) parkingspace if a station is located at an underground garage (thisis related to whether or not the station is easy to find) andif a station has limited accessibility because it is located atan internal location of companies or government departmentwhere nonemployees are not allowed to enter for a specifictime period or permanently Other stations nearby can becalculated based on the coordinates data using GeographicInformation System (GIS) software A station includes severalparking lots (from two to dozens) Transaction data includesmembership number vehicle license plate number timeand station name for pick-up and drop-off trip duration
fee deductions of the booking fee and the reason for thereduction Given that the data are commercially sensitivethese data are only permitted to display directly a limited partof the operator request
The second data source is composed of Point of Interest(POI) data in Shanghai which was collected through AMAPAPI inNovember 2016 AMAP is a Google-Maps-like E-mapwhich is popular andwidely used inChinaThe total collectedPOIs reached 427210 which are categorized into eleventypes based on land use attributes residential area cultureplaces business medical service public authority socialand recreation shopping college and university industrytransportation and intersections Among these types collegeand university is used instead of a larger type of educationbecause in addition to being found in preresearch othereducational areas such as middle school primary schoolkindergarten and training institution have no statisticallysignificant effect on the demand characteristics Howevercollege and university plays an important role therefore itreplaces the educational typeThe transportation type is veryimportant therefore it is separately categorized into morespecific subtypes including airport train station intercitycoach station metro station bus stop and traditional carrental station Intersection POIs are used to represent thestreet design Moreover POI-mixed entropy is calculated
4 Journal of Advanced Transportation
based on POIs which refers to the entropy form of research[25]
The third data source is part of the latest census dataof Shanghai which was collected in 2010 by the ShanghaiMunicipal Bureau of Statistics Resident density job densityand average trip distance of each block were used in thisresearch Average trip distance is used to represent theaccessibility of a destination
Finally the fourth data source is the information of roadnetwork in Shanghai which includes freeway arterial roadsecondary road local street one-way road and two-wayroad We counted the length of each type of road in spatialunit to represent characteristics of local road network
The four data sources are reorganized into three groupsof features The first feature is the stationrsquos attributes (119878) Thesecond feature is the transportation facilities The number ofeach type of transportation facilities located within the bufferof the carsharing station is used to present the transportationenvironment within the spatial unit The third group offeatures refers to the 5119863 principle of the built environment[26 27] However the fifth ldquo119863rdquo (distance to transit) ispresented as the distance between a residential place or workplace to the nearest public transportation station whichis not appropriate for this research Therefore we use thetransportation facilities instead of the ldquodistance to transitrdquoThus independent variables shall be categorized and groupedas ldquo119878+119879+4119863rdquoThedescriptive statistics of features and targetsare summarized in Table 2
For sample selection stable operational stations areselected in this study Many stations have short operationperiod with data that ended on 10 December 2016 Thesestations are not well known to users therefore they arenot included in this research because of their unstableperformance Data indicate that the use of vehicles becomesrelatively stable when the station is operated into the fourthmonth (the booking changing rate is about 05 The bookingchanging rate reaches 71 in the second month and 078 inthe third month) In this research we only include stationswithmore than threemonths of operation while the bookingchanging rate is less than 03 in three consecutive monthsfromOctober to December 2016These stations are judged asrelatively stable Eventually 551 stations participated in thisresearch Figure 1 shows the spatial distribution of all stationsoperating in Shanghai as well as the stations participating inthis research The proportion of stations participating in thisresearch is the same as the proportion of the whole stationat each area The central areas have less than 4 stations andparking spaces while 95 of the stations are located at theoutskirts area We divided Shanghai into a hexagonal gridwhere the area of the hexagonal spatial unit is 2 km2 (similarto a circle with 800-meter radius in the area) The stationsparticipating in this research are assigned to this hexagon unitbased on spatial location The hexagonal spatial unit is usedas an analysis unit in this research
Since the EVCARD station as basic object in thisresearch is selected according to the fact that bookingschanging rate is relatively stable during the three monthsfrom November to December 2016 the dependent variable
in this research also requires the relative stability of stationoperation performance where the collecting time of POI datashall be within this time range Therefore only the three-month transaction data from October to December 2016 isinvestigated which covers about 40 of the total bookings
311 Temporal Dynamics of Demand To investigate thetemporal dynamics of demand the data is divided into twosubsets workday and nonworkday In China given thatadjustment is made when national holidays overlap week-days some weekends become workdays and some workdaysbecome holidays Therefore workday and nonworkday areused to separate time instead of weekday and weekend
Based on these conditions the two subsets of data aredivided into five time sections of the day which are as followsaccording to the trip temporal distribution of the EVCARD(Figure 2) early morning (EM) (200ndash600 4 h) morningpeak (MP) (600ndash1000 4 h) off-peak (OP) (1000ndash16006 h) evening peak (EP) (1600ndash2000 4 h) and night (NT)(2000ndash200 6 h) Given the few bookings in EM which arequite random the later four time sections are included in thisresearch Finally eight time sections were obtained and thesetime sections are mathematically expressed as follows
where 119860 = workday nonworkday and 119861 = MP OP EPNT32 Methodology Two dependent variables are used as prox-ies representing demand monthly usage intensity and degreeof usage imbalance Twomachine-learning models are estab-lished in this research The monthly usage intensity modelis used to evaluate the effect of factors influencing usageintensity of exclusive parking space in the analysis spatialunits per month The usage imbalance model is developed toestimate the factors affecting the degree of imbalance of pick-up and drop-off at spatial units The research framework isshown in Figure 3
321 Features Selection Extremely randomized trees algo-rithm (ET) is used to select the important features to avoidoverfitting The extra-trees algorithm builds an ensembleregression trees Its two main differences from other tree-based ensemblemethods such as random forest [28] are thatit divides the nodes by choosing the cut-points completelyat random and that it uses the complete learning sampleinstead of a bootstrap replica to grow the trees The completeextra-trees algorithm is described in [29] Relative variancereduction is used to denote the goodness of point splittingFor a sample 119878 and a splitting point s (a feature selectedrandomly) the goodness of point splitting is expressed asfollows
119866 (119904 119878)= var 119910 | 119878 minus (10038161003816100381610038161198781198971003816100381610038161003816 |119878|) var 119910 | 119878119897 minus (10038161003816100381610038161198781199031003816100381610038161003816 |119878|) var 119910 | 119878119903
var 119910 | 119878 (2)
where var119910 | 119878 is the mean squared error of output 119910 in thesample 119878 119878119897 and 119878119903 are two subsets of sample 119878
Journal of Advanced Transportation 5
Table 2 Descriptive statistics of variables
Variable type Variable name Abbr Variable type Mean Std Min Med Max
Usage intensity
Workday 06ndash09 h I W0609 Numerical 19013 26016 0 105 2879Workday 10ndash15 h I W1015 Numerical 27126 41904 2 152 5287Workday 16ndash19 h I W1619 Numerical 28009 48029 2 140 6214Workday 20ndash01 h I W2001 Numerical 28528 43063 0 114 3987
Nonworkday 06ndash09 h I NW0609 Numerical 6113 8656 0 32 948Nonworkday 10ndash15 h I NW1015 Numerical 14921 23139 0 80 2962Nonworkday 16ndash19 h I NW1619 Numerical 11131 21055 0 54 2925Nonworkday 20ndash01 h I NW2001 Numerical 1107 17335 0 44 1860
Stations Participating in ResearchAll Station in Shanghai
Figure 1 Spatial distribution of stations in Shanghai
012345678
200
300
400
500
600
700
800
900
100
011
00
120
013
00
140
015
00
160
017
00
180
019
00
200
021
00
220
023
00
000
100
Book
ing
Perc
enta
ge
Time
Work DayNon-work Day
02ndash05 h
06ndash09 h10ndash15 h
16ndash19 h
20ndash01 h
Figure 2 EVCARD trip temporal distribution and time division
Let 119866119894(119904 119878) be the goodness of 119894th ET the featureimportance is the average goodness of each ET which can beexpressed as follows
FI = 1119873119873sum119894=1
119866119894 (119904 119878) (3)
where FI denotes the feature importance and119873 is the numberof ET
Unimportant features have little to no effect on themean squared error model while important features shouldsignificantly decrease it
322 Monthly Usage Intensity Model and Usage ImbalanceModel Considering the unequal duration of time sections120591119879 can be the duration of each 119879 where the unit is hour
Journal of Advanced Transportation 7
Exclusive Parking SpaceOperational Age
Limited-Access Parking SpaceUnderground-Garage Parking Space
POI-Mixed Entropy
Residential Public Authority
College and University
Medical Service
Culture
Business
Social and Recreation
Industry
Intersection Density
Average Trip Distance
Train Station
Intercity Coach Station
Metro
Bus Stop
Car Rental StationMonthly usage intensity
Imbalance degree
Demand Estimation
Station Location Vehicle Relocation
support support
inputFeature selection
Freeway Length
Road Network Density
Shopping
Arterial Road LengthSecondary Road LengthLocal Street LengthOne-way lengthTwo-way length
Local Road Network Feature
Public Parking Space
Built Environment
Density
Diversity
Design
DestinationAccessibility
Transportation
Area Attributes
Feat
ures
Transaction data
EvcardDatabase
Targets
Monthly usageintensity model
Usage imbalance model
Figure 3 Research framework
Usage intensity is defined as average value per hour of the totalamount for pick-up and drop-off in specific time sections ofthe spatial unit The calculation formula is shown as follows
where 119868 represents the usage intensity and 119901119879 and 119889119879 are theamount of pick-up and drop-off at specific time-interval 119897means 119897 monthsrsquo transaction data is involved in this research119897 is 3 here
We use usage intensity as a proxy of demand rather thanthe number of bookings because the station is viewed as bothorigin and destination for the carsharing trip It not onlygenerates carsharing trip but also attracts it Therefore usageintensity is an appropriate index for carsharing demand
Usage imbalance degree is defined as the ratio of thedifference between pick-up and drop-off to the sum of pick-up and drop-off in specific time sections of the spatial unitHalf an hour is used as the statistic time interval then thedifference between pick-up and drop-off of each statisticinterval is aggregated by corresponding 119879 time section andfurther divided with the sum of pick-up and drop-off in timesection T
where 119875119879 and 119863119879 are the pick-up matrix and drop-off matrixin time section 119879 with half-hour statistic interval
IM119879 = sum119898119879119894=1sum119899119879119895=1 10038161003816100381610038161003816119901119894119895 minus 11988911989411989510038161003816100381610038161003816sum119898119879119894=1sum119899119879119895=1 119901119894119895 + 119889119894119895 (6)
where IM119879 is the imbalance degree of the spatial unit 119898119879 isthe amount of day in time section 119879 119899119879 is the statistic unitin time section 119879 and 119901119894119895 and 119889119894119895 are the pick-up and drop-off quantity in statistic unit of 119894th day and 119895th day respec-tively
Given that several types of POIs coexist in the samearea with varying degrees features have multiple collinearityproblems that cannot be ignored Meanwhile there are manyfeatures for samplesTherefore a linear regressionmodelwith1198711 and 1198712 prior as a regularizer called adaptive elastic net(AEN) regression [30] was developed to predict the usageintensity and the degree of imbalance The method can beviewed as a combination of elastic net [31] and the adaptiveleast absolute shrinkage and selection operator (LASSO) [32]which overcome the lack of adaptive LASSO (instability for
8 Journal of Advanced Transportation
high-dimensional data) and lack of the oracle property for theelastic net The AEN is defined as follows
(AEN) = (1 + 1205822119899 )
sdot argmin120573
1003817100381710038171003817119910 minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 120582lowast1119901sum119895=1
1003817100381710038171003817y minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 1205821 100381710038171003817100381712057310038171003817100381710038171 (8)
where (EN) is an elastic net algorithm 119901 denotes thenumber of features 1205731 = sum119901119895=1 |120573|1 is the 1198971-norm and 12057322is 1198972-norm 1205821 1205822 and 120582lowast1 are weights for the 1198971-norm 1198972-norm and optimal 1205821 respectively 119908119895119901119895=1 are the adaptivedata-driven weights
4 Results and Discussion
A total of 1500 ETs are used to estimate the feature importanceso that the ranks and scores of featuresrsquo importance are stableWe drop the features with importance score not greater than0015The information of featuresrsquo importance indicating thatthe contribution of each feature to prediction is provided inFigures 4 and 5 The features that are not filtered are used forbuilding prediction models The results of models are shownin Tables 3 and 4
41 Usage Intensity Model The result shows that 1198772 ofintensity models for eight time sections are between 0404and 0583 The goodness of fit is better than other researcheson this issue [15 23 24] Some features show the samedirection of influence on usage intensity However otherfactors play a positive or negative effect on the demanddepending on the time period Meanwhile all factors havedifferent weights for demand across different time periodsThis causes the usage intensity to be different across thewholeday as shown in Figure 2
411 Station-Related Factors The operational attributes ofthe spatial unit play an important role in usage intensityacross all periods The longer the first station operates inspatial unit the greater the intensity is because the serviceand location of a station are getting familiarized by usersgradually Limited-access parking space has constantly pro-vided negative effect on usage intensity Exclusive parkingspace represents the supply level partly It positively affectsusage intensity However these factors are endogenous itemsand the carsharing operator can improve these factors aspossible as it can The more important factors are exogenous
variables such as the built environment and transportationin the spatial unit
412 Built Environment Factors Built environment factorsshow the diverse effect on usage intensity in different timeperiods College and university constantly has positive influ-ence In contrast the industrial area shows negative effectResidential culture public authority and medical hygienearea impose negative influence on usage intensity in specialtime period Other factors have opposite effect depending onthe time period Recreation area has a negative influence onusage intensity in workdayrsquos early peak and positive influenceduring nonworking time sections More POI-related shop-ping is within the spatial unit and usage intensity showsmoreincrease in evening peak and night
POI-mixed entropy plays a positive role in the usageintensity during working day at around 1600ndash200 Theaverage trip distance is a special factor It has positive effect onthe evening peak of working days and night of nonworkingdays which implies that users might tend to use carsharingfor further distance trip during these periods Additionallyit has negative effect during 600ndash1000 in nonworking dayswhich indicates that short distance trip of carsharing tends tobe at 600ndash1000 in nonworking days
Areas with higher intersection density mainly duringnonworking time sections improve usage intensity becauseof good accessibility It is unexpected that the length of thelocal street and one-way road generally has negative effectand two-way road has a positive effect on usage intensityGiven that the local street and one-way street are more walk-friendly the result is opposite to some research This findingmay be attributed to our use of intensity which includespick-up and drop-off instead of pick-up only to count as anindicator of demand Many local streets and one-way streetswould make it more difficult for users to find parking spaceIn contrast more two-way roads but less one-way and localstreets means simpler road network
413 Transportation Factors Considering the transportationfactors unexpectedly traditional car rental station has apositive effect on station usage intensity at 600ndash1600 onworking days and 1000ndash2000 on nonworking days Despitebeing shown in literature that car rental and carsharing havecompetitive relationship in medium distance trip [1] thisresult indicates a more complicated relationship betweencarsharing and car rental Moreover intercity coach stationhas key role and positive relation during 600ndash1600 onnonworking days which has not been reported in any currentliterature Intercity coach stations are generally far from thecenter of the city and passengers have no personal car whiletaking some packages which could be themain reason for thedemand in using carsharing to connect with intercity coach
For public transportation the existence of metro stationnegatively affects carsharing station usage intensity duringmorning peak and evening peak on working days Thisfinding could be attributed to the belief that the metro ismore reliable for commuting compared to ground trafficand commute to work is time-limited Meanwhile the metroimposes positive effect during nonworkday which implies
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
2 Journal of Advanced Transportation
OWC and FFC allow one-way trip which creates theproblem of spatiotemporal imbalance of demand This prob-lem can be handled using different approaches FFC canonly maintain system balance by dispatching techniquesHowever OWC is a station-based system therefore it canmitigate the problem of demand imbalance by appropriatelocation of stations Supposing that there is a more demand-balanced area the pick-up and drop-off at a station can becounteracted during a given temporal intervalTherefore thelayout of stations in the more balanced areas can strategicallyreduce vehicle relocation In the long term it can reducedispatching cost Therefore it is essential to find a small-scale area with more balanced demand in OWC Howeverthe authors found that studies focusing on this issue are rareThus this research will address this gap in literature
Given that station usage intensity is related to the opera-torrsquos operational profits and station usage imbalance is relatedto operational cost finding areas with higher carsharingusage intensity but lower degree of station usage imbalance issignificantly important to the station-based OWC programTherefore this research uses station usage intensity anddegree of station usage imbalance as proxies to representthe demand characteristics to explore how factors affect thecarsharing demand characteristics in different time period ofthe day
Section 2 reviews the literature related to the identifica-tion of factors affecting the demand characteristics in the car-sharing field Section 3 describes the data profile and developsthe monthly usage intensity model and usage imbalancemodel Section 4 presents the result of the models and thendiscusses the factors affecting the demand characteristicsof carsharing and provides result of station layout FinallySection 5 provides a summary and the limitations of thisstudy
2 Literature Review
The research on carsharing has increased rapidly since 2010[10] The current document mainly contains seven subfieldsthe usersrsquo attitude and behavior demand analysis depotlocation vehicle relocation benefit evaluation businessmod-els and policy Two issues are included in the demandanalysis (1) user characteristics or potential demand and (2)identification of factors that influence the current demandThis work aims to address the second issue
Table 1 shows the current references based on analysisunits methods source of variable data dependent variablesand the division stage according to the operating model
The current references mainly focus on RTC [11ndash16] andFFC [17ndash22] but the study on OWC is insufficient In basicanalysis unit the user is always selected for the RTC model[11ndash13]The spatial scales of a single station the station clusterto land parcel (including grid census tracts transportationanalysis zone and district) are in increasing order [14ndash24]The smaller the spatial scale of the analysis unit is the greaterthe model can reflect the demand characteristics of the localspace However the negative effect is the larger proportionof unexplained variance [23] When the amount of datacan ensure that the proportion of random change cannot
obviously affect the model fitting the analysis should beconducted using smaller space scale The study using vehicleas an analysis unit mainly focuses on analyzing the usage rateof different types of vehicles [9]
Given the privacy protection and business sensitivityearly researchers cannot obtain operational data but discon-tinuously collect some sparse data [14 15] These researchstudies assume that operators can dispatch a vehicle fullyaccording to demand where the vehicle number within thespace unit can be used as proxy of the demand in that spatialunit [23] Studies based on transaction data and operationrecord log increased However given that the demand isrestrained by insufficient supply the study on the demand canonly be conducted using actual usage as proxy of the demandwhich is more reliable than studies using supply as proxy fordemand
Dependent variables can be divided into usage frequencyand usage duration which are common dimensions ofdemand characteristics Most studies take usage frequencyand its various variants (such as vehicle usage rate) asdependent variables These studies take booking quantity asusage frequency [9 11ndash13 15 16 19 21] but do not considerpick-up and drop-off mainly because the pick-up and drop-off of RTC are in the same station On the contrary there isno station in FFC In OWC the station not only generatescarsharing trip but also is attracting it as well Completinga carsharing trip based on two stations requires a separatepick-up and drop-off In addition vehicles of RTC belongto one station therefore the effects of supply on demandcan be eliminated by controlling the variable of the vehiclenumber The vehicle usage rate can be used as dependentvariable [14] Without station in FFC one method to obtainusage frequency is by calculating the bookings in area unitwhich makes booking density the most common variable inFFC [17 20ndash22] including usage duration However no studyhas been found on station usage imbalance
Only few research studies use a simple method ofdescriptive statistics [11] similar to the method used inthis study Most studies directly use multiple linear models[12 14ndash17 19 23 24] while other studies exchange naturallogarithms for dependent variables to make them in normaldistribution [23] Given that dependent variables meet withother exponential family distribution the generalized linearmodel is used [12] Moreover given that an independentvariable has large zero value the zero-inflated model [13 20]is used The Dynamic Ordered Probit Model is used becausedecisions of multiple frequencies of usage are cumulativelyrelated to each other [13]
In summary the research on station-based OWC is rareand this work will focus on this operational mode Moreoverfor the dependent variable the station usage intensity usedin this study comprehensively considers the demands forpick-up and drop-off However the degree of station usageimbalance has not yet been studied Finally the analysis ofdemand characteristics across different time period is notsufficient in the current literature In this study time isdivided into several parts and each time section of the dayis modeled individually
Journal of Advanced Transportation 3
Table 1 Summary of document on identification of factors affecting demand characteristics
Item Category Literature
Operational modeRTC [11ndash16]FFC [17ndash22]
NEM (not explicitly mentioned) [9 23 24]
Basic analysis unit
User (RTC) [11ndash13]Single station (RTC) [14]Station cluster (NEM) [23]
Data sources of dependent variable Collected by authors [14 15]Transaction data or operation log [9 11ndash13 16ndash24]
Dependent variable
Usage frequency [RTC and FFC] [11ndash13 16 19 21]Total vehicle hour travel [RTC] [12]Monthly hours per vehicle [RTC] [14]Vehicle usage rate [RTC and NEM] [9 15]
Booking density [FFC] [17 20ndash22]Number of vehicles [NEM] [24]
Vehicle unused duration [FFC] [18]Usage hours of station [NEM] [23]
Method
Descriptive statistics [11]Generalized linear model [9 12 20 21]Multiple linear model [12 14ndash17 19 23 24]
Dynamic ordered probit model [13]Duration model technique [18]
Hot spots by ArcGIS [22]Develop individually models for several time periods of the day [15 21 22]
3 Data and Methodology
31 Data Preparation The four data sources used in thisstudy are shown in Figure 3 The first data source is providedby the Shanghai EVCARD which includes station infor-mation and transaction data starting from January 2015 toDecember 2016 EVCARD is the largest carsharing programin China It is a station-based one-way system which allowsusers to drop off car at any station after a trip All vehiclesin operation are Pure Battery Electric Vehicles (PBEVs)and the price is only based on the trip duration The usageprocedure is completely self-service Station informationincludes station coordinates station age (number of monthsfrom the start of operation until December 2016) parkingspace if a station is located at an underground garage (thisis related to whether or not the station is easy to find) andif a station has limited accessibility because it is located atan internal location of companies or government departmentwhere nonemployees are not allowed to enter for a specifictime period or permanently Other stations nearby can becalculated based on the coordinates data using GeographicInformation System (GIS) software A station includes severalparking lots (from two to dozens) Transaction data includesmembership number vehicle license plate number timeand station name for pick-up and drop-off trip duration
fee deductions of the booking fee and the reason for thereduction Given that the data are commercially sensitivethese data are only permitted to display directly a limited partof the operator request
The second data source is composed of Point of Interest(POI) data in Shanghai which was collected through AMAPAPI inNovember 2016 AMAP is a Google-Maps-like E-mapwhich is popular andwidely used inChinaThe total collectedPOIs reached 427210 which are categorized into eleventypes based on land use attributes residential area cultureplaces business medical service public authority socialand recreation shopping college and university industrytransportation and intersections Among these types collegeand university is used instead of a larger type of educationbecause in addition to being found in preresearch othereducational areas such as middle school primary schoolkindergarten and training institution have no statisticallysignificant effect on the demand characteristics Howevercollege and university plays an important role therefore itreplaces the educational typeThe transportation type is veryimportant therefore it is separately categorized into morespecific subtypes including airport train station intercitycoach station metro station bus stop and traditional carrental station Intersection POIs are used to represent thestreet design Moreover POI-mixed entropy is calculated
4 Journal of Advanced Transportation
based on POIs which refers to the entropy form of research[25]
The third data source is part of the latest census dataof Shanghai which was collected in 2010 by the ShanghaiMunicipal Bureau of Statistics Resident density job densityand average trip distance of each block were used in thisresearch Average trip distance is used to represent theaccessibility of a destination
Finally the fourth data source is the information of roadnetwork in Shanghai which includes freeway arterial roadsecondary road local street one-way road and two-wayroad We counted the length of each type of road in spatialunit to represent characteristics of local road network
The four data sources are reorganized into three groupsof features The first feature is the stationrsquos attributes (119878) Thesecond feature is the transportation facilities The number ofeach type of transportation facilities located within the bufferof the carsharing station is used to present the transportationenvironment within the spatial unit The third group offeatures refers to the 5119863 principle of the built environment[26 27] However the fifth ldquo119863rdquo (distance to transit) ispresented as the distance between a residential place or workplace to the nearest public transportation station whichis not appropriate for this research Therefore we use thetransportation facilities instead of the ldquodistance to transitrdquoThus independent variables shall be categorized and groupedas ldquo119878+119879+4119863rdquoThedescriptive statistics of features and targetsare summarized in Table 2
For sample selection stable operational stations areselected in this study Many stations have short operationperiod with data that ended on 10 December 2016 Thesestations are not well known to users therefore they arenot included in this research because of their unstableperformance Data indicate that the use of vehicles becomesrelatively stable when the station is operated into the fourthmonth (the booking changing rate is about 05 The bookingchanging rate reaches 71 in the second month and 078 inthe third month) In this research we only include stationswithmore than threemonths of operation while the bookingchanging rate is less than 03 in three consecutive monthsfromOctober to December 2016These stations are judged asrelatively stable Eventually 551 stations participated in thisresearch Figure 1 shows the spatial distribution of all stationsoperating in Shanghai as well as the stations participating inthis research The proportion of stations participating in thisresearch is the same as the proportion of the whole stationat each area The central areas have less than 4 stations andparking spaces while 95 of the stations are located at theoutskirts area We divided Shanghai into a hexagonal gridwhere the area of the hexagonal spatial unit is 2 km2 (similarto a circle with 800-meter radius in the area) The stationsparticipating in this research are assigned to this hexagon unitbased on spatial location The hexagonal spatial unit is usedas an analysis unit in this research
Since the EVCARD station as basic object in thisresearch is selected according to the fact that bookingschanging rate is relatively stable during the three monthsfrom November to December 2016 the dependent variable
in this research also requires the relative stability of stationoperation performance where the collecting time of POI datashall be within this time range Therefore only the three-month transaction data from October to December 2016 isinvestigated which covers about 40 of the total bookings
311 Temporal Dynamics of Demand To investigate thetemporal dynamics of demand the data is divided into twosubsets workday and nonworkday In China given thatadjustment is made when national holidays overlap week-days some weekends become workdays and some workdaysbecome holidays Therefore workday and nonworkday areused to separate time instead of weekday and weekend
Based on these conditions the two subsets of data aredivided into five time sections of the day which are as followsaccording to the trip temporal distribution of the EVCARD(Figure 2) early morning (EM) (200ndash600 4 h) morningpeak (MP) (600ndash1000 4 h) off-peak (OP) (1000ndash16006 h) evening peak (EP) (1600ndash2000 4 h) and night (NT)(2000ndash200 6 h) Given the few bookings in EM which arequite random the later four time sections are included in thisresearch Finally eight time sections were obtained and thesetime sections are mathematically expressed as follows
where 119860 = workday nonworkday and 119861 = MP OP EPNT32 Methodology Two dependent variables are used as prox-ies representing demand monthly usage intensity and degreeof usage imbalance Twomachine-learning models are estab-lished in this research The monthly usage intensity modelis used to evaluate the effect of factors influencing usageintensity of exclusive parking space in the analysis spatialunits per month The usage imbalance model is developed toestimate the factors affecting the degree of imbalance of pick-up and drop-off at spatial units The research framework isshown in Figure 3
321 Features Selection Extremely randomized trees algo-rithm (ET) is used to select the important features to avoidoverfitting The extra-trees algorithm builds an ensembleregression trees Its two main differences from other tree-based ensemblemethods such as random forest [28] are thatit divides the nodes by choosing the cut-points completelyat random and that it uses the complete learning sampleinstead of a bootstrap replica to grow the trees The completeextra-trees algorithm is described in [29] Relative variancereduction is used to denote the goodness of point splittingFor a sample 119878 and a splitting point s (a feature selectedrandomly) the goodness of point splitting is expressed asfollows
119866 (119904 119878)= var 119910 | 119878 minus (10038161003816100381610038161198781198971003816100381610038161003816 |119878|) var 119910 | 119878119897 minus (10038161003816100381610038161198781199031003816100381610038161003816 |119878|) var 119910 | 119878119903
var 119910 | 119878 (2)
where var119910 | 119878 is the mean squared error of output 119910 in thesample 119878 119878119897 and 119878119903 are two subsets of sample 119878
Journal of Advanced Transportation 5
Table 2 Descriptive statistics of variables
Variable type Variable name Abbr Variable type Mean Std Min Med Max
Usage intensity
Workday 06ndash09 h I W0609 Numerical 19013 26016 0 105 2879Workday 10ndash15 h I W1015 Numerical 27126 41904 2 152 5287Workday 16ndash19 h I W1619 Numerical 28009 48029 2 140 6214Workday 20ndash01 h I W2001 Numerical 28528 43063 0 114 3987
Nonworkday 06ndash09 h I NW0609 Numerical 6113 8656 0 32 948Nonworkday 10ndash15 h I NW1015 Numerical 14921 23139 0 80 2962Nonworkday 16ndash19 h I NW1619 Numerical 11131 21055 0 54 2925Nonworkday 20ndash01 h I NW2001 Numerical 1107 17335 0 44 1860
Stations Participating in ResearchAll Station in Shanghai
Figure 1 Spatial distribution of stations in Shanghai
012345678
200
300
400
500
600
700
800
900
100
011
00
120
013
00
140
015
00
160
017
00
180
019
00
200
021
00
220
023
00
000
100
Book
ing
Perc
enta
ge
Time
Work DayNon-work Day
02ndash05 h
06ndash09 h10ndash15 h
16ndash19 h
20ndash01 h
Figure 2 EVCARD trip temporal distribution and time division
Let 119866119894(119904 119878) be the goodness of 119894th ET the featureimportance is the average goodness of each ET which can beexpressed as follows
FI = 1119873119873sum119894=1
119866119894 (119904 119878) (3)
where FI denotes the feature importance and119873 is the numberof ET
Unimportant features have little to no effect on themean squared error model while important features shouldsignificantly decrease it
322 Monthly Usage Intensity Model and Usage ImbalanceModel Considering the unequal duration of time sections120591119879 can be the duration of each 119879 where the unit is hour
Journal of Advanced Transportation 7
Exclusive Parking SpaceOperational Age
Limited-Access Parking SpaceUnderground-Garage Parking Space
POI-Mixed Entropy
Residential Public Authority
College and University
Medical Service
Culture
Business
Social and Recreation
Industry
Intersection Density
Average Trip Distance
Train Station
Intercity Coach Station
Metro
Bus Stop
Car Rental StationMonthly usage intensity
Imbalance degree
Demand Estimation
Station Location Vehicle Relocation
support support
inputFeature selection
Freeway Length
Road Network Density
Shopping
Arterial Road LengthSecondary Road LengthLocal Street LengthOne-way lengthTwo-way length
Local Road Network Feature
Public Parking Space
Built Environment
Density
Diversity
Design
DestinationAccessibility
Transportation
Area Attributes
Feat
ures
Transaction data
EvcardDatabase
Targets
Monthly usageintensity model
Usage imbalance model
Figure 3 Research framework
Usage intensity is defined as average value per hour of the totalamount for pick-up and drop-off in specific time sections ofthe spatial unit The calculation formula is shown as follows
where 119868 represents the usage intensity and 119901119879 and 119889119879 are theamount of pick-up and drop-off at specific time-interval 119897means 119897 monthsrsquo transaction data is involved in this research119897 is 3 here
We use usage intensity as a proxy of demand rather thanthe number of bookings because the station is viewed as bothorigin and destination for the carsharing trip It not onlygenerates carsharing trip but also attracts it Therefore usageintensity is an appropriate index for carsharing demand
Usage imbalance degree is defined as the ratio of thedifference between pick-up and drop-off to the sum of pick-up and drop-off in specific time sections of the spatial unitHalf an hour is used as the statistic time interval then thedifference between pick-up and drop-off of each statisticinterval is aggregated by corresponding 119879 time section andfurther divided with the sum of pick-up and drop-off in timesection T
where 119875119879 and 119863119879 are the pick-up matrix and drop-off matrixin time section 119879 with half-hour statistic interval
IM119879 = sum119898119879119894=1sum119899119879119895=1 10038161003816100381610038161003816119901119894119895 minus 11988911989411989510038161003816100381610038161003816sum119898119879119894=1sum119899119879119895=1 119901119894119895 + 119889119894119895 (6)
where IM119879 is the imbalance degree of the spatial unit 119898119879 isthe amount of day in time section 119879 119899119879 is the statistic unitin time section 119879 and 119901119894119895 and 119889119894119895 are the pick-up and drop-off quantity in statistic unit of 119894th day and 119895th day respec-tively
Given that several types of POIs coexist in the samearea with varying degrees features have multiple collinearityproblems that cannot be ignored Meanwhile there are manyfeatures for samplesTherefore a linear regressionmodelwith1198711 and 1198712 prior as a regularizer called adaptive elastic net(AEN) regression [30] was developed to predict the usageintensity and the degree of imbalance The method can beviewed as a combination of elastic net [31] and the adaptiveleast absolute shrinkage and selection operator (LASSO) [32]which overcome the lack of adaptive LASSO (instability for
8 Journal of Advanced Transportation
high-dimensional data) and lack of the oracle property for theelastic net The AEN is defined as follows
(AEN) = (1 + 1205822119899 )
sdot argmin120573
1003817100381710038171003817119910 minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 120582lowast1119901sum119895=1
1003817100381710038171003817y minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 1205821 100381710038171003817100381712057310038171003817100381710038171 (8)
where (EN) is an elastic net algorithm 119901 denotes thenumber of features 1205731 = sum119901119895=1 |120573|1 is the 1198971-norm and 12057322is 1198972-norm 1205821 1205822 and 120582lowast1 are weights for the 1198971-norm 1198972-norm and optimal 1205821 respectively 119908119895119901119895=1 are the adaptivedata-driven weights
4 Results and Discussion
A total of 1500 ETs are used to estimate the feature importanceso that the ranks and scores of featuresrsquo importance are stableWe drop the features with importance score not greater than0015The information of featuresrsquo importance indicating thatthe contribution of each feature to prediction is provided inFigures 4 and 5 The features that are not filtered are used forbuilding prediction models The results of models are shownin Tables 3 and 4
41 Usage Intensity Model The result shows that 1198772 ofintensity models for eight time sections are between 0404and 0583 The goodness of fit is better than other researcheson this issue [15 23 24] Some features show the samedirection of influence on usage intensity However otherfactors play a positive or negative effect on the demanddepending on the time period Meanwhile all factors havedifferent weights for demand across different time periodsThis causes the usage intensity to be different across thewholeday as shown in Figure 2
411 Station-Related Factors The operational attributes ofthe spatial unit play an important role in usage intensityacross all periods The longer the first station operates inspatial unit the greater the intensity is because the serviceand location of a station are getting familiarized by usersgradually Limited-access parking space has constantly pro-vided negative effect on usage intensity Exclusive parkingspace represents the supply level partly It positively affectsusage intensity However these factors are endogenous itemsand the carsharing operator can improve these factors aspossible as it can The more important factors are exogenous
variables such as the built environment and transportationin the spatial unit
412 Built Environment Factors Built environment factorsshow the diverse effect on usage intensity in different timeperiods College and university constantly has positive influ-ence In contrast the industrial area shows negative effectResidential culture public authority and medical hygienearea impose negative influence on usage intensity in specialtime period Other factors have opposite effect depending onthe time period Recreation area has a negative influence onusage intensity in workdayrsquos early peak and positive influenceduring nonworking time sections More POI-related shop-ping is within the spatial unit and usage intensity showsmoreincrease in evening peak and night
POI-mixed entropy plays a positive role in the usageintensity during working day at around 1600ndash200 Theaverage trip distance is a special factor It has positive effect onthe evening peak of working days and night of nonworkingdays which implies that users might tend to use carsharingfor further distance trip during these periods Additionallyit has negative effect during 600ndash1000 in nonworking dayswhich indicates that short distance trip of carsharing tends tobe at 600ndash1000 in nonworking days
Areas with higher intersection density mainly duringnonworking time sections improve usage intensity becauseof good accessibility It is unexpected that the length of thelocal street and one-way road generally has negative effectand two-way road has a positive effect on usage intensityGiven that the local street and one-way street are more walk-friendly the result is opposite to some research This findingmay be attributed to our use of intensity which includespick-up and drop-off instead of pick-up only to count as anindicator of demand Many local streets and one-way streetswould make it more difficult for users to find parking spaceIn contrast more two-way roads but less one-way and localstreets means simpler road network
413 Transportation Factors Considering the transportationfactors unexpectedly traditional car rental station has apositive effect on station usage intensity at 600ndash1600 onworking days and 1000ndash2000 on nonworking days Despitebeing shown in literature that car rental and carsharing havecompetitive relationship in medium distance trip [1] thisresult indicates a more complicated relationship betweencarsharing and car rental Moreover intercity coach stationhas key role and positive relation during 600ndash1600 onnonworking days which has not been reported in any currentliterature Intercity coach stations are generally far from thecenter of the city and passengers have no personal car whiletaking some packages which could be themain reason for thedemand in using carsharing to connect with intercity coach
For public transportation the existence of metro stationnegatively affects carsharing station usage intensity duringmorning peak and evening peak on working days Thisfinding could be attributed to the belief that the metro ismore reliable for commuting compared to ground trafficand commute to work is time-limited Meanwhile the metroimposes positive effect during nonworkday which implies
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Data sources of dependent variable Collected by authors [14 15]Transaction data or operation log [9 11ndash13 16ndash24]
Dependent variable
Usage frequency [RTC and FFC] [11ndash13 16 19 21]Total vehicle hour travel [RTC] [12]Monthly hours per vehicle [RTC] [14]Vehicle usage rate [RTC and NEM] [9 15]
Booking density [FFC] [17 20ndash22]Number of vehicles [NEM] [24]
Vehicle unused duration [FFC] [18]Usage hours of station [NEM] [23]
Method
Descriptive statistics [11]Generalized linear model [9 12 20 21]Multiple linear model [12 14ndash17 19 23 24]
Dynamic ordered probit model [13]Duration model technique [18]
Hot spots by ArcGIS [22]Develop individually models for several time periods of the day [15 21 22]
3 Data and Methodology
31 Data Preparation The four data sources used in thisstudy are shown in Figure 3 The first data source is providedby the Shanghai EVCARD which includes station infor-mation and transaction data starting from January 2015 toDecember 2016 EVCARD is the largest carsharing programin China It is a station-based one-way system which allowsusers to drop off car at any station after a trip All vehiclesin operation are Pure Battery Electric Vehicles (PBEVs)and the price is only based on the trip duration The usageprocedure is completely self-service Station informationincludes station coordinates station age (number of monthsfrom the start of operation until December 2016) parkingspace if a station is located at an underground garage (thisis related to whether or not the station is easy to find) andif a station has limited accessibility because it is located atan internal location of companies or government departmentwhere nonemployees are not allowed to enter for a specifictime period or permanently Other stations nearby can becalculated based on the coordinates data using GeographicInformation System (GIS) software A station includes severalparking lots (from two to dozens) Transaction data includesmembership number vehicle license plate number timeand station name for pick-up and drop-off trip duration
fee deductions of the booking fee and the reason for thereduction Given that the data are commercially sensitivethese data are only permitted to display directly a limited partof the operator request
The second data source is composed of Point of Interest(POI) data in Shanghai which was collected through AMAPAPI inNovember 2016 AMAP is a Google-Maps-like E-mapwhich is popular andwidely used inChinaThe total collectedPOIs reached 427210 which are categorized into eleventypes based on land use attributes residential area cultureplaces business medical service public authority socialand recreation shopping college and university industrytransportation and intersections Among these types collegeand university is used instead of a larger type of educationbecause in addition to being found in preresearch othereducational areas such as middle school primary schoolkindergarten and training institution have no statisticallysignificant effect on the demand characteristics Howevercollege and university plays an important role therefore itreplaces the educational typeThe transportation type is veryimportant therefore it is separately categorized into morespecific subtypes including airport train station intercitycoach station metro station bus stop and traditional carrental station Intersection POIs are used to represent thestreet design Moreover POI-mixed entropy is calculated
4 Journal of Advanced Transportation
based on POIs which refers to the entropy form of research[25]
The third data source is part of the latest census dataof Shanghai which was collected in 2010 by the ShanghaiMunicipal Bureau of Statistics Resident density job densityand average trip distance of each block were used in thisresearch Average trip distance is used to represent theaccessibility of a destination
Finally the fourth data source is the information of roadnetwork in Shanghai which includes freeway arterial roadsecondary road local street one-way road and two-wayroad We counted the length of each type of road in spatialunit to represent characteristics of local road network
The four data sources are reorganized into three groupsof features The first feature is the stationrsquos attributes (119878) Thesecond feature is the transportation facilities The number ofeach type of transportation facilities located within the bufferof the carsharing station is used to present the transportationenvironment within the spatial unit The third group offeatures refers to the 5119863 principle of the built environment[26 27] However the fifth ldquo119863rdquo (distance to transit) ispresented as the distance between a residential place or workplace to the nearest public transportation station whichis not appropriate for this research Therefore we use thetransportation facilities instead of the ldquodistance to transitrdquoThus independent variables shall be categorized and groupedas ldquo119878+119879+4119863rdquoThedescriptive statistics of features and targetsare summarized in Table 2
For sample selection stable operational stations areselected in this study Many stations have short operationperiod with data that ended on 10 December 2016 Thesestations are not well known to users therefore they arenot included in this research because of their unstableperformance Data indicate that the use of vehicles becomesrelatively stable when the station is operated into the fourthmonth (the booking changing rate is about 05 The bookingchanging rate reaches 71 in the second month and 078 inthe third month) In this research we only include stationswithmore than threemonths of operation while the bookingchanging rate is less than 03 in three consecutive monthsfromOctober to December 2016These stations are judged asrelatively stable Eventually 551 stations participated in thisresearch Figure 1 shows the spatial distribution of all stationsoperating in Shanghai as well as the stations participating inthis research The proportion of stations participating in thisresearch is the same as the proportion of the whole stationat each area The central areas have less than 4 stations andparking spaces while 95 of the stations are located at theoutskirts area We divided Shanghai into a hexagonal gridwhere the area of the hexagonal spatial unit is 2 km2 (similarto a circle with 800-meter radius in the area) The stationsparticipating in this research are assigned to this hexagon unitbased on spatial location The hexagonal spatial unit is usedas an analysis unit in this research
Since the EVCARD station as basic object in thisresearch is selected according to the fact that bookingschanging rate is relatively stable during the three monthsfrom November to December 2016 the dependent variable
in this research also requires the relative stability of stationoperation performance where the collecting time of POI datashall be within this time range Therefore only the three-month transaction data from October to December 2016 isinvestigated which covers about 40 of the total bookings
311 Temporal Dynamics of Demand To investigate thetemporal dynamics of demand the data is divided into twosubsets workday and nonworkday In China given thatadjustment is made when national holidays overlap week-days some weekends become workdays and some workdaysbecome holidays Therefore workday and nonworkday areused to separate time instead of weekday and weekend
Based on these conditions the two subsets of data aredivided into five time sections of the day which are as followsaccording to the trip temporal distribution of the EVCARD(Figure 2) early morning (EM) (200ndash600 4 h) morningpeak (MP) (600ndash1000 4 h) off-peak (OP) (1000ndash16006 h) evening peak (EP) (1600ndash2000 4 h) and night (NT)(2000ndash200 6 h) Given the few bookings in EM which arequite random the later four time sections are included in thisresearch Finally eight time sections were obtained and thesetime sections are mathematically expressed as follows
where 119860 = workday nonworkday and 119861 = MP OP EPNT32 Methodology Two dependent variables are used as prox-ies representing demand monthly usage intensity and degreeof usage imbalance Twomachine-learning models are estab-lished in this research The monthly usage intensity modelis used to evaluate the effect of factors influencing usageintensity of exclusive parking space in the analysis spatialunits per month The usage imbalance model is developed toestimate the factors affecting the degree of imbalance of pick-up and drop-off at spatial units The research framework isshown in Figure 3
321 Features Selection Extremely randomized trees algo-rithm (ET) is used to select the important features to avoidoverfitting The extra-trees algorithm builds an ensembleregression trees Its two main differences from other tree-based ensemblemethods such as random forest [28] are thatit divides the nodes by choosing the cut-points completelyat random and that it uses the complete learning sampleinstead of a bootstrap replica to grow the trees The completeextra-trees algorithm is described in [29] Relative variancereduction is used to denote the goodness of point splittingFor a sample 119878 and a splitting point s (a feature selectedrandomly) the goodness of point splitting is expressed asfollows
119866 (119904 119878)= var 119910 | 119878 minus (10038161003816100381610038161198781198971003816100381610038161003816 |119878|) var 119910 | 119878119897 minus (10038161003816100381610038161198781199031003816100381610038161003816 |119878|) var 119910 | 119878119903
var 119910 | 119878 (2)
where var119910 | 119878 is the mean squared error of output 119910 in thesample 119878 119878119897 and 119878119903 are two subsets of sample 119878
Journal of Advanced Transportation 5
Table 2 Descriptive statistics of variables
Variable type Variable name Abbr Variable type Mean Std Min Med Max
Usage intensity
Workday 06ndash09 h I W0609 Numerical 19013 26016 0 105 2879Workday 10ndash15 h I W1015 Numerical 27126 41904 2 152 5287Workday 16ndash19 h I W1619 Numerical 28009 48029 2 140 6214Workday 20ndash01 h I W2001 Numerical 28528 43063 0 114 3987
Nonworkday 06ndash09 h I NW0609 Numerical 6113 8656 0 32 948Nonworkday 10ndash15 h I NW1015 Numerical 14921 23139 0 80 2962Nonworkday 16ndash19 h I NW1619 Numerical 11131 21055 0 54 2925Nonworkday 20ndash01 h I NW2001 Numerical 1107 17335 0 44 1860
Stations Participating in ResearchAll Station in Shanghai
Figure 1 Spatial distribution of stations in Shanghai
012345678
200
300
400
500
600
700
800
900
100
011
00
120
013
00
140
015
00
160
017
00
180
019
00
200
021
00
220
023
00
000
100
Book
ing
Perc
enta
ge
Time
Work DayNon-work Day
02ndash05 h
06ndash09 h10ndash15 h
16ndash19 h
20ndash01 h
Figure 2 EVCARD trip temporal distribution and time division
Let 119866119894(119904 119878) be the goodness of 119894th ET the featureimportance is the average goodness of each ET which can beexpressed as follows
FI = 1119873119873sum119894=1
119866119894 (119904 119878) (3)
where FI denotes the feature importance and119873 is the numberof ET
Unimportant features have little to no effect on themean squared error model while important features shouldsignificantly decrease it
322 Monthly Usage Intensity Model and Usage ImbalanceModel Considering the unequal duration of time sections120591119879 can be the duration of each 119879 where the unit is hour
Journal of Advanced Transportation 7
Exclusive Parking SpaceOperational Age
Limited-Access Parking SpaceUnderground-Garage Parking Space
POI-Mixed Entropy
Residential Public Authority
College and University
Medical Service
Culture
Business
Social and Recreation
Industry
Intersection Density
Average Trip Distance
Train Station
Intercity Coach Station
Metro
Bus Stop
Car Rental StationMonthly usage intensity
Imbalance degree
Demand Estimation
Station Location Vehicle Relocation
support support
inputFeature selection
Freeway Length
Road Network Density
Shopping
Arterial Road LengthSecondary Road LengthLocal Street LengthOne-way lengthTwo-way length
Local Road Network Feature
Public Parking Space
Built Environment
Density
Diversity
Design
DestinationAccessibility
Transportation
Area Attributes
Feat
ures
Transaction data
EvcardDatabase
Targets
Monthly usageintensity model
Usage imbalance model
Figure 3 Research framework
Usage intensity is defined as average value per hour of the totalamount for pick-up and drop-off in specific time sections ofthe spatial unit The calculation formula is shown as follows
where 119868 represents the usage intensity and 119901119879 and 119889119879 are theamount of pick-up and drop-off at specific time-interval 119897means 119897 monthsrsquo transaction data is involved in this research119897 is 3 here
We use usage intensity as a proxy of demand rather thanthe number of bookings because the station is viewed as bothorigin and destination for the carsharing trip It not onlygenerates carsharing trip but also attracts it Therefore usageintensity is an appropriate index for carsharing demand
Usage imbalance degree is defined as the ratio of thedifference between pick-up and drop-off to the sum of pick-up and drop-off in specific time sections of the spatial unitHalf an hour is used as the statistic time interval then thedifference between pick-up and drop-off of each statisticinterval is aggregated by corresponding 119879 time section andfurther divided with the sum of pick-up and drop-off in timesection T
where 119875119879 and 119863119879 are the pick-up matrix and drop-off matrixin time section 119879 with half-hour statistic interval
IM119879 = sum119898119879119894=1sum119899119879119895=1 10038161003816100381610038161003816119901119894119895 minus 11988911989411989510038161003816100381610038161003816sum119898119879119894=1sum119899119879119895=1 119901119894119895 + 119889119894119895 (6)
where IM119879 is the imbalance degree of the spatial unit 119898119879 isthe amount of day in time section 119879 119899119879 is the statistic unitin time section 119879 and 119901119894119895 and 119889119894119895 are the pick-up and drop-off quantity in statistic unit of 119894th day and 119895th day respec-tively
Given that several types of POIs coexist in the samearea with varying degrees features have multiple collinearityproblems that cannot be ignored Meanwhile there are manyfeatures for samplesTherefore a linear regressionmodelwith1198711 and 1198712 prior as a regularizer called adaptive elastic net(AEN) regression [30] was developed to predict the usageintensity and the degree of imbalance The method can beviewed as a combination of elastic net [31] and the adaptiveleast absolute shrinkage and selection operator (LASSO) [32]which overcome the lack of adaptive LASSO (instability for
8 Journal of Advanced Transportation
high-dimensional data) and lack of the oracle property for theelastic net The AEN is defined as follows
(AEN) = (1 + 1205822119899 )
sdot argmin120573
1003817100381710038171003817119910 minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 120582lowast1119901sum119895=1
1003817100381710038171003817y minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 1205821 100381710038171003817100381712057310038171003817100381710038171 (8)
where (EN) is an elastic net algorithm 119901 denotes thenumber of features 1205731 = sum119901119895=1 |120573|1 is the 1198971-norm and 12057322is 1198972-norm 1205821 1205822 and 120582lowast1 are weights for the 1198971-norm 1198972-norm and optimal 1205821 respectively 119908119895119901119895=1 are the adaptivedata-driven weights
4 Results and Discussion
A total of 1500 ETs are used to estimate the feature importanceso that the ranks and scores of featuresrsquo importance are stableWe drop the features with importance score not greater than0015The information of featuresrsquo importance indicating thatthe contribution of each feature to prediction is provided inFigures 4 and 5 The features that are not filtered are used forbuilding prediction models The results of models are shownin Tables 3 and 4
41 Usage Intensity Model The result shows that 1198772 ofintensity models for eight time sections are between 0404and 0583 The goodness of fit is better than other researcheson this issue [15 23 24] Some features show the samedirection of influence on usage intensity However otherfactors play a positive or negative effect on the demanddepending on the time period Meanwhile all factors havedifferent weights for demand across different time periodsThis causes the usage intensity to be different across thewholeday as shown in Figure 2
411 Station-Related Factors The operational attributes ofthe spatial unit play an important role in usage intensityacross all periods The longer the first station operates inspatial unit the greater the intensity is because the serviceand location of a station are getting familiarized by usersgradually Limited-access parking space has constantly pro-vided negative effect on usage intensity Exclusive parkingspace represents the supply level partly It positively affectsusage intensity However these factors are endogenous itemsand the carsharing operator can improve these factors aspossible as it can The more important factors are exogenous
variables such as the built environment and transportationin the spatial unit
412 Built Environment Factors Built environment factorsshow the diverse effect on usage intensity in different timeperiods College and university constantly has positive influ-ence In contrast the industrial area shows negative effectResidential culture public authority and medical hygienearea impose negative influence on usage intensity in specialtime period Other factors have opposite effect depending onthe time period Recreation area has a negative influence onusage intensity in workdayrsquos early peak and positive influenceduring nonworking time sections More POI-related shop-ping is within the spatial unit and usage intensity showsmoreincrease in evening peak and night
POI-mixed entropy plays a positive role in the usageintensity during working day at around 1600ndash200 Theaverage trip distance is a special factor It has positive effect onthe evening peak of working days and night of nonworkingdays which implies that users might tend to use carsharingfor further distance trip during these periods Additionallyit has negative effect during 600ndash1000 in nonworking dayswhich indicates that short distance trip of carsharing tends tobe at 600ndash1000 in nonworking days
Areas with higher intersection density mainly duringnonworking time sections improve usage intensity becauseof good accessibility It is unexpected that the length of thelocal street and one-way road generally has negative effectand two-way road has a positive effect on usage intensityGiven that the local street and one-way street are more walk-friendly the result is opposite to some research This findingmay be attributed to our use of intensity which includespick-up and drop-off instead of pick-up only to count as anindicator of demand Many local streets and one-way streetswould make it more difficult for users to find parking spaceIn contrast more two-way roads but less one-way and localstreets means simpler road network
413 Transportation Factors Considering the transportationfactors unexpectedly traditional car rental station has apositive effect on station usage intensity at 600ndash1600 onworking days and 1000ndash2000 on nonworking days Despitebeing shown in literature that car rental and carsharing havecompetitive relationship in medium distance trip [1] thisresult indicates a more complicated relationship betweencarsharing and car rental Moreover intercity coach stationhas key role and positive relation during 600ndash1600 onnonworking days which has not been reported in any currentliterature Intercity coach stations are generally far from thecenter of the city and passengers have no personal car whiletaking some packages which could be themain reason for thedemand in using carsharing to connect with intercity coach
For public transportation the existence of metro stationnegatively affects carsharing station usage intensity duringmorning peak and evening peak on working days Thisfinding could be attributed to the belief that the metro ismore reliable for commuting compared to ground trafficand commute to work is time-limited Meanwhile the metroimposes positive effect during nonworkday which implies
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
4 Journal of Advanced Transportation
based on POIs which refers to the entropy form of research[25]
The third data source is part of the latest census dataof Shanghai which was collected in 2010 by the ShanghaiMunicipal Bureau of Statistics Resident density job densityand average trip distance of each block were used in thisresearch Average trip distance is used to represent theaccessibility of a destination
Finally the fourth data source is the information of roadnetwork in Shanghai which includes freeway arterial roadsecondary road local street one-way road and two-wayroad We counted the length of each type of road in spatialunit to represent characteristics of local road network
The four data sources are reorganized into three groupsof features The first feature is the stationrsquos attributes (119878) Thesecond feature is the transportation facilities The number ofeach type of transportation facilities located within the bufferof the carsharing station is used to present the transportationenvironment within the spatial unit The third group offeatures refers to the 5119863 principle of the built environment[26 27] However the fifth ldquo119863rdquo (distance to transit) ispresented as the distance between a residential place or workplace to the nearest public transportation station whichis not appropriate for this research Therefore we use thetransportation facilities instead of the ldquodistance to transitrdquoThus independent variables shall be categorized and groupedas ldquo119878+119879+4119863rdquoThedescriptive statistics of features and targetsare summarized in Table 2
For sample selection stable operational stations areselected in this study Many stations have short operationperiod with data that ended on 10 December 2016 Thesestations are not well known to users therefore they arenot included in this research because of their unstableperformance Data indicate that the use of vehicles becomesrelatively stable when the station is operated into the fourthmonth (the booking changing rate is about 05 The bookingchanging rate reaches 71 in the second month and 078 inthe third month) In this research we only include stationswithmore than threemonths of operation while the bookingchanging rate is less than 03 in three consecutive monthsfromOctober to December 2016These stations are judged asrelatively stable Eventually 551 stations participated in thisresearch Figure 1 shows the spatial distribution of all stationsoperating in Shanghai as well as the stations participating inthis research The proportion of stations participating in thisresearch is the same as the proportion of the whole stationat each area The central areas have less than 4 stations andparking spaces while 95 of the stations are located at theoutskirts area We divided Shanghai into a hexagonal gridwhere the area of the hexagonal spatial unit is 2 km2 (similarto a circle with 800-meter radius in the area) The stationsparticipating in this research are assigned to this hexagon unitbased on spatial location The hexagonal spatial unit is usedas an analysis unit in this research
Since the EVCARD station as basic object in thisresearch is selected according to the fact that bookingschanging rate is relatively stable during the three monthsfrom November to December 2016 the dependent variable
in this research also requires the relative stability of stationoperation performance where the collecting time of POI datashall be within this time range Therefore only the three-month transaction data from October to December 2016 isinvestigated which covers about 40 of the total bookings
311 Temporal Dynamics of Demand To investigate thetemporal dynamics of demand the data is divided into twosubsets workday and nonworkday In China given thatadjustment is made when national holidays overlap week-days some weekends become workdays and some workdaysbecome holidays Therefore workday and nonworkday areused to separate time instead of weekday and weekend
Based on these conditions the two subsets of data aredivided into five time sections of the day which are as followsaccording to the trip temporal distribution of the EVCARD(Figure 2) early morning (EM) (200ndash600 4 h) morningpeak (MP) (600ndash1000 4 h) off-peak (OP) (1000ndash16006 h) evening peak (EP) (1600ndash2000 4 h) and night (NT)(2000ndash200 6 h) Given the few bookings in EM which arequite random the later four time sections are included in thisresearch Finally eight time sections were obtained and thesetime sections are mathematically expressed as follows
where 119860 = workday nonworkday and 119861 = MP OP EPNT32 Methodology Two dependent variables are used as prox-ies representing demand monthly usage intensity and degreeof usage imbalance Twomachine-learning models are estab-lished in this research The monthly usage intensity modelis used to evaluate the effect of factors influencing usageintensity of exclusive parking space in the analysis spatialunits per month The usage imbalance model is developed toestimate the factors affecting the degree of imbalance of pick-up and drop-off at spatial units The research framework isshown in Figure 3
321 Features Selection Extremely randomized trees algo-rithm (ET) is used to select the important features to avoidoverfitting The extra-trees algorithm builds an ensembleregression trees Its two main differences from other tree-based ensemblemethods such as random forest [28] are thatit divides the nodes by choosing the cut-points completelyat random and that it uses the complete learning sampleinstead of a bootstrap replica to grow the trees The completeextra-trees algorithm is described in [29] Relative variancereduction is used to denote the goodness of point splittingFor a sample 119878 and a splitting point s (a feature selectedrandomly) the goodness of point splitting is expressed asfollows
119866 (119904 119878)= var 119910 | 119878 minus (10038161003816100381610038161198781198971003816100381610038161003816 |119878|) var 119910 | 119878119897 minus (10038161003816100381610038161198781199031003816100381610038161003816 |119878|) var 119910 | 119878119903
var 119910 | 119878 (2)
where var119910 | 119878 is the mean squared error of output 119910 in thesample 119878 119878119897 and 119878119903 are two subsets of sample 119878
Journal of Advanced Transportation 5
Table 2 Descriptive statistics of variables
Variable type Variable name Abbr Variable type Mean Std Min Med Max
Usage intensity
Workday 06ndash09 h I W0609 Numerical 19013 26016 0 105 2879Workday 10ndash15 h I W1015 Numerical 27126 41904 2 152 5287Workday 16ndash19 h I W1619 Numerical 28009 48029 2 140 6214Workday 20ndash01 h I W2001 Numerical 28528 43063 0 114 3987
Nonworkday 06ndash09 h I NW0609 Numerical 6113 8656 0 32 948Nonworkday 10ndash15 h I NW1015 Numerical 14921 23139 0 80 2962Nonworkday 16ndash19 h I NW1619 Numerical 11131 21055 0 54 2925Nonworkday 20ndash01 h I NW2001 Numerical 1107 17335 0 44 1860
Stations Participating in ResearchAll Station in Shanghai
Figure 1 Spatial distribution of stations in Shanghai
012345678
200
300
400
500
600
700
800
900
100
011
00
120
013
00
140
015
00
160
017
00
180
019
00
200
021
00
220
023
00
000
100
Book
ing
Perc
enta
ge
Time
Work DayNon-work Day
02ndash05 h
06ndash09 h10ndash15 h
16ndash19 h
20ndash01 h
Figure 2 EVCARD trip temporal distribution and time division
Let 119866119894(119904 119878) be the goodness of 119894th ET the featureimportance is the average goodness of each ET which can beexpressed as follows
FI = 1119873119873sum119894=1
119866119894 (119904 119878) (3)
where FI denotes the feature importance and119873 is the numberof ET
Unimportant features have little to no effect on themean squared error model while important features shouldsignificantly decrease it
322 Monthly Usage Intensity Model and Usage ImbalanceModel Considering the unequal duration of time sections120591119879 can be the duration of each 119879 where the unit is hour
Journal of Advanced Transportation 7
Exclusive Parking SpaceOperational Age
Limited-Access Parking SpaceUnderground-Garage Parking Space
POI-Mixed Entropy
Residential Public Authority
College and University
Medical Service
Culture
Business
Social and Recreation
Industry
Intersection Density
Average Trip Distance
Train Station
Intercity Coach Station
Metro
Bus Stop
Car Rental StationMonthly usage intensity
Imbalance degree
Demand Estimation
Station Location Vehicle Relocation
support support
inputFeature selection
Freeway Length
Road Network Density
Shopping
Arterial Road LengthSecondary Road LengthLocal Street LengthOne-way lengthTwo-way length
Local Road Network Feature
Public Parking Space
Built Environment
Density
Diversity
Design
DestinationAccessibility
Transportation
Area Attributes
Feat
ures
Transaction data
EvcardDatabase
Targets
Monthly usageintensity model
Usage imbalance model
Figure 3 Research framework
Usage intensity is defined as average value per hour of the totalamount for pick-up and drop-off in specific time sections ofthe spatial unit The calculation formula is shown as follows
where 119868 represents the usage intensity and 119901119879 and 119889119879 are theamount of pick-up and drop-off at specific time-interval 119897means 119897 monthsrsquo transaction data is involved in this research119897 is 3 here
We use usage intensity as a proxy of demand rather thanthe number of bookings because the station is viewed as bothorigin and destination for the carsharing trip It not onlygenerates carsharing trip but also attracts it Therefore usageintensity is an appropriate index for carsharing demand
Usage imbalance degree is defined as the ratio of thedifference between pick-up and drop-off to the sum of pick-up and drop-off in specific time sections of the spatial unitHalf an hour is used as the statistic time interval then thedifference between pick-up and drop-off of each statisticinterval is aggregated by corresponding 119879 time section andfurther divided with the sum of pick-up and drop-off in timesection T
where 119875119879 and 119863119879 are the pick-up matrix and drop-off matrixin time section 119879 with half-hour statistic interval
IM119879 = sum119898119879119894=1sum119899119879119895=1 10038161003816100381610038161003816119901119894119895 minus 11988911989411989510038161003816100381610038161003816sum119898119879119894=1sum119899119879119895=1 119901119894119895 + 119889119894119895 (6)
where IM119879 is the imbalance degree of the spatial unit 119898119879 isthe amount of day in time section 119879 119899119879 is the statistic unitin time section 119879 and 119901119894119895 and 119889119894119895 are the pick-up and drop-off quantity in statistic unit of 119894th day and 119895th day respec-tively
Given that several types of POIs coexist in the samearea with varying degrees features have multiple collinearityproblems that cannot be ignored Meanwhile there are manyfeatures for samplesTherefore a linear regressionmodelwith1198711 and 1198712 prior as a regularizer called adaptive elastic net(AEN) regression [30] was developed to predict the usageintensity and the degree of imbalance The method can beviewed as a combination of elastic net [31] and the adaptiveleast absolute shrinkage and selection operator (LASSO) [32]which overcome the lack of adaptive LASSO (instability for
8 Journal of Advanced Transportation
high-dimensional data) and lack of the oracle property for theelastic net The AEN is defined as follows
(AEN) = (1 + 1205822119899 )
sdot argmin120573
1003817100381710038171003817119910 minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 120582lowast1119901sum119895=1
1003817100381710038171003817y minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 1205821 100381710038171003817100381712057310038171003817100381710038171 (8)
where (EN) is an elastic net algorithm 119901 denotes thenumber of features 1205731 = sum119901119895=1 |120573|1 is the 1198971-norm and 12057322is 1198972-norm 1205821 1205822 and 120582lowast1 are weights for the 1198971-norm 1198972-norm and optimal 1205821 respectively 119908119895119901119895=1 are the adaptivedata-driven weights
4 Results and Discussion
A total of 1500 ETs are used to estimate the feature importanceso that the ranks and scores of featuresrsquo importance are stableWe drop the features with importance score not greater than0015The information of featuresrsquo importance indicating thatthe contribution of each feature to prediction is provided inFigures 4 and 5 The features that are not filtered are used forbuilding prediction models The results of models are shownin Tables 3 and 4
41 Usage Intensity Model The result shows that 1198772 ofintensity models for eight time sections are between 0404and 0583 The goodness of fit is better than other researcheson this issue [15 23 24] Some features show the samedirection of influence on usage intensity However otherfactors play a positive or negative effect on the demanddepending on the time period Meanwhile all factors havedifferent weights for demand across different time periodsThis causes the usage intensity to be different across thewholeday as shown in Figure 2
411 Station-Related Factors The operational attributes ofthe spatial unit play an important role in usage intensityacross all periods The longer the first station operates inspatial unit the greater the intensity is because the serviceand location of a station are getting familiarized by usersgradually Limited-access parking space has constantly pro-vided negative effect on usage intensity Exclusive parkingspace represents the supply level partly It positively affectsusage intensity However these factors are endogenous itemsand the carsharing operator can improve these factors aspossible as it can The more important factors are exogenous
variables such as the built environment and transportationin the spatial unit
412 Built Environment Factors Built environment factorsshow the diverse effect on usage intensity in different timeperiods College and university constantly has positive influ-ence In contrast the industrial area shows negative effectResidential culture public authority and medical hygienearea impose negative influence on usage intensity in specialtime period Other factors have opposite effect depending onthe time period Recreation area has a negative influence onusage intensity in workdayrsquos early peak and positive influenceduring nonworking time sections More POI-related shop-ping is within the spatial unit and usage intensity showsmoreincrease in evening peak and night
POI-mixed entropy plays a positive role in the usageintensity during working day at around 1600ndash200 Theaverage trip distance is a special factor It has positive effect onthe evening peak of working days and night of nonworkingdays which implies that users might tend to use carsharingfor further distance trip during these periods Additionallyit has negative effect during 600ndash1000 in nonworking dayswhich indicates that short distance trip of carsharing tends tobe at 600ndash1000 in nonworking days
Areas with higher intersection density mainly duringnonworking time sections improve usage intensity becauseof good accessibility It is unexpected that the length of thelocal street and one-way road generally has negative effectand two-way road has a positive effect on usage intensityGiven that the local street and one-way street are more walk-friendly the result is opposite to some research This findingmay be attributed to our use of intensity which includespick-up and drop-off instead of pick-up only to count as anindicator of demand Many local streets and one-way streetswould make it more difficult for users to find parking spaceIn contrast more two-way roads but less one-way and localstreets means simpler road network
413 Transportation Factors Considering the transportationfactors unexpectedly traditional car rental station has apositive effect on station usage intensity at 600ndash1600 onworking days and 1000ndash2000 on nonworking days Despitebeing shown in literature that car rental and carsharing havecompetitive relationship in medium distance trip [1] thisresult indicates a more complicated relationship betweencarsharing and car rental Moreover intercity coach stationhas key role and positive relation during 600ndash1600 onnonworking days which has not been reported in any currentliterature Intercity coach stations are generally far from thecenter of the city and passengers have no personal car whiletaking some packages which could be themain reason for thedemand in using carsharing to connect with intercity coach
For public transportation the existence of metro stationnegatively affects carsharing station usage intensity duringmorning peak and evening peak on working days Thisfinding could be attributed to the belief that the metro ismore reliable for commuting compared to ground trafficand commute to work is time-limited Meanwhile the metroimposes positive effect during nonworkday which implies
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Stations Participating in ResearchAll Station in Shanghai
Figure 1 Spatial distribution of stations in Shanghai
012345678
200
300
400
500
600
700
800
900
100
011
00
120
013
00
140
015
00
160
017
00
180
019
00
200
021
00
220
023
00
000
100
Book
ing
Perc
enta
ge
Time
Work DayNon-work Day
02ndash05 h
06ndash09 h10ndash15 h
16ndash19 h
20ndash01 h
Figure 2 EVCARD trip temporal distribution and time division
Let 119866119894(119904 119878) be the goodness of 119894th ET the featureimportance is the average goodness of each ET which can beexpressed as follows
FI = 1119873119873sum119894=1
119866119894 (119904 119878) (3)
where FI denotes the feature importance and119873 is the numberof ET
Unimportant features have little to no effect on themean squared error model while important features shouldsignificantly decrease it
322 Monthly Usage Intensity Model and Usage ImbalanceModel Considering the unequal duration of time sections120591119879 can be the duration of each 119879 where the unit is hour
Journal of Advanced Transportation 7
Exclusive Parking SpaceOperational Age
Limited-Access Parking SpaceUnderground-Garage Parking Space
POI-Mixed Entropy
Residential Public Authority
College and University
Medical Service
Culture
Business
Social and Recreation
Industry
Intersection Density
Average Trip Distance
Train Station
Intercity Coach Station
Metro
Bus Stop
Car Rental StationMonthly usage intensity
Imbalance degree
Demand Estimation
Station Location Vehicle Relocation
support support
inputFeature selection
Freeway Length
Road Network Density
Shopping
Arterial Road LengthSecondary Road LengthLocal Street LengthOne-way lengthTwo-way length
Local Road Network Feature
Public Parking Space
Built Environment
Density
Diversity
Design
DestinationAccessibility
Transportation
Area Attributes
Feat
ures
Transaction data
EvcardDatabase
Targets
Monthly usageintensity model
Usage imbalance model
Figure 3 Research framework
Usage intensity is defined as average value per hour of the totalamount for pick-up and drop-off in specific time sections ofthe spatial unit The calculation formula is shown as follows
where 119868 represents the usage intensity and 119901119879 and 119889119879 are theamount of pick-up and drop-off at specific time-interval 119897means 119897 monthsrsquo transaction data is involved in this research119897 is 3 here
We use usage intensity as a proxy of demand rather thanthe number of bookings because the station is viewed as bothorigin and destination for the carsharing trip It not onlygenerates carsharing trip but also attracts it Therefore usageintensity is an appropriate index for carsharing demand
Usage imbalance degree is defined as the ratio of thedifference between pick-up and drop-off to the sum of pick-up and drop-off in specific time sections of the spatial unitHalf an hour is used as the statistic time interval then thedifference between pick-up and drop-off of each statisticinterval is aggregated by corresponding 119879 time section andfurther divided with the sum of pick-up and drop-off in timesection T
where 119875119879 and 119863119879 are the pick-up matrix and drop-off matrixin time section 119879 with half-hour statistic interval
IM119879 = sum119898119879119894=1sum119899119879119895=1 10038161003816100381610038161003816119901119894119895 minus 11988911989411989510038161003816100381610038161003816sum119898119879119894=1sum119899119879119895=1 119901119894119895 + 119889119894119895 (6)
where IM119879 is the imbalance degree of the spatial unit 119898119879 isthe amount of day in time section 119879 119899119879 is the statistic unitin time section 119879 and 119901119894119895 and 119889119894119895 are the pick-up and drop-off quantity in statistic unit of 119894th day and 119895th day respec-tively
Given that several types of POIs coexist in the samearea with varying degrees features have multiple collinearityproblems that cannot be ignored Meanwhile there are manyfeatures for samplesTherefore a linear regressionmodelwith1198711 and 1198712 prior as a regularizer called adaptive elastic net(AEN) regression [30] was developed to predict the usageintensity and the degree of imbalance The method can beviewed as a combination of elastic net [31] and the adaptiveleast absolute shrinkage and selection operator (LASSO) [32]which overcome the lack of adaptive LASSO (instability for
8 Journal of Advanced Transportation
high-dimensional data) and lack of the oracle property for theelastic net The AEN is defined as follows
(AEN) = (1 + 1205822119899 )
sdot argmin120573
1003817100381710038171003817119910 minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 120582lowast1119901sum119895=1
1003817100381710038171003817y minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 1205821 100381710038171003817100381712057310038171003817100381710038171 (8)
where (EN) is an elastic net algorithm 119901 denotes thenumber of features 1205731 = sum119901119895=1 |120573|1 is the 1198971-norm and 12057322is 1198972-norm 1205821 1205822 and 120582lowast1 are weights for the 1198971-norm 1198972-norm and optimal 1205821 respectively 119908119895119901119895=1 are the adaptivedata-driven weights
4 Results and Discussion
A total of 1500 ETs are used to estimate the feature importanceso that the ranks and scores of featuresrsquo importance are stableWe drop the features with importance score not greater than0015The information of featuresrsquo importance indicating thatthe contribution of each feature to prediction is provided inFigures 4 and 5 The features that are not filtered are used forbuilding prediction models The results of models are shownin Tables 3 and 4
41 Usage Intensity Model The result shows that 1198772 ofintensity models for eight time sections are between 0404and 0583 The goodness of fit is better than other researcheson this issue [15 23 24] Some features show the samedirection of influence on usage intensity However otherfactors play a positive or negative effect on the demanddepending on the time period Meanwhile all factors havedifferent weights for demand across different time periodsThis causes the usage intensity to be different across thewholeday as shown in Figure 2
411 Station-Related Factors The operational attributes ofthe spatial unit play an important role in usage intensityacross all periods The longer the first station operates inspatial unit the greater the intensity is because the serviceand location of a station are getting familiarized by usersgradually Limited-access parking space has constantly pro-vided negative effect on usage intensity Exclusive parkingspace represents the supply level partly It positively affectsusage intensity However these factors are endogenous itemsand the carsharing operator can improve these factors aspossible as it can The more important factors are exogenous
variables such as the built environment and transportationin the spatial unit
412 Built Environment Factors Built environment factorsshow the diverse effect on usage intensity in different timeperiods College and university constantly has positive influ-ence In contrast the industrial area shows negative effectResidential culture public authority and medical hygienearea impose negative influence on usage intensity in specialtime period Other factors have opposite effect depending onthe time period Recreation area has a negative influence onusage intensity in workdayrsquos early peak and positive influenceduring nonworking time sections More POI-related shop-ping is within the spatial unit and usage intensity showsmoreincrease in evening peak and night
POI-mixed entropy plays a positive role in the usageintensity during working day at around 1600ndash200 Theaverage trip distance is a special factor It has positive effect onthe evening peak of working days and night of nonworkingdays which implies that users might tend to use carsharingfor further distance trip during these periods Additionallyit has negative effect during 600ndash1000 in nonworking dayswhich indicates that short distance trip of carsharing tends tobe at 600ndash1000 in nonworking days
Areas with higher intersection density mainly duringnonworking time sections improve usage intensity becauseof good accessibility It is unexpected that the length of thelocal street and one-way road generally has negative effectand two-way road has a positive effect on usage intensityGiven that the local street and one-way street are more walk-friendly the result is opposite to some research This findingmay be attributed to our use of intensity which includespick-up and drop-off instead of pick-up only to count as anindicator of demand Many local streets and one-way streetswould make it more difficult for users to find parking spaceIn contrast more two-way roads but less one-way and localstreets means simpler road network
413 Transportation Factors Considering the transportationfactors unexpectedly traditional car rental station has apositive effect on station usage intensity at 600ndash1600 onworking days and 1000ndash2000 on nonworking days Despitebeing shown in literature that car rental and carsharing havecompetitive relationship in medium distance trip [1] thisresult indicates a more complicated relationship betweencarsharing and car rental Moreover intercity coach stationhas key role and positive relation during 600ndash1600 onnonworking days which has not been reported in any currentliterature Intercity coach stations are generally far from thecenter of the city and passengers have no personal car whiletaking some packages which could be themain reason for thedemand in using carsharing to connect with intercity coach
For public transportation the existence of metro stationnegatively affects carsharing station usage intensity duringmorning peak and evening peak on working days Thisfinding could be attributed to the belief that the metro ismore reliable for commuting compared to ground trafficand commute to work is time-limited Meanwhile the metroimposes positive effect during nonworkday which implies
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
6 Journal of Advanced Transportation
Analysis UnitArea
N
0 5 10 20 30 40
(Kilometers)
Stations Participating in ResearchAll Station in Shanghai
Figure 1 Spatial distribution of stations in Shanghai
012345678
200
300
400
500
600
700
800
900
100
011
00
120
013
00
140
015
00
160
017
00
180
019
00
200
021
00
220
023
00
000
100
Book
ing
Perc
enta
ge
Time
Work DayNon-work Day
02ndash05 h
06ndash09 h10ndash15 h
16ndash19 h
20ndash01 h
Figure 2 EVCARD trip temporal distribution and time division
Let 119866119894(119904 119878) be the goodness of 119894th ET the featureimportance is the average goodness of each ET which can beexpressed as follows
FI = 1119873119873sum119894=1
119866119894 (119904 119878) (3)
where FI denotes the feature importance and119873 is the numberof ET
Unimportant features have little to no effect on themean squared error model while important features shouldsignificantly decrease it
322 Monthly Usage Intensity Model and Usage ImbalanceModel Considering the unequal duration of time sections120591119879 can be the duration of each 119879 where the unit is hour
Journal of Advanced Transportation 7
Exclusive Parking SpaceOperational Age
Limited-Access Parking SpaceUnderground-Garage Parking Space
POI-Mixed Entropy
Residential Public Authority
College and University
Medical Service
Culture
Business
Social and Recreation
Industry
Intersection Density
Average Trip Distance
Train Station
Intercity Coach Station
Metro
Bus Stop
Car Rental StationMonthly usage intensity
Imbalance degree
Demand Estimation
Station Location Vehicle Relocation
support support
inputFeature selection
Freeway Length
Road Network Density
Shopping
Arterial Road LengthSecondary Road LengthLocal Street LengthOne-way lengthTwo-way length
Local Road Network Feature
Public Parking Space
Built Environment
Density
Diversity
Design
DestinationAccessibility
Transportation
Area Attributes
Feat
ures
Transaction data
EvcardDatabase
Targets
Monthly usageintensity model
Usage imbalance model
Figure 3 Research framework
Usage intensity is defined as average value per hour of the totalamount for pick-up and drop-off in specific time sections ofthe spatial unit The calculation formula is shown as follows
where 119868 represents the usage intensity and 119901119879 and 119889119879 are theamount of pick-up and drop-off at specific time-interval 119897means 119897 monthsrsquo transaction data is involved in this research119897 is 3 here
We use usage intensity as a proxy of demand rather thanthe number of bookings because the station is viewed as bothorigin and destination for the carsharing trip It not onlygenerates carsharing trip but also attracts it Therefore usageintensity is an appropriate index for carsharing demand
Usage imbalance degree is defined as the ratio of thedifference between pick-up and drop-off to the sum of pick-up and drop-off in specific time sections of the spatial unitHalf an hour is used as the statistic time interval then thedifference between pick-up and drop-off of each statisticinterval is aggregated by corresponding 119879 time section andfurther divided with the sum of pick-up and drop-off in timesection T
where 119875119879 and 119863119879 are the pick-up matrix and drop-off matrixin time section 119879 with half-hour statistic interval
IM119879 = sum119898119879119894=1sum119899119879119895=1 10038161003816100381610038161003816119901119894119895 minus 11988911989411989510038161003816100381610038161003816sum119898119879119894=1sum119899119879119895=1 119901119894119895 + 119889119894119895 (6)
where IM119879 is the imbalance degree of the spatial unit 119898119879 isthe amount of day in time section 119879 119899119879 is the statistic unitin time section 119879 and 119901119894119895 and 119889119894119895 are the pick-up and drop-off quantity in statistic unit of 119894th day and 119895th day respec-tively
Given that several types of POIs coexist in the samearea with varying degrees features have multiple collinearityproblems that cannot be ignored Meanwhile there are manyfeatures for samplesTherefore a linear regressionmodelwith1198711 and 1198712 prior as a regularizer called adaptive elastic net(AEN) regression [30] was developed to predict the usageintensity and the degree of imbalance The method can beviewed as a combination of elastic net [31] and the adaptiveleast absolute shrinkage and selection operator (LASSO) [32]which overcome the lack of adaptive LASSO (instability for
8 Journal of Advanced Transportation
high-dimensional data) and lack of the oracle property for theelastic net The AEN is defined as follows
(AEN) = (1 + 1205822119899 )
sdot argmin120573
1003817100381710038171003817119910 minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 120582lowast1119901sum119895=1
1003817100381710038171003817y minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 1205821 100381710038171003817100381712057310038171003817100381710038171 (8)
where (EN) is an elastic net algorithm 119901 denotes thenumber of features 1205731 = sum119901119895=1 |120573|1 is the 1198971-norm and 12057322is 1198972-norm 1205821 1205822 and 120582lowast1 are weights for the 1198971-norm 1198972-norm and optimal 1205821 respectively 119908119895119901119895=1 are the adaptivedata-driven weights
4 Results and Discussion
A total of 1500 ETs are used to estimate the feature importanceso that the ranks and scores of featuresrsquo importance are stableWe drop the features with importance score not greater than0015The information of featuresrsquo importance indicating thatthe contribution of each feature to prediction is provided inFigures 4 and 5 The features that are not filtered are used forbuilding prediction models The results of models are shownin Tables 3 and 4
41 Usage Intensity Model The result shows that 1198772 ofintensity models for eight time sections are between 0404and 0583 The goodness of fit is better than other researcheson this issue [15 23 24] Some features show the samedirection of influence on usage intensity However otherfactors play a positive or negative effect on the demanddepending on the time period Meanwhile all factors havedifferent weights for demand across different time periodsThis causes the usage intensity to be different across thewholeday as shown in Figure 2
411 Station-Related Factors The operational attributes ofthe spatial unit play an important role in usage intensityacross all periods The longer the first station operates inspatial unit the greater the intensity is because the serviceand location of a station are getting familiarized by usersgradually Limited-access parking space has constantly pro-vided negative effect on usage intensity Exclusive parkingspace represents the supply level partly It positively affectsusage intensity However these factors are endogenous itemsand the carsharing operator can improve these factors aspossible as it can The more important factors are exogenous
variables such as the built environment and transportationin the spatial unit
412 Built Environment Factors Built environment factorsshow the diverse effect on usage intensity in different timeperiods College and university constantly has positive influ-ence In contrast the industrial area shows negative effectResidential culture public authority and medical hygienearea impose negative influence on usage intensity in specialtime period Other factors have opposite effect depending onthe time period Recreation area has a negative influence onusage intensity in workdayrsquos early peak and positive influenceduring nonworking time sections More POI-related shop-ping is within the spatial unit and usage intensity showsmoreincrease in evening peak and night
POI-mixed entropy plays a positive role in the usageintensity during working day at around 1600ndash200 Theaverage trip distance is a special factor It has positive effect onthe evening peak of working days and night of nonworkingdays which implies that users might tend to use carsharingfor further distance trip during these periods Additionallyit has negative effect during 600ndash1000 in nonworking dayswhich indicates that short distance trip of carsharing tends tobe at 600ndash1000 in nonworking days
Areas with higher intersection density mainly duringnonworking time sections improve usage intensity becauseof good accessibility It is unexpected that the length of thelocal street and one-way road generally has negative effectand two-way road has a positive effect on usage intensityGiven that the local street and one-way street are more walk-friendly the result is opposite to some research This findingmay be attributed to our use of intensity which includespick-up and drop-off instead of pick-up only to count as anindicator of demand Many local streets and one-way streetswould make it more difficult for users to find parking spaceIn contrast more two-way roads but less one-way and localstreets means simpler road network
413 Transportation Factors Considering the transportationfactors unexpectedly traditional car rental station has apositive effect on station usage intensity at 600ndash1600 onworking days and 1000ndash2000 on nonworking days Despitebeing shown in literature that car rental and carsharing havecompetitive relationship in medium distance trip [1] thisresult indicates a more complicated relationship betweencarsharing and car rental Moreover intercity coach stationhas key role and positive relation during 600ndash1600 onnonworking days which has not been reported in any currentliterature Intercity coach stations are generally far from thecenter of the city and passengers have no personal car whiletaking some packages which could be themain reason for thedemand in using carsharing to connect with intercity coach
For public transportation the existence of metro stationnegatively affects carsharing station usage intensity duringmorning peak and evening peak on working days Thisfinding could be attributed to the belief that the metro ismore reliable for commuting compared to ground trafficand commute to work is time-limited Meanwhile the metroimposes positive effect during nonworkday which implies
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Advanced Transportation 7
Exclusive Parking SpaceOperational Age
Limited-Access Parking SpaceUnderground-Garage Parking Space
POI-Mixed Entropy
Residential Public Authority
College and University
Medical Service
Culture
Business
Social and Recreation
Industry
Intersection Density
Average Trip Distance
Train Station
Intercity Coach Station
Metro
Bus Stop
Car Rental StationMonthly usage intensity
Imbalance degree
Demand Estimation
Station Location Vehicle Relocation
support support
inputFeature selection
Freeway Length
Road Network Density
Shopping
Arterial Road LengthSecondary Road LengthLocal Street LengthOne-way lengthTwo-way length
Local Road Network Feature
Public Parking Space
Built Environment
Density
Diversity
Design
DestinationAccessibility
Transportation
Area Attributes
Feat
ures
Transaction data
EvcardDatabase
Targets
Monthly usageintensity model
Usage imbalance model
Figure 3 Research framework
Usage intensity is defined as average value per hour of the totalamount for pick-up and drop-off in specific time sections ofthe spatial unit The calculation formula is shown as follows
where 119868 represents the usage intensity and 119901119879 and 119889119879 are theamount of pick-up and drop-off at specific time-interval 119897means 119897 monthsrsquo transaction data is involved in this research119897 is 3 here
We use usage intensity as a proxy of demand rather thanthe number of bookings because the station is viewed as bothorigin and destination for the carsharing trip It not onlygenerates carsharing trip but also attracts it Therefore usageintensity is an appropriate index for carsharing demand
Usage imbalance degree is defined as the ratio of thedifference between pick-up and drop-off to the sum of pick-up and drop-off in specific time sections of the spatial unitHalf an hour is used as the statistic time interval then thedifference between pick-up and drop-off of each statisticinterval is aggregated by corresponding 119879 time section andfurther divided with the sum of pick-up and drop-off in timesection T
where 119875119879 and 119863119879 are the pick-up matrix and drop-off matrixin time section 119879 with half-hour statistic interval
IM119879 = sum119898119879119894=1sum119899119879119895=1 10038161003816100381610038161003816119901119894119895 minus 11988911989411989510038161003816100381610038161003816sum119898119879119894=1sum119899119879119895=1 119901119894119895 + 119889119894119895 (6)
where IM119879 is the imbalance degree of the spatial unit 119898119879 isthe amount of day in time section 119879 119899119879 is the statistic unitin time section 119879 and 119901119894119895 and 119889119894119895 are the pick-up and drop-off quantity in statistic unit of 119894th day and 119895th day respec-tively
Given that several types of POIs coexist in the samearea with varying degrees features have multiple collinearityproblems that cannot be ignored Meanwhile there are manyfeatures for samplesTherefore a linear regressionmodelwith1198711 and 1198712 prior as a regularizer called adaptive elastic net(AEN) regression [30] was developed to predict the usageintensity and the degree of imbalance The method can beviewed as a combination of elastic net [31] and the adaptiveleast absolute shrinkage and selection operator (LASSO) [32]which overcome the lack of adaptive LASSO (instability for
8 Journal of Advanced Transportation
high-dimensional data) and lack of the oracle property for theelastic net The AEN is defined as follows
(AEN) = (1 + 1205822119899 )
sdot argmin120573
1003817100381710038171003817119910 minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 120582lowast1119901sum119895=1
1003817100381710038171003817y minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 1205821 100381710038171003817100381712057310038171003817100381710038171 (8)
where (EN) is an elastic net algorithm 119901 denotes thenumber of features 1205731 = sum119901119895=1 |120573|1 is the 1198971-norm and 12057322is 1198972-norm 1205821 1205822 and 120582lowast1 are weights for the 1198971-norm 1198972-norm and optimal 1205821 respectively 119908119895119901119895=1 are the adaptivedata-driven weights
4 Results and Discussion
A total of 1500 ETs are used to estimate the feature importanceso that the ranks and scores of featuresrsquo importance are stableWe drop the features with importance score not greater than0015The information of featuresrsquo importance indicating thatthe contribution of each feature to prediction is provided inFigures 4 and 5 The features that are not filtered are used forbuilding prediction models The results of models are shownin Tables 3 and 4
41 Usage Intensity Model The result shows that 1198772 ofintensity models for eight time sections are between 0404and 0583 The goodness of fit is better than other researcheson this issue [15 23 24] Some features show the samedirection of influence on usage intensity However otherfactors play a positive or negative effect on the demanddepending on the time period Meanwhile all factors havedifferent weights for demand across different time periodsThis causes the usage intensity to be different across thewholeday as shown in Figure 2
411 Station-Related Factors The operational attributes ofthe spatial unit play an important role in usage intensityacross all periods The longer the first station operates inspatial unit the greater the intensity is because the serviceand location of a station are getting familiarized by usersgradually Limited-access parking space has constantly pro-vided negative effect on usage intensity Exclusive parkingspace represents the supply level partly It positively affectsusage intensity However these factors are endogenous itemsand the carsharing operator can improve these factors aspossible as it can The more important factors are exogenous
variables such as the built environment and transportationin the spatial unit
412 Built Environment Factors Built environment factorsshow the diverse effect on usage intensity in different timeperiods College and university constantly has positive influ-ence In contrast the industrial area shows negative effectResidential culture public authority and medical hygienearea impose negative influence on usage intensity in specialtime period Other factors have opposite effect depending onthe time period Recreation area has a negative influence onusage intensity in workdayrsquos early peak and positive influenceduring nonworking time sections More POI-related shop-ping is within the spatial unit and usage intensity showsmoreincrease in evening peak and night
POI-mixed entropy plays a positive role in the usageintensity during working day at around 1600ndash200 Theaverage trip distance is a special factor It has positive effect onthe evening peak of working days and night of nonworkingdays which implies that users might tend to use carsharingfor further distance trip during these periods Additionallyit has negative effect during 600ndash1000 in nonworking dayswhich indicates that short distance trip of carsharing tends tobe at 600ndash1000 in nonworking days
Areas with higher intersection density mainly duringnonworking time sections improve usage intensity becauseof good accessibility It is unexpected that the length of thelocal street and one-way road generally has negative effectand two-way road has a positive effect on usage intensityGiven that the local street and one-way street are more walk-friendly the result is opposite to some research This findingmay be attributed to our use of intensity which includespick-up and drop-off instead of pick-up only to count as anindicator of demand Many local streets and one-way streetswould make it more difficult for users to find parking spaceIn contrast more two-way roads but less one-way and localstreets means simpler road network
413 Transportation Factors Considering the transportationfactors unexpectedly traditional car rental station has apositive effect on station usage intensity at 600ndash1600 onworking days and 1000ndash2000 on nonworking days Despitebeing shown in literature that car rental and carsharing havecompetitive relationship in medium distance trip [1] thisresult indicates a more complicated relationship betweencarsharing and car rental Moreover intercity coach stationhas key role and positive relation during 600ndash1600 onnonworking days which has not been reported in any currentliterature Intercity coach stations are generally far from thecenter of the city and passengers have no personal car whiletaking some packages which could be themain reason for thedemand in using carsharing to connect with intercity coach
For public transportation the existence of metro stationnegatively affects carsharing station usage intensity duringmorning peak and evening peak on working days Thisfinding could be attributed to the belief that the metro ismore reliable for commuting compared to ground trafficand commute to work is time-limited Meanwhile the metroimposes positive effect during nonworkday which implies
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
1003817100381710038171003817y minus X120573100381710038171003817100381722 + 1205822 1003817100381710038171003817120573100381710038171003817100381722 + 1205821 100381710038171003817100381712057310038171003817100381710038171 (8)
where (EN) is an elastic net algorithm 119901 denotes thenumber of features 1205731 = sum119901119895=1 |120573|1 is the 1198971-norm and 12057322is 1198972-norm 1205821 1205822 and 120582lowast1 are weights for the 1198971-norm 1198972-norm and optimal 1205821 respectively 119908119895119901119895=1 are the adaptivedata-driven weights
4 Results and Discussion
A total of 1500 ETs are used to estimate the feature importanceso that the ranks and scores of featuresrsquo importance are stableWe drop the features with importance score not greater than0015The information of featuresrsquo importance indicating thatthe contribution of each feature to prediction is provided inFigures 4 and 5 The features that are not filtered are used forbuilding prediction models The results of models are shownin Tables 3 and 4
41 Usage Intensity Model The result shows that 1198772 ofintensity models for eight time sections are between 0404and 0583 The goodness of fit is better than other researcheson this issue [15 23 24] Some features show the samedirection of influence on usage intensity However otherfactors play a positive or negative effect on the demanddepending on the time period Meanwhile all factors havedifferent weights for demand across different time periodsThis causes the usage intensity to be different across thewholeday as shown in Figure 2
411 Station-Related Factors The operational attributes ofthe spatial unit play an important role in usage intensityacross all periods The longer the first station operates inspatial unit the greater the intensity is because the serviceand location of a station are getting familiarized by usersgradually Limited-access parking space has constantly pro-vided negative effect on usage intensity Exclusive parkingspace represents the supply level partly It positively affectsusage intensity However these factors are endogenous itemsand the carsharing operator can improve these factors aspossible as it can The more important factors are exogenous
variables such as the built environment and transportationin the spatial unit
412 Built Environment Factors Built environment factorsshow the diverse effect on usage intensity in different timeperiods College and university constantly has positive influ-ence In contrast the industrial area shows negative effectResidential culture public authority and medical hygienearea impose negative influence on usage intensity in specialtime period Other factors have opposite effect depending onthe time period Recreation area has a negative influence onusage intensity in workdayrsquos early peak and positive influenceduring nonworking time sections More POI-related shop-ping is within the spatial unit and usage intensity showsmoreincrease in evening peak and night
POI-mixed entropy plays a positive role in the usageintensity during working day at around 1600ndash200 Theaverage trip distance is a special factor It has positive effect onthe evening peak of working days and night of nonworkingdays which implies that users might tend to use carsharingfor further distance trip during these periods Additionallyit has negative effect during 600ndash1000 in nonworking dayswhich indicates that short distance trip of carsharing tends tobe at 600ndash1000 in nonworking days
Areas with higher intersection density mainly duringnonworking time sections improve usage intensity becauseof good accessibility It is unexpected that the length of thelocal street and one-way road generally has negative effectand two-way road has a positive effect on usage intensityGiven that the local street and one-way street are more walk-friendly the result is opposite to some research This findingmay be attributed to our use of intensity which includespick-up and drop-off instead of pick-up only to count as anindicator of demand Many local streets and one-way streetswould make it more difficult for users to find parking spaceIn contrast more two-way roads but less one-way and localstreets means simpler road network
413 Transportation Factors Considering the transportationfactors unexpectedly traditional car rental station has apositive effect on station usage intensity at 600ndash1600 onworking days and 1000ndash2000 on nonworking days Despitebeing shown in literature that car rental and carsharing havecompetitive relationship in medium distance trip [1] thisresult indicates a more complicated relationship betweencarsharing and car rental Moreover intercity coach stationhas key role and positive relation during 600ndash1600 onnonworking days which has not been reported in any currentliterature Intercity coach stations are generally far from thecenter of the city and passengers have no personal car whiletaking some packages which could be themain reason for thedemand in using carsharing to connect with intercity coach
For public transportation the existence of metro stationnegatively affects carsharing station usage intensity duringmorning peak and evening peak on working days Thisfinding could be attributed to the belief that the metro ismore reliable for commuting compared to ground trafficand commute to work is time-limited Meanwhile the metroimposes positive effect during nonworkday which implies
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Advanced Transportation 9
TSUGFWARSRCRJD
CUMSPULAINRC
JHRORRD
INUSH
NASBUPATRRSLS
MHID
ATDBS
PMEICSUNPSSA
Feat
ure
TSFWUGARCRSRINSH
NASJD
PULACU
JHRRDTR
MHORBURC
INUMSRSPA
ATDLSBSID
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRINLACUJDSHPUOR
NASMS
MHBU
INUATD
RDRC
JHRTRRSIDPALSBS
PMEICSUNPSSA
Feat
ure
TSFWUGARSRCRORPUCUICS
NASJHR
INJDTR
INUSH
MHBURDLSRSBS
ATDIDPA
PMERCLA
UNPS
MSSA
Feat
ure
TSUGFWARMSSR
CULAJD
CRRDRCPUINSHRS
JHRBU
INUPAORUN
NASTR
ATDICSBS
MHLSID
PMEPSSA
Feat
ure
TSFWUGARSRLAMSCUJDIN
RDRCPU
MHCRSH
JHRINU
BURSPAORBSTRICS
LSNAS
IDATDPME
UNPSSA
Feat
ure
TSFWARUGSRLACUINPUJD
CRSHRCORMS
MHJHR
PARDTRBU
NASINU
RSIDLSBS
ATDPME
ICSUNSAPS
Feat
ure
TSFWUGARCRSR
ORJHRPU
NASCUICSJDINTRRD
INUBUSH
MHLSBSRS
ATDIDPA
PMEPS
UNRCMSLASA
Feat
ure
005 010 015000
Importance Score
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score002 004 006 008000
Importance Score
002
004
006
008
010
000
Importance Score
005 010 015000
Importance Score005 010 015000
Importance Score0000
0025
0050
0075
0100
0125
Feature importances for usage intensityMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 4 Feature importance for usage intensity
that carsharing users are using the metro to connect carshar-ing during nonworkdayThemetro competes with carsharingin rush hours but they cooperate with each other duringnonworkdays This relationship shows a policy potentialfor the government to promote diversified mobility withoutdeteriorating ground traffic condition
The bus stop has a positive effect during 600ndash2000 inboth workday and nonworkday which is similar to literature[20] This could be attributed to the good accessibility ofthe area near bus stops Therefore the exposed rate ofcarsharing will be high if the station is placed in nearby busstop This explains the nonsignificance during 1000ndash2000in nonworkdays because the main purpose of nonworkdaysis leisure which requires higher sensitivity to comfort andlower sensitivity to price
Public parking space has a significantly negative impacton usage intensity The result implies that more publicparking spaces result in more private vehicle trips rather thancarsharing Since private vehicle is very inefficient in usingparking space if a part of the public parking space is replacedwith carsharing exclusive parking space gradually it can (1)save huge area of high-value land in the center of the city and(2) reduce private vehicle usage
42 Usage Imbalance Model 1198772 of usage imbalance modelsfor eight time sections are between 0217 and 0514 which areworse than the usage intensity models The worst imbalancemodel is that of the morning peak in a nonworkday The lessusage in the early morning of the nonworkday results in a fewfactors showing significant effect
421 Station-Related Factors With increasing operation agethe degree of imbalance decreases for all time periods Aspatial unit with more limited-access parking space meansthat it only serves lower proportion of users and the demanddiversity (purpose departure time and arriving time) withinthe spatial unit is lowerThe same reason results in the similarappearance effect of underground-garage parking spacesTherefore the operator should locate less parking space onlimited access and underground garage
422 Built Environment Factors At the built environmentfactor residential public authority business and industrialarea continually play a negative role to increase the degreeof imbalance in special time period Recreation medicalhygiene university and shopping area have different effectin different time periods Among them the university has a
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
10 Journal of Advanced Transportation
UGFWMSLAARCRSRJD
UNCURD
JHRICSRCINBUORRSSH
ATDPA
INUBSTRPU
PMELS
MHNAS
SAIDPS
TS
000 002 004 006
Importance Score
Importance Score
Feat
ure
Feat
ure
TSUGICSCRARFWSR
JHRUN
NASOR
INUJD
RDPU
MHBUCUBSSH
ATDIDTRMSLSPSINPASARS
PMERCLA
TSUGFWARCRCURDSRJDSHPUOR
INUICSBURS
JHRMHNAS
TRIN
PMEMSLS
RCATD
PSBSPALA
UNIDSA
Feat
ure
TSUGFWMSCRARSRLASH
ICSRSRCJD
MHRDBUCUORPA
INUJHRPULSIDIN
ATDNAS
TRBS
PMEUNSAPS
Feat
ure
002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score002 004 006000
Importance Score
002 004 006000
Importance Score002 004 006000
Importance Score
TSUGICSUNCRARFWSR
ORSHPS
JHRINTRPUBUCU
ATDRDPA
MHRS
NASJD
INUBSLS
MSRC
PMEIDSALA
Feat
ure
TSFWCRARUGSR
ICSPU
JHRCU
MHNAS
SHJD
RDMSBUOR
INUPME
BSATD
PSPARCLSRSTRINIDSA
UNLA
Feat
ure
TSUGCRARFWICSSR
MSBURSJDSH
MHRD
ATDINU
PSPU
NASOR
JHRCULSBSTRRCPA
PMEIDSAIN
UNLA
Feat
ure
TSMSUNICSUGARFWCRSRJDSH
NASRD
MHPS
JHRPABUORIDTRLS
RCPUCU
INUBSLARS
ATDPME
INSA
Feat
ure
000
001
002
003
004
005
Feature importances for usage imbalanceMP (600ndash1000) Work day OP (1000ndash1600) Work day EP (1600ndash2000) Work day NT (2000ndash200) Work day
MP (600ndash1000) Non-Work day OP (1000ndash1600) Non-Work day EP (1600ndash2000) Non-Work day NT (2000ndash200) Non-Work day
Figure 5 Feature importance for usage imbalance
positive effect on station balance in most of the time sectionsgiven that many adult students live in this area These peoplehave less time constraint flexible travel time and diversetravel purpose However it appears as a negative effect during600ndash1000 on nonworkday which could mean that peopleliving in these areas tend to go out of campus during this timesection
Intersection density represents accessibility partially Itcan influence people who are unfamiliar with station locationto access the station More people using the station cangenerate and attract diversified and compensative usage ofpick-up and drop-off Our new finding is that a longer arterialroad and secondary road lead to higher degree of usageimbalance in the spatial unit By contrast the local streetresults inmore balanced carsharing spatial unit Analogouslymore two-way roads show higher degree of imbalance andmore one-way streets result in lower imbalance This couldbe attributed to the increased possibility of imbalance in aspecified area because of higher usage intensity
POI-mixed entropy reduces the degree of imbalanceduring morning peak and night of workdays and non-workday nights The diversified demand can reduce theusage imbalance However it increases the degree of stationimbalance during 1000ndash1600 on working days which couldbe attributed to the low usage intensity of the high diversity
areas in this time section indicative of scattered pick-up anddrop-off It causes the imbalance of the demand for drop-offand pick-up during the statistical interval (half hour)
423 Transportation Variables For transportation variablesmetro stations play a significantly positive role in stationusage balance because of the huge crowd nearby metro andthe trip purpose and time are diverse On the other handit is implied that carsharing has a closer relationship withmetro and there might be a demand for connection betweencarsharing and metro Therefore it is implied that those twomodes can compensate each other By contrast the bus stopshows opposite effect (negative) on usage imbalance Giventhat the main difference between the bus and the metro isthat the former runs on the ground where the uncertainty oftrip duration is larger the significant finding implies that car-sharing attracts a part of the bus passengers unidirectionallyeven in early peak and evening peak on workday Thereforefrom the government viewpoint carsharing station shouldnot be located near a bus stop which results in a transit triptransferring to a car trip Besides car rental station appearsto have positive effect on station usage balance in partialtime section It is implied that there is a demand of usingcarsharing to connect with car rental
The results of imbalance model are shown in Table 4
Journal of Advanced Transportation 11
Table 3 The result of monthly usage intensity model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Combining these two models the significant features canbe arranged as shown in Figure 6 The features within therange of the dotted lines and located on 119909-axis or 119910-axisonly have significant impact on single dependent variablesMeanwhile the others in the outer side beyond the dottedrange are significant on both dependent variables
Given an average value of area attributes as shown inTable 2 we get some appropriate location of carsharingstation based on the result of usage intensity model andusage imbalance model respectively We divide location ofShanghai into three levels in proportion as 25 50 and25 firstly Then combining results of two models thelocation can be divided into five levels prior recommendedmedium not recommended avoid as Figure 7 shows Wefind that central area takes a relatively large proportion ofprior level area to locating Given that a lot of central areas
of city are appropriate to locating carsharing station andcarsharing ismore efficient in using parking space we suggestthat more carsharing exclusive parking space can be usedto replace public parking space to decrease usage of privatevehicle and save parking space Moreover many suburblocations are evaluated as prior or recommended level bymodels This means that the usage scenarios of carsharingare wider than central area If these suburb areas can bedeveloped adequately the usage scenarios related to outskirtswill take on more trips
5 Conclusions
This study focused on the largest station-based OWC pro-gram in Shanghai China There are many approaches toestimate carsharing demand according to research objects
12 Journal of Advanced Transportation
Table 4 The results of imbalance model
Features Workday Nonworkday06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h 06ndash09 h 10ndash15 h 16ndash19 h 20ndash01 h
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
However the station-based one-way system is rarely investi-gated Meanwhile many research investigations focus on theusage rate vehicle hour traveled (VHT) and many othersbut the station usage imbalance has not yet been investigatedThis study addressed this gap
In this study multiple linear regression models and betaregression model are developed to analyze how differentfactors affect station usage intensity and degree of stationimbalance across different periodsThe conclusions are sum-marized as follows
(1) The attributes of spatial unit constantly appear tohave significant effect on the demand characteristicsHowevermany built environment and transportationfactors have a different effect on the demand indifferent time periods This is the main reason whycarsharing demand appears to be dynamic across timeperiods
(2) For usage intensity the university high POI-mixedentropy high intersection density area and areaincluding a metro station bus stop car rental stationand intercity coach have positive influence on usageintensity However industrial residential culturepublic authority and medical hygiene areas shownegative effect in different time periods in whichlayout should be avoided by carsharing stations
(3) For the degree of usage imbalance it will decreasealong with the increase in operation age Limited-access parking space enhances usage imbalance Res-idential public authority business and industrialareas continually play a negative role to increase thedegree of imbalance in special time period The areawith the university high intersection density highPOI-mixed entropy and more local streets and one-way roads lead to more balanced operational area
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Advanced Transportation 13
BalanceImbalance
High intensity
Low intensity
Operational age
College and University
POI-mixed entropy
Intersection density
Average trip distance
Car rental
Metro station
Business
Social amp Recreation
Limited access
Residential
Industry
Exclusive parking space
Medical
Public authority
Intercity coach
Bus stop
Culture Public parking
Shopping
Local street
One-way road
Two-way road
Arterial road
Secondary road
Area attributesBuilt EnvironmentTransportation
Figure 6 Influence diagram of statistically significant independent variables
(4) Areas with adequate public parking space will attractmore personal vehicle use rather than carsharingtrip Given that carsharing is more efficient in usingparking space we suggest that public parking spacesshould be gradually converted to carsharing exclusiveparking space This will increase the usage efficiencyof the limited number of parking spaces and reducepersonal vehicle usage while having a flexible car tripstill available
(5) For public transportation the metro and bus aresignificantly different for carsharing The metro has astrong advantage over carsharing in the morning andevening peak on workdays because of its certainty oftrip durationThus carsharing cannot attract passen-gers from the metro in rush hour Meanwhile theyappear to connect with each other in another timeperiod which is a complementary relationship How-ever the bus is similar to carsharing which runs onthe ground but lacks the comfort and personality ofcarsharing Thus carsharing has a related advantageover the bus which results in some bus passengerstransferring to carsharing unidirectionallyThereforewe suggest that the government should encouragecarsharing station layout near a metro station but nota bus stop
Usage intensity is related to profits and the degree of stationimbalance is related to dispatching cost From the carsharingoperator viewpoint the purpose of the carsharing station isto minimize the cost to obtain the maximum benefit Thusthe results shown in Figure 6 can be viewed as a guidanceof carsharing station layout for maximizing benefit Thefeatures in the first quadrant lead to higher usage intensityand lower imbalance degree meanwhile features in the thirdquadrant result in lower usage intensity and higher imbalancedegreeTherefore carsharing station should be given priorityto locating at area with features in the first quadrant andsetting up stations in areas with features in the third quadrantshould be avoidedOther factors can be selected as secondarysuch as stations nearby metro stations which only decreasestation usage intensity during peak time section onworkdaysHowever it might be a good choice to select the station nearother stations so that the imbalance level can be dramaticallydecreased during most of the time sections
The method of modeling for different time sectionsreveals to a certain extent the temporal dynamics patternsof the demand which can provide guidance for vehiclerelocation In college and university areas the imbalance levelis high at 600ndash900 on nonworking days which shows thatextra dispatch is needed during this time section Howeverthe research conclusion is built upon long-termmeasurement
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
14 Journal of Advanced Transportation
Intensity
1886ndash37801320ndash18860ndash1320
LevelAvoidNot recommendedMedium
RecommendedPrior
Imbalance020ndash040040ndash075075ndash100
Figure 7 Combining usage intensity model and imbalance model to locating carsharing station in Shanghai
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Advanced Transportation 15
(three months) Thus it can provide a noninstant dispatchstrategy We believe that it is strategically advantageousto arrange vehicle in advance based on demand dynamicspattern concluded by this research Then an instant dispatchmethod is used for adjustment accordingly
There are three main limitations in this research
(1) The statistics radium station is 800m and it onlyrefers to the value in the research of public transitAlthough the range of 800m iswidely used in carshar-ing areas [24] the service range of carsharing stationsin different zones and different traffic conditions canvary
(2) The categorization of time section is only based on thetime distribution feature of bookings but more rea-sonable time categorization shall be an improvementdirection
(3) In the calculation of station imbalance level statistictime interval is very important Too small intervalmight cause high imbalance level while too biginterval may cause low level of imbalance We inferthat statistic time interval should depend on differentusage intensities in each spatial unit but this limita-tion will be improved in future research
Conflicts of Interest
The authors declare that they have no conflicts of interestregarding the publication of this paper
Acknowledgments
The authors would like to acknowledge the Shanghai Inter-national Automobile City Co Ltd and Global Carsharing ampRental Co Ltd for providing the precious data of EVCARDin this researchThis study is supported by theNational Natu-ral Science Foundation of China (71734004) China NationalKey Technology RampD Program (2015BAG11B01) and OpenResearch Funding of ldquoGaofengrdquo Discipline (2016J012307)
References
[1] A Millard-Ball ldquoWhere and how it succeedsrdquo TransportationResearch Board 2005
[2] E Martin S Shaheen and J Lidicker ldquoImpact of carsharingon household vehicle holdings Results from North Americanshared-use vehicle surveyrdquo Transportation Research RecordJournal of the Transportation Research Board vol 2143 pp 150ndash158 2010
[3] J T Schure F Napolitan and R Hutchinson ldquoCumulativeimpacts of carsharing and unbundled parking on vehicle own-ership and mode choicerdquo Transportation Research Record no2319 pp 96ndash104 2012
[4] S A Shaheen C Rodier and G Murray Carsharing and PublicParking Policies Assessing Benefits Costs and Best Practices inNorth America 2010
[5] E W Martin and S A Shaheen ldquoGreenhouse gas emissionimpacts of carsharing in North Americardquo IEEE Transactions on
Intelligent Transportation Systems vol 12 no 4 pp 1074ndash10862011
[6] HNijland J VanMeerkerk andAHoen Impact of Car Sharingon Mobility and CO2 Emissions PBL Note 2015
[7] A Bieszczat and J Schwieterman Are Taxes on CarsharingToo High A Review of the Public Benefits and Tax Burdenof an Expanding Transportation Sector Chaddick Institute forMetropolitan Development DePaul University 2011
[8] J Firnkorn and M Muller ldquoFree-floating electric carsharing-fleets in smart cities The dawning of a post-private car era inurban environmentsrdquo Environmental Science amp Policy vol 45pp 30ndash40 2015
[9] G D Kim J Park and J D Woo Investigating the Charac-teristics of Carsharing Usage Pattern for Public Rental HousingComplexes A Case Study in South Korea 2017
[10] F Ferrero G Perboli and A Vesco Car-Sharing ServicesmdashParta Taxonomy and Annotated Review Montreal Canada 2015
[11] R Katzev ldquoCar Sharing ANewApproach toUrban Transporta-tion Problemsrdquo Analyses of Social Issues and Public Policy vol3 no 1 pp 65ndash86 2003
[12] C Costain C Ardron and K N Habib ldquoSynopsis of usersrsquobehaviour of a carsharing program A case study in TorontordquoTransportation Research Part A Policy and Practice vol 46 no3 pp 421ndash434 2012
[13] K M N Habib C Morency M T Islam and V Grasset ldquoMod-elling usersrsquo behaviour of a carsharing program Application ofa joint hazard and zero inflated dynamic ordered probabilitymodelrdquo Transportation Research Part A Policy and Practice vol46 no 2 pp 241ndash254 2012
[14] A De Lorimier and A M El-Geneidy ldquoUnderstanding thefactors affecting vehicle usage and availability in carsharingnetworks a case study of communauto carsharing systemfrom Montreal Canadardquo International Journal of SustainableTransportation vol 7 no 1 pp 35ndash51 2012
[15] K Kim ldquoCan carsharing meet the mobility needs for thelow-income neighborhoods Lessons from carsharing usagepatterns in New York Cityrdquo Transportation Research Part APolicy and Practice vol 77 pp 249ndash260 2015
[16] J Kang K Hwang and S Park ldquoFinding factors that influencecarsharing usage Case study in seoulrdquo Sustainability vol 8 no8 p 709 2016
[17] R Seign and K Bogenberger ldquoModel-Based Design of Free-Floating Carsharing Systemsrdquo in Proceedings of the Transporta-tion Research Board 94th Annual Meeting 2015
[18] M Khan and R MachemehlThe Impact of Land-Use Variableson Free-Floating Carsharing Vehicle Rental Choice and ParkingDuration Seeing Cities Through Big Data Springer Interna-tional Publishing 2017
[19] S Schmoller and K Bogenberger ldquoAnalyzing External Factorson the Spatial and Temporal Demand of Car Sharing SystemsrdquoProcedia - Social and Behavioral Sciences vol 111 pp 8ndash17 2014
[20] S Wagner T Brandt and D Neumann ldquoIn free float Devel-oping Business Analytics support for carsharing providersrdquoOMEGA -The International Journal ofManagement Science vol59 pp 4ndash14 2016
[21] K Klemmer S Wagner C Willing and T Brandt ExplainingSpatio-Temporal Dynamics in Carsharing A Case Study ofAmsterdam 2016
[22] S Schmoller SWeikl JMuller andK Bogenberger ldquoEmpiricalanalysis of free-floating carsharing usage The munich andberlin caserdquoTransportation Research Part C Emerging Technolo-gies vol 56 pp 34ndash51 2015
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
16 Journal of Advanced Transportation
[23] T Stillwater P L Mokhtarian and S A Shaheen ldquoCarsharingand the built environment Geographic information systembased study of one US operatorrdquo Transportation ResearchRecord Journal of the Transportation Research Board vol 2110pp 27ndash34 2009
[24] C Celsor and A Millard-Ball ldquoWhere does carsharing workUsing geographic information systems to assess market poten-tialrdquo Transportation Research Record Journal of the Transporta-tion Research Board vol 1992 pp 61ndash69 2007
[25] Y Jiang P Gu F Chen et al Measuring Transit-OrientedDevelopment in Quantity and Quality A Case of 24 Cities withUrban Rail Systems in China 2017
[26] R Cervero and K Kockelman ldquoTravel demand and the 3Dsdensity diversity and designrdquo Transportation Research Part DTransport and Environment vol 2 no 3 pp 199ndash219 1997
[27] R Ewing and R Cervero ldquoTravel and the built environmenta meta-analysisrdquo Journal of the American Planning Associationvol 76 no 3 pp 265ndash294 2010
[28] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001
[29] P Geurts D Ernst and L Wehenkel ldquoExtremely randomizedtreesrdquoMachine Learning vol 63 no 1 pp 3ndash42 2006
[30] H Zou and H H Zhang ldquoOn the adaptive elastic-net with adiverging number of parametersrdquoAnnals of Statistics vol 37 no4 pp 1733ndash1751 2009
[31] H Zou and T Hastie ldquoRegularization and variable selection viathe elastic netrdquo Journal of the Royal Statistical Society vol 67 no2 pp 768-768 2005
[32] H Zou ldquoThe Adaptive Lasso and Its Oracle Propertiesrdquo Publi-cations of the American Statistical Association vol 101 no 476pp 1418ndash1429 2006
