The feeder network design problem:Application to container services in the BlackSea region
O l c a y P o l a t a, H a n s - O t t o Gü n t h e r b a n d O sma n Ku l a ka
aDepartment of Industrial Engineering, Pamukkale University, Kinikli,Denizli , 20070 Turkey.E-mail: [email protected] of Industrial Engineering, Seoul National University,Seoul 151-744, Korea.
Abst rac t Global containership liners design their transportation service as hub-and-
spoke networks to improve the access to local transportation markets and to reduce
operational costs by using short-sea connections for low-volume transportation lanes.
These connections from the hub ports to the regional ports constitute the feeder network
that is serviced by small- or medium-sized feeder containerships. In our case study
investigation, we assume the feeder network design problem of a Turkish short-sea
shipping company, in view of the opening of the new Candarli port near Izmir. The cost
performance of three alternate feeder network configurations serving the Black Sea region is
compared. For this purpose, a mixed-integer linear programming model is developed and an
adaptive neighbourhood search algorithm is applied in order to determine the feeder ship
fleet size and mix with routes and voyage schedules to minimize operational costs for a given
planning period. Numerical results show that the new Candarli port has great potential as
Ocean shipping is the most important mode of transport in international trade.More than 80 per cent of international trade in 2010 was carried overseas(UNCTAD, 2012). In comparison to other modes of freight transport like truck,air, rail and pipeline, ships are preferred for moving large amounts of cargo overlong distances. Among the different modes of ocean going transportation,container shipping is by far the most important one in terms of value of thecarried goods compared to bulk, tanker and general cargo shipping. Consideringthe large amounts of cargo and the continually increasing size of containervessels, global container shipping liners typically operate roundtrip lines, linkingmajor ports in different continents.
The introduction of mega containerships on the main international searoutes between major seaports has made it necessary to temporarily storecontainers in a specific region and to distribute them on short-sea routes.Therefore, in addition to the location of hub ports, regional feeder containershipservice is a critical issue in designing global networks of shipping lines. Inconceptual terms, the feeder containership service collects and drops containersin a specific region with small- and medium-sized container ships and feedsmega containerships, so as to avoid their calling at too many ports. In certainregions, it was the containership feeder line that made the entire containerservice economically rational, efficient and more profitable, consequentlycheaper and timely for the end-users (Rudić and Hlača, 2005).
Feeder networks are often operated by subsidiary or third-party companiesspecializing in short-sea shipping. For establishing a feeder service system, forexample, in conjunction with a new port development project, an aspiringcompany faces the main decision problems highlighted in Figure 1. For a generaloverview of planning problems in liner shipping networks, cf. Christiansen et al(2007).
At the strategic planning level, the targeted markets and trade lanes have tobe selected and respective long-term forecasts of the transportation volume have
to be developed. Considering the economic progress of a region and theprospects of international trade relationships, as well as trends in the choice oftransportation modes, scenarios reflecting the future development of demand forcontainer traffic between the regional ports and destinations outside the regionhave to be defined. These scenarios are used to evaluate different feeder networkconfigurations, in particular the strategic options for the selection of hub ports.The latter decision is not only made based on quantitative factors like costs andshipping distances, but also on the competitive situation in the regional short-seashipping market. According to the hub port options, ship types and theirownership, company-owned or chartered, have to be determined.
A key decision problem at the tactical level is contract management thatinvolves the analysis and the development of the existing contract relationshipswith ports in the region and with cooperating companies in the network, forexample, service providers in other transportation modes. Accordingly, targetshave to be defined for the container transportation volume handled by theprojected feeder network and for the necessary service frequencies. A key issue atthe tactical level is ship routing and scheduling. In this step, the sequence of portsto be visited has to be assigned to the individual routes and the timing of the portcalls has to be determined. Fleet deployment is concerned with the allocation ofships in the fleet to routes in the network.
Finally, at the operational planning level, the most economic sailing speedand the exact route, considering environmental factors like water depths, wind,wave and currents, have to be determined along with minor schedule adjust-ments with respect to the actual transportation orders. Additional planning tasksinclude stowage planning, that is, positioning of containers on board the vessel,and repositioning of empty containers.
In literature, most of the contributions focus on specific issues at one of theplanning levels while an integrated view, for example, including aspects of strategicand tactical decision making is missing (Fagerholt et al, 2010). In our article, weaddress the strategic choice of the hub port, decisions on the size and compositionof the fleet of containerships, and ship routing and scheduling as an integratedplanning problem. We consider the Black Sea region as an application example toanalyse the design problem of container feeder networks from the perspective of afeeder company commencing its services from a newly constructed port. The articleshows how decision support can be provided by application of a tailoredoptimization algorithm that determines, among others, the routes in the regionalnetwork and the allocation of shipping volumes between the different routes.
Feeder networks are mostly organized as hub-and-spoke systems whereregional ports are serviced from a centralized terminal by small- or medium-sizedcontainerships. The hub also serves as a transhipment terminal for the exchange ofcargo between the intercontinental lines and the regional ports. Figure 2 shows a
typical hub-and-spoke feeder network consisting of both shuttle and cyclic routes.There are two shuttle services connecting the hub to the feeder ports 5 and 11,respectively, and several cyclic routes that include a number of feeder ports. Shuttleroutes are usually established between ports showing a relatively high transporta-tion volume. They offer the lowest transit time, but typically require more feedersand smaller feeder ships, while cyclic routes are more appropriate for feeder portswith lower transportation volume. Cyclic routes benefit from economies of feedership size, but incur longer shipping distances and longer transit times.
In this article, we specifically consider the design of hub-and-spoke networksas shown in Figure 2. Feeder ships simultaneously deliver and pick-up containersat each port in the route. Hence, the loading capacity of the vessels has to beconsidered on each arc of the route. Further side constraints arise from the timedeadlines for arriving at a feeder port and for returning to the hub port. Sincecontainer shipping involves considerable capital investments and daily operatingcosts, the appropriate feeder network design (FND) is a crucial factor for theprofitability of the liner companies.
The remainder of this study is structured as follows. In the second section,the relevant literature is briefly reviewed. In the third section, a mathematicalproblem formulation of the FND problem is given. Next, a heuristic solutionprocedure is proposed in the fourth section. The fifth section introduces the casestudy application and presents detailed numerical results. Finally, concludingremarks are given in the sixth section.
Hub port
Feeder port Feeder service - Cyclic
Feeder service - Shuttle
8
9
10
7
0
11
6
4
1
5
3
2
Figure 2: Feeder service network organized as hub-and-spoke system.
Liner shipping has become a popular topic of academic research worldwide.Hence, a wealth of papers has been published, focusing on different planningaspects in this area. A large number of papers addresses ship routing andscheduling (cf. Ronen, 1983, 1993; Christiansen et al, 2004, 2013; Kjeldsen,2011 for comprehensive reviews). Additional review papers appeared oncontainer shipping (Notteboom, 2004); fleet composition and routing (Hoffet al, 2010); and liner shipping network design (Ducruet and Notteboom, 2012;Yang et al, 2012). We therefore do not provide a comprehensive literature reviewof liner shipping, but refer only to these review papers.
Nowadays feeder services play an irreplaceable role in global shipping net-works. The related literature primarily addresses regional conditions and prospectsin the development of feeder services. For instance, Robinson (1998) discusses thedevelopment of Asian ports between 1970 and 2000. Frankel (2002) analyses thefeeder service structure of three Caribbean ports and compares the economies ofshuttle, cyclic and pendulum feeder services with direct trunk connections. Ridolfi(1999) and Jadrijević and Tomašević (2011) examine problems and prospects ofglobal and regional shipping lines in the Mediterranean region. Further, Bukša andKos (2005), Rudić and Hlača (2005) and Hlača and Babić (2006) evaluate the effectof feeder services, current conditions and future prospects for the Adriatic region.More recently, Varbanova (2011a, 2011b) analyses transportation conditions andefficiency factors of feeder lines in the Black Sea region. The author discussesoperational planning problems of container feeder lines, for example, the scheduleintegrity between feeder and trunk lines, shipping capacity utilization, environ-mental issues, and shipping policy initiatives of the European Union.
Just a few papers deal with real feeder service network settings and considerexisting or projected hub locations. In Table 1, an overview of these papers isgiven. For instance, Sambracos et al (2004) present a case study of dispatchingsmall packages via coastal freight liners from a hub port to 12 Greek island ports.In this study, total operating costs, including fuel consumption and port charges,are minimized assuming a homogeneous fleet and a given traffic demand. Theauthors formulate a linear programming model which is solved by use of a list-based threshold acceptance heuristic. Later, Karlaftis et al (2009) generalized thiscontainer dispatching problem by minimizing total travel distances with simulta-neous container pick-up and delivery operations and time deadlines. To solve theproblem, they propose a mixed-integer programming formulation and a geneticalgorithm assuming soft time limits, that is, tolerating violations of certainconstraints. As a case study, they deal with a feeder network in the Aegean Seawith 26 ports including 1 mainland hub port. However, neither study considersheterogeneous ship types and the operating costs of the fleet.
Most recently, environmental issues have received considerable attention innetwork design and optimization (Windeck, 2013). The respective research hasbeen driven by the pressure on ship operators to reduce fuel consumption andthereby greenhouse gas emissions. These goals can be reached by seeking acompromise between economic and environmental factors in determining thesailing speed (cf. Psaraftis and Kontovas, 2013 for a survey and an overview ofrespective model formulations). In addition, routing and scheduling decisionscan be affected by waves, ocean currents, wind and tides, for instance. Theseissues, however, are disregarded in our investigation because the primary focusis on the cost performance of hub–and-spoke networks.
In the academic literature, only a few studies treated the feeder serviceproblem as a variant of vehicle routing with simultaneous pick-up and delivery(VRPSPD). One example is the paper by Fagerholt (1999, 2004) who proposes tosolve the problem by first initializing all feasible single-ship routes and then, inthe second step, allocating ship types to these routes. Contrary to the classicvehicle routing problem (VRP), ships can operate on more than one route withinthe allowed time limit (VRPMTTL). The author solved instances with up to 40ports and 19 ships. Jin et al (2005) address the feeder ship routing problem as aVRP with pick-up and delivery with time windows (VRPPDTW). The authorspropose a mixed-integer programming model with the objective to minimize theweighed sum of total travel times, the number of ships used, total waiting timeand total tardiness. The proposed model is solved for instances of up to 100 portsby using variable neighborhood search (VNS) and tabu search. Later, Sun and Li(2006) handled the same problem by using an Immune Genetic Algorithm andimproved the solutions to Jin et al’s (2005) test instances.
However, so far none of the feeder service network design studies haveconsidered detailed variable and fixed cost structures for a heterogeneous vesselfleet over a complete sailing horizon. In this study, we propose a mixed-integerlinear programming (MILP) model to simultaneously determine fleet size andmix, fleet deployment, ship routing and ship scheduling in feeder servicenetwork design by minimizing total network costs in a sailing season. A hybridsolution heuristic called ANS is proposed to solve the joint problem using theBlack Sea region as a case study.
The FND Prob lem
In a feeder network, ships visit a number of ports along the predefined routesconnecting ports in the region. In the design of the feeder network factors such asthe capacity of feeder ships, characteristics of the ports, transport demand at thevarious ports, as well as bunker costs and operating and chartering costs of the
ships have to be considered. Specifically, the FND problem can be described asfollows. A set of feeder ports is located on a distribution network where feederports require both delivery and pick-up operations. Each feeder port has to beserved once for both operations with a given fleet of heterogeneous feeder ships.Each ship leaves the hub port carrying the total amount of containers it has todeliver and returns to the hub port carrying the total amount of containers pickedup on the voyage. Each feeder or hub port also has a specific operational efficiencyfor loading and unloading containers. The service time of a port depends on itsoperational efficiency, ship size, the number of loaded and unloaded containersand the pilotage time for entering and exiting the port. Therefore, total voyageduration consists of the total travel time of the route and the total service time at thehub and the feeder ports. The voyage starts in the hub port with the commence-ment of loading operations, and ends also there with the completion of unloadingoperations. Each vessel has to finish its voyage before the allowed time deadline isreached. Before starting a new voyage, the ship needs an off-hire interval forpossible repairs, cleaning, waste disposal and so on.
According to these considerations, the FND problem has similarities withthe vehicle routing problem with simultaneous pick-up and delivery with timelimit (VRPSPDTL). For a review and classifications of VRPs, see, for example,Berbeglia et al (2007) and Parragh et al (2008). While the VRPSPDTL minimizestotal voyage distance, the FND problem aims to serve all contracted feeder portsby minimizing total operational costs for a given planning period. For a feedernetwork provider, operational costs include containership related fixed costsfor the necessary number of ships (chartering, operating and so on) and totalservice-related variable costs (bunker costs at sea and in port, port chargesdepending on the ports visited in a route). Table 2 shows the related basic costdefinitions.
Since our investigation is concerned with the design of a real-worldcontainer feeder network, some assumptions have to be made in order to excludeelements of minor relevance and to focus only on the salient aspects.
Table 2: Basic definitions of total costs for a sailing season
Parameter Basic calculation
Total costs Fixed costs+ Variable costsFixed costs Number of necessary ships*(Chartering+ Operating costs)Variable costs Number of services*(Bunker (sea)+Bunker (port)+ Port charges)Number of required ships |(Voyage duration+ Off-hire duration)/Service frequency|Number of services |Planning period/Service frequency|Voyage duration At-sea duration+ In-port duration (feeder)+ In-port duration (hub)Idle duration Number of necessary ships*Service frequency− (Voyage+Off-hire duration)Ship total duration Voyage duration+Off-hire duration+ Idle duration
● All parameter values are deterministic, that is, we exclude weather andseasonal effects.
● No direct delivery takes place between feeder ports.● Queuing time at ports is not considered.● Time windows for berthing at a port are disregarded as they are not known in
advance for a complete sailing season.● Demand and supply quantities of feeder ports cannot be split between
different ships.● The fleet of ships is heterogeneous. However, ships of the same type are
identical regarding their carrying capacity.● Bunker costs are the same in all ports.● We only consider chartered ships.● Set-up time (pilotage, berthing, cleaning and so on) of a port depends only on
the type of ship.● Effects of speed-dependent fuel costs as well as strait/canal durations and
costs are not considered.
A MILP formulation of the FND problem is presented using the followingnotation:
Indices and sets
i, j∈N The set of ports (0 represents the hub port)s∈ S The set of containership types(i, j)∈ L The set of allowed voyage legs between portsr∈R The set of routes (|R|<|N|)
Parameters
f Service frequency daysγ Number of services in a sailing seasonD Duration of the planning period daysK Maximum allowed voyage duration hoursvis Vessel set-up duration of ship type s in port i (pilotage,
berthing, cleaning and so on)hours
us Off-hire duration of ship type s hoursms Available number of containerships of ship type s shipsqs Loading capacity of ship type s TEUhs Average travel speed of ship type s Knotsosi Operational efficiency of port i for ship type s TEU/hourwij Distance between ports i and j n.miletsi Berthing duration of ship s at port i hoursdi Daily container demand (delivery) of port i TEU
pi Daily container supply (pick-up) of port i TEUα Main fuel oil price $/tonβ Auxiliary fuel oil price (distillate) $/tonccs Chartering cost of ship type s $/shipfcs Operating cost of ship s (administration, maintenance,
lubricants, insurance and so on)$/ship
as Main fuel consumption of ship type s at sea ton/n.milebs Auxiliary fuel consumption of ship type s at berth ton/hoursbcsi Port charges for ship type s at port i $/ship
Decision variables
xijrs 1, if the arc between ports i and j belongs to route r served by ship
type s (0, otherwise)binary
yij containers picked up from ports up to port i and transported fromport i to j
integer
zij containers to be delivered to ports routed after port i and transportedbetween port i and j
integer
ers Required number of ships of type s on route r shipscrs Voyage cycle time of route r with ship type s hoursFC Total fixed costs of a sailing season $VC Total variable costs of a sailing season $
xrsij 2 0; 1f g; yij; zij; ers 2 Z + ; crs ⩾ 0 8i; j 2 N; i; jð Þ 2 L; r 2 R; s 2 S (16)
The objective function (1) minimizes total costs of the network for a sailingseason. Equations (2) and (3) define fixed and variable costs, respectively. Thenecessary number of ships needed for a full service cycle on each route iscalculated in Equation (4). Equation (5) determines the cycle time of ships oneach route (berthing duration+service duration+voyage duration). Equation(6) ensures that each feeder port is served by only one type of ship and one route.Equation (7) guarantees that a ship arrives at and departs from each feeder porton each route. Equations (8) and (9) impose a similar condition for the hub portat which the route starts and ends. Equation (10) is the vehicle sub-tour
elimination constraints according to Karlaftis et al (2009). Constraint (11)represents an upper bound for the number of ships employed from each type.Equations (12) and (13) satisfy pick-up and delivery demand of containers at thefeeder ports, respectively. Equation (14) is the ship capacity constraints.Equation (15) represents the maximum voyage duration constraints. Finally,Constraint (16) defines the variable domains. In general, the constraints ensurethat each ship departs from the hub with a load equivalent to the total delivery ofcontainers, and each ship returns to the hub with a load equivalent to the totalpick-up containers from feeder ports in the route served by that ship.
So lut ion Approach
Methodology
The FND problem presented in the previous section is a highly complexcombinatorial optimization problem and thus hard to solve by use of standardoptimization software. Exact methods for solving the FND problem are generallynot practical for large instances because of problem complexity. In this study, wetherefore employ an ANS heuristic that has shown to be very efficient for solvingthe VRPSPDTL (Polat et al, 2012). In each solution step, total costs expressed inEquations (2) and (3) of the optimization model are applied. By use of the ANSheuristic, the fleet size and mix along with the routes and schedules of the vesselsin the feeder network are determined. In the case study application presented inthe next section, the ANS heuristic is employed in order to determine theresulting costs and network configuration for each of the considered hub options.
The proposed ANS approach applies the Savings Algorithm of Clarke andWright (1964) in order to gain a fast and effective initial solution. This classicheuristic aims at merging sub-tours based on cost savings which can be achievedby combining two sub-tours to be served by one vehicle. In the literature, someenhancements of the Clarke andWright savings algorithm have been suggested byadding new terms and parameterizing the savings formula. Since the FND problemis a generalization of the VRP, we construct our initial solution according to thesavings formula proposed for the VRP by Altinel and Öncan (2005).
In the next stage, the initial solution is improved with enhanced variableneighborhood search (EVNS). The EVNS is an adapted version of the VNSapproach of Mladenović and Hansen (1997). VNS is based on the idea ofsystematically changing the neighbourhoods in order to improve the currentsolution and aims to explore the solution space that cannot be searched by localsearch (Hansen et al, 2010). Kytöjokia et al (2007), Hemmelmayr et al (2009) andStenger et al (2012) showed the effectiveness of VNS in VRP applications.
Shaking, local search and move or not operators are used in the implementationof the VNS. The shaking operator defines the search direction of the VNS byusing the set of neighbourhoods. The possibility of reaching a global solutionincreases when combining the shaking operator with local search, rather thanusing a single shaking operator. Therefore, each solution obtained through theshaking operator is used in the local search operator in order to explore newpromising neighbourhoods of the current solution. In this study, we implemen-ted the Variable Neighborhood Descend (VND) algorithm as the local searchoperator. The VND aims to combine the set of neighbourhoods in a deterministicway, as using more than one neighborhood structure could obtain a bettersolution (Hansen and Mladenović, 2001).
The previously obtained local optimum solution combines global statisticalinformation and local information of good individual solutions. In this study, thecurrent solution is therefore used to develop a novel perturbation method calledAdaptive Perturbation Mechanism. This perturbation mechanism runs after anumber of non-improving iterations counted from the last improving iteration. Inaddition to the perturbation move, a local optimization method, with thepreviously defined four intra-route neighborhood structures, is applied in orderto improve the perturbed solution quality.
Implementation
The candidate networks created by the ANS are evaluated using a fitnessfunction. Since ANS is originally intended to solve the VRPSPDTL with homo-genous vehicles under the objective to minimize the total travel distance withinthe network, it is necessary to adjust the fitness function of the ANS. In ourimplementation of the ANS, total operating costs of all routes for the entire sailingseason, according to the cost functions (2) and (3) of the third section, are usedas fitness function. The respective procedure for calculating the fitness values issummarized in Figure 3. In the VRPSPDTL application, candidate routes aregenerated with the help of neighborhood structures. In this step, Constraints (6)–(16) of the optimization model are checked in order to achieve feasible solutions.
Apart from the network routes, the ANS determines the fleet mix, thenumber of required ships according to Equation (4) and their deployment toroutes in the candidate network. On the basis of these data, the total voyage cycleof a ship on a route is achieved as given by Equation (5), that is, considering therelated port service times, travel times between ports, off-hire times and so on.Figure 4 shows an example of a route-ship-port schedule for 30 days of operationwith 5 days service frequency, three feeder ports in sequence and three requiredships for a route in the network.
The case of a feeder network in the Black Sea region
The fact that the considered region is surrounded by several seas – the BlackSea, Mediterranean Sea, Adriatic Sea, Ionian Sea, Aegean Sea and Marmara
1 procedure: fitness function for ANS 2 input: candidate network 3 output: total network costs of candidate network(dNC) 4 start5 for each route (r) in the candidate network6 initialize a big number for total route cost (RCr) 7 for each ship type (s)8 if (hub and each feeder port departure and arrival loads are feasible for ship type s on
route r) [Eq.(12), (13), (14)] 9 calculate voyage cycle time of route r with ship type s specifications [Eq. (5)]10 if (voyage cycle time is feasible by considering maximum voyage dur.limit) [Eq. (15)] 11 calculate required ship number of ship type s for route r [Eq. (4)]12 calculate variable costs for route r operated with ship type s [Eq. (3)]13 calculate fixed costs for route r operated with ship type s [Eq. (2)]14 calculate total costs of route r operated with ship type s (dRCr)15 if (dRCr < RCr )16 update RCr with dRCr
17 end if18 end if19 end if20 end for21 end for22 calculate total network costs of candidate network (dNC = RC ) [Eq. (1)]23 end
Figure 3: Calculation of the fitness function.
Figure 4: An example of a route-ship-port schedule.H.L.: Loading time at hub port; H-1, 1-2, 2-3, 3-H: Port-to-port travel time; 1, 2, 3: Feeder port servicetime including, loading, unloading and set-up times; H.U.: Unloading time at hub port; O.H.: Off-hiretime of a ship until the next voyage.
Sea –makes maritime shipping a prime area for sustained growth (see Figure 5).Container feeder shipping lines offer crucial transport connections between thehinterland of this region and global trunk shipping lines. The feeder shippingdynamics of the region are mainly related to container transportation of the trunkshipping lines between Far East and Europe. In recent years, in parallel to thistraffic, an increase in total container port throughput has also been observed inthe regional feeder ports. This is particularly true for ports in the Black Searegion. Hence, the outlook for the maritime transportation market in the region isvery promising (Varbanova, 2011a).
Turkey’s ideal location between Asia and Europe gives its ports a competi-tive advantage and opportunity to develop into major transhipment hub ports.However, so far Turkish ports primarily serve their national needs and remain
MersinAntalya
CandarliIzmir
Gebze
Ambarli Haydarpasa
Gemlik
Thessaloniki
Piraeus
Limassol
Lattakia
Beirut
Haifa
Ashdod
Port SaidAlexandria
Burgas
Varna
IlyichevskOdessa
Novorossiysk
Poti
Trabzon
Aliaga
Aegean Sea
Mediterranean Sea
Black Sea
Sea of Azov
Constantza
TURKEY
Damietta
Batumi
Sea of Marmara
Figure 5: Hub port locations (A= Port Said and B= Candarli) and regional feeder ports numbered 1–26.
outside the major trunk lines (Kulak et al, 2013). This situation results in maritimecontainer transport mainly carried by feeder lines that serve the Turkish ports fromthe East Mediterranean hub ports. In this regard, Turkey has significant potentialfor getting more involved in regional maritime transport, and consequently severalprojects for the development of intermodal transport are being initiated. One ofthese projects is the construction of a hub port in Izmir’s Candarli district in order toimprove Turkey’s hub port potential in the East Mediterranean and especially in theBlack Sea region. According to the project plan, the Northern Aegean Candarli portwill take its place among the world’s largest ports after its first part is completed in2013, and it will be able to handle 12 million tons of container freight annually in itsultimate configuration. The potential market areas of Candarli as a hub port canbe categorized into four sub-regions: the Black Sea, the Marmara Sea, the EastMediterranean and the Aegean Sea.
A particular company currently operates a feeder network with Port Said inNorthern Egypt as its hub. However, after the opening of the new Candarli port,the company will possibly redesign its current feeder network with Candarli as itsnew hub. Therefore, in this study three different strategic options for hub ports,faced by the company, are considered.
● The first strategic option corresponds to the current configuration with PortSaid as hub for feeder ports in the Black Sea region. The distinguishingcharacteristic of this option is the closeness to the Suez Canal through whichalmost all of the Asia–Europe shipping routes pass.
● In the second strategic option, the new Candarli port replaces Port Said as a hubport for the Black Sea region. This option is based on the assumption thatCandarli will be a firm part of global shipping routes.
● The third strategic option is a mixed case in which two hub ports areestablished. Namely, Port Said serves as a link to the main global shippinglines and at the same time as regional hub port for the East Mediterraneanports. Candarli will serve as a second regional hub port for the Black Sea, theSea of Marmara and the Aegean Sea ports and with daily direct connections toPort Said via mid-sized ships.
These strategic options are tested under different time deadline and servicefrequency conditions for a 52-week sailing season. In this region, the feedercompany concerned has 36 contracted container terminals at 26 feeder ports,with a total daily demand of 3321 TEU and a total daily supply of 2151 TEU onaverage (see Table A1 in the Appendix for details of the ports). Because of thelimited berth depth at some regional ports and well-known traffic bottlenecks atthe Bosporus and Dardanelles straights, ships of three different sizes areconsidered in the numerical experiment. The major cost parameters for all shiptypes are shown in Table 3.
In order to provide decision support for the FND problem faced by the Turkishcompany, we proceed with our experiments in the following order. First, thestrategic options for choosing the hub port are evaluated (the section ‘Strategicoptions for feeder networks’). Second, the impact of different scenarios for thelong-term development of transportation volume in the Black Sea region isanalysed (the section ‘Demand scenarios’).
Since solution times of the optimization model formulated in the thirdsection turned out to be prohibitive except for very small problem instances, wedecided to apply the heuristic solution algorithm outlined in the fourth section.The algorithm was coded using Matlab R2010b/Visual C# 4.0 and executed on anIntel Core 2 Duo T5750 2.0 GHz processor with 3 GB RAM. For each probleminstance, the algorithm was run 10 times with different random number seeds.Computational times depending on the structure of the feeder network variedbetween 10 and 60 seconds.
Strategic options for feeder networksThe three basic strategic options to be considered are the locations of hub port inPort Said and Candarli, respectively, and a combined network design with thesetwo transhipment hubs connected by a shuttle service of feeder ships. In theexperiments, two additional network design parameters are evaluated. Forshipping frequencies, we compare the departure of services every 7 or 3.5 days,respectively. For each frequency, the time deadline for voyages is varied between3 and 4.5 weeks.
The specific research issues addressed in our numerical investigation are thefollowing.
● Does the time deadline imposed on the voyages represent a major factor in thedesign of the network configuration?
● How does the voyage frequency impact the cost performance of the variousnetwork configurations?
● Which of the three strategic options for hub ports would be favourable interms total yearly costs?
Table 4 shows the total costs of the current and alternate hub port optionsunder various time deadline and service frequencies. Total costs includechartering costs, operating costs, administration costs bunker costs, at sea andin port, and port charges for a 52-week planning period.
The first conclusion that can be drawn from the results displayed in Table 4is that the effect of the time deadline is practically negligible. Even the largestdeviation observed for Candarli and the 7-days frequency options (no. 9–12) areless than 1 per cent.
Table 4: Scenario results for alternative feeder network configurations
Option number Hub port Frequency (days) Deadline (weeks) Total cost (′000 $)
1 Port Said 7 3 286 548.472 Port Said 7 3.5 286 911.483 Port Said 7 4 286 420.044 Port Said 7 4.5 287 388.275 Port Said 3.5 3 339 726.796 Port Said 3.5 3.5 339 726.797 Port Said 3.5 4 339 726.798 Port Said 3.5 4.5 339 726.909 Candarli 7 3 257 526.7410 Candarli 7 3.5 255 733.2011 Candarli 7 4 255 341.8712 Candarli 7 4.5 255 341.8713 Candarli 3.5 3 296 789.8114 Candarli 3.5 3.5 298 157.2715 Candarli 3.5 4 296 831.3116 Candarli 3.5 4.5 296 789.8117 Mixed 7 3 368 529.8818 Mixed 7 3.5 367 462.3119 Mixed 7 4 367 099.6920 Mixed 7 4.5 367 462.3121 Mixed 3.5 3 403 355.7622 Mixed 3.5 3.5 401 789.6023 Mixed 3.5 4 401 789.6024 Mixed 3.5 4.5 401 789.60
Minimum costs for a hub port location highlighted in bold.
However, voyage frequencies have a major impact on cost performance.Reducing the voyage frequency for Port Said from 7 to 3.5 days causes a costincrease of 18.65 per cent. Respective values are 16.2 per cent for Candarli and9.44 per cent for the configuration with two hubs.
The main research question addresses the choice of the hub location. It can beseen from the results shown in Table 4 that the mixed hub option causes total costsof US$367 099 690 (option no. 19 with 7-day service frequency and 4 weeksdeadline) and thus is clearly outperformed by the single-hub configurations. Thiscost disadvantage is mainly due to the additional transhipment operations atCandarli. As for the single-hub configurations, the existing hub port option of PortSaid shows minimum total costs of $286 420 040 (option no. 3 with 7-day servicefrequency and 4 weeks deadline) while the projected hub port of Candarli achievesminimum total cost of $255 341 870 (option no. 11 with 7-day service frequencyand 4 weeks deadline). Considering only network-wide cost figures, the Candarlioption would allow cost savings of 12.2 per cent compared to the existing feedernetwork configuration with Port Said as hub. The resulting feeder routes for PortSaid (option no. 3) and Candarli (option no. 11) are shown in Figure 6.
Table 5 presents a comparison of costs, fleet and voyage characteristics ofthe two single-hub configurations. As in global trunk lines, feeder shipment ishighly sensitive to bunker fuel costs as they represent 26.67 per cent (Port Said)and 20.83 per cent (Candarli) of total costs. However, these shares aresignificantly lower compared to global trunk lines due to the density of thenetwork and the relatively short transportation distances. In turn, feeder net-works show a higher share of ship-based fixed costs such as chartering, operatingand port charges. Since Candarli has shorter distances to regional feeder ports,relatively small containerships are employed. In contrast, the Port Said-basedfeeder network utilizes slightly more mid-sized containerships. Large-sizedfeeder ships of 4300 TEU are not appropriate for either hub alternative because
0 20 40 60 80 100 120 1400
20
40
60
80
100
120
140
0 20 40 60 80 100 120 1400
20
40
60
80
100
120
140
Figure 6: Feeder route networks for Port Said (left) and Candarli (right).
of the relatively high fixed costs. It could be expected, however, that in case thenetwork dimension is enlarged and total demand increases, larger ships willbecome more attractive in order to meet the balance between fixed and variablecosts. It is also shown in Table 3 that total voyage durations of 297.23 hours areslightly lower for the Port Said option compared to Candarli with 308 hours. Inboth cases, the major share of voyage durations of more than 60 per cent occursfor the stay in the hub and in the feeder ports. As expected, the at-sea voyageduration is lower for the Candarli option due to its geographical location closer tothe Black Sea region. The best solution achieved for Candarli is given in detail inTable A2 in the Appendix.
A specific drawback of the Candarli option compared to Port Said is certainlyits location of about 220 nautical miles farther away from the main globalshipping lines. Under one daily East-Westbound and West-Eastbound serviceassumption, the extra costs for operating this transhipment service would almostcompensate the saving in operational costs.
Demand scenariosFor the future development of the feeder network, the expected growth of thetransportation market in the Black Sea region is an essential factor. According toforecasting reports of Ocean Shipping Consultants (2011), container handling
Table 5: Feeder network comparison of the Port Said and Candarli options
Parameter Port Said Candarli
CostsTotal costs (′000 $) 286 420.04 255 341.87Chartering costs (in percentage) 20.80 22.30Operating costs (in percentage) 23.62 25.42Bunker costs (on sea) (in percentage) 26.67 20.83Bunker costs (at port) (in percentage) 4.83 5.38Port charges (in percentage) 24.09 26.07
FleetNumber of routes 13 12Total number of ships 23 221200 TEU (in percentage) 20.44 27.272600 TEU (in percentage) 79.56 72.734300 TEU (in percentage) 0.00 0.00
VoyagesTotal average duration (hour) 297.23 308.00On sea (in percentage) 23.71 17.69In feeder ports (in percentage) 41.31 43.17In hub port (in percentage) 20.39 21.27Off-hire times (in percentage) 6.96 6.82Idle times (in percentage) 7.64 11.06
demand in the region will continue to increase yearly by 25 per cent till 2025.Therefore, a sensitivity analysis is performed to assess the influence of this factoron the cost performance of Port Said and Candarli network configurations. Onthe basis of this expectation for four sub-regions, namely, the Black Sea region,the Sea of Marmara region, the Aegean Sea and the East Med. sea region, 16different market scenarios are created to evaluate the network costs of changingthe current hub port. Scenario 1 corresponds to the current market situation. Infurther scenarios, combinations of market volume increase in one, two and threeregions, respectively, are assumed. Finally, Scenario 16 corresponds to a 25 percent volume increase in all four regions.
Results of the sensitivity analysis summarized in Table 6 show network costsfor the two candidate hub ports under equivalent demand increase assumptions.According to the results of the sensitivity analysis, Candarli outperforms PortSaid in all demand scenarios because of its advantageous geographical position.Candarli’s superiority, however, is considerably smaller when only a marketvolume increase in the East Mediterranean and the Aegean Sea region isassumed. Otherwise, Candarli benefits from increased market volumes in theBlack Sea and the Sea of Marmara region.
Table 6: Sensitivity analysis of market volume increase.
Scenarionumber
Assumed market volume increasea Total costs for alternative hub portsb
Black Searegion
Sea ofMarmararegion
AegeanSea region
EastMediterraneansea region
Port Said(´000 $)
Candarli(´000 $)
Difference(´000 $)
1 o O o o 286 420.04 255 341.87 31 078.172 + O o o 309 076.85 269 818.90 39 257.953 o + o o 298 126.92 268 435.94 29 690.984 o O + o 295 888.30 264 097.58 31 790.725 o O o + 298 868.95 273 022.57 25 846.386 + + o o 323 199.03 283 684.09 39 514.947 o + + o 310 919.54 275 603.45 35 316.098 o O + + 306 960.82 280 852.65 26 108.179 + O + o 318 259.61 279 308.36 38 951.2510 + O o + 320 500.69 288 631.05 31 869.6411 o + o + 312 501.39 283 657.26 28 844.1312 + + + o 332 576.84 291 443.00 41 133.8413 o + + + 322 131.10 291 084.21 31 046.8914 + + o + 334 665.42 299 057.31 35 608.1115 + O + + 333 242.55 295 495.20 37 747.3516 + + + + 344 605.35 308 725.32 35 880.03
ao indicates that the company will maintain its current market share; + indicates that the company willincrease its current market share in the region by 25 per cent.bThe results show the best of 10 replications of the heuristic for alternative hub ports with 4 weeks deadlineand 7-day service frequency for a 52-week sailing season.
In this study, we focus on the potential hub role of a new port (Candarli) inthe East Mediterranean and Black Sea region and apply a heuristic procedure tosolve the FND problem faced by a short-sea shipping company. On the basisof the container transportation demand at feeder ports, the feeder network andfleet mix, the composition of routes and the schedule of the vessels operatingon these routes are determined by minimizing total operating costs. A mathe-matical model of the FND problem has been developed. Because of the complex-ity of the optimization problem, an efficient heuristic solution procedure wasapplied.
In the numerical investigation, the cost performance of three strategicoptions for hub port configurations has been compared. From the numericalresults, it can be concluded that Candarli as a new hub port offers significant costsavings compared to Port Said which is currently used as a hub port by thecompany in question. However, these cost savings could be compensated withadditional transhipment cost for the Port Said–Candarli services that are neededto connect Candarli to the global trunk shipping lines. Therefore, additionalfactors like service quality and handling efficiency at the hub ports, as well aswaiting time in the queue of the hub ports play an important role in thedevelopment of the company’s feeder network configuration. Certainly, thenew Candarli port has great market potential as long as port authorities keepcontainer handling costs and service quality at a favourable level.
Acknowledgements
This work was partially supported by the German Academic Exchange Service(DAAD) with the grant number A/08/77565 and Ministry of Transport, MaritimeAffairs and Communications of the Republic of Turkey.
References
Altınel, K. and Öncan, T. (2005) A new enhancement of the Clarke and Wright savings heuristicfor the capacitated vehicle routing problem. Journal of the Operational Research Society 56(8):954–961.
Andersen, M.W. (2010) Service network design and management in liner container shippingapplications (Chapter 5). PhD thesis, Technical University of Denmark, Lyngby, Denmark.
Bendall, H.B. and Stent, A.F. (2001) A scheduling model for a high speed containership service: A huband spoke short-sea application. International Journal of Maritime Economics 3(3): 262–277.
Berbeglia, G., Cordeau, J.-F., Gribkovskaia, I. and Laporte, G. (2007) Static pickup and deliveryproblems: A classification scheme and survey. TOP 15(1): 1–31.
Bukša, J. and Kos, S. (2005) Analysis of the structure of transported containers in the Adriatic feederservice with a retrospect on the proportion and effect of empty containers on the service. ScientificJournal of Maritime Research 19(1): 89–99.
BunkerIndex. (2012) South Europe Bunker Prices, http://www.bunkerindex.com/prices/seurope.php, accessed 15 June 2012.
Catalani, M. (2009) Ship scheduling and routing optimization: An application to Western Mediterra-nean area. European Transport 42: 67–82.
Chou, M., Song, M. and Teo, C. (2003) Inventory-Routing Problem in Sea Freight: Direct VersusTransshipment Model. Asia Pacific: The Logistics Institute. Technical Report.
Christiansen, M., Fagerholt, K., Nygreen, B. and Ronen, D. (2007) Maritime transportation. In:C. Barnhart and G. Laporte (eds.) Transportation: Handbooks in Operations Research andManagement Science. Vol. 14. North-Holland, Amsterdam: Elsevier B.V., pp. 189–284.
Christiansen, M., Fagerholt, K., Nygreen, B. and Ronen, D. (2013) Ship routing and scheduling in thenew millennium. European Journal of Operational Research 228(3): 467–483.
Christiansen, M., Fagerholt, K. and Ronen, D. (2004) Ship routing and scheduling: Status andperspective. Transportation Science 38(1): 1–18.
Clarke, G. andWright, J.W. (1964) Scheduling of vehicles from a central depot to a number of deliverypoints. Operations Research 12(4): 568–581.
Ducruet, C. and Notteboom, T. (2012) Developing liner service networks in container shipping. In:D.W. Song and P. Panayides (eds.) Maritime Logistics: A Complete Guide to Effective Shipping andPort Management. London: Kogan Page, pp. 77–110.
Fagerholt, K. (1999) Optimal fleet design in a ship routing problem. International Transactions inOperations Research 6(5): 453–464.
Fagerholt, K. (2004) Designing optimal routes in a liner shipping problem. Maritime Policy &Management 31(4): 259–268.
Fagerholt, K., Christiansen, M., Hvattum, L.M., Johnsen, T.A.V. and Vabø, T.J. (2010) A decisionsupport methodology for strategic planning in maritime transportation. Omega 38(6):465–474.
Frankel, E.G. (2002) The challenge of container transshipment in the Caribbean. Paper presented atthe International Association of Maritime Economist Conference, Panama.
Hansen, P. and Mladenović, N. (2001) Variable neighborhood search: Principles and applications.European Journal of Operational Research 130(3): 449–467.
Hansen, P., Mladenović, N. and Pérez, J.M. (2010) Developments of variable neighborhood search.Annals of Operations Research 175(1): 367–407.
Hemmelmayr, V.C., Doerner, K.F. and Hartl, R.F. (2009) A variable neighborhood search heuristic forperiodic routing problems. European Journal of Operational Research 195(3): 791–802.
Hlača, B. and Babić, A.O. (2006) Feeder service – Logistic support to the port of Rijeka. Promet –Traffic and Transportation 18(3): 229–233.
Hoff, A., Andersson, H., Christiansen, M., Hasle, G. and Løkettangen, A. (2010) Industrial aspectsand literature survey: Fleet composition and routing. Computers & Operations Research 37(12):2041–2061.
Jadrijević, N. and Tomašević, M. (2011) Feeder service development in Mediterranean ports towardmore efficient transport technologies. Paper presented at the 14th International Conference onTransport Science – ICTS Portoroz, Slovenia.
Jin, Z.-H., You, Q.-H. and Wang, J. (2005) The container ship routing problem and its practicalsolutions. Paper presented at the International Conference on Management Science and Engineer-ing Incheon/Korea.
Karlaftis, M.G., Kepaptsoglou, K. and Sambracos, E. (2009) Containership routing with time deadlinesand simultaneous deliveries and pick-ups. Transportation Research Part E 45(1): 210–221.
Kjeldsen, K.H. (2011) Classification of ship routing and scheduling problems in liner shipping. INFOR:Information Systems and Operational Research 49(2): 139–152.
Kulak, O., Polat, O., Gujjula, R. and Günther, H.-O. (2013) Strategies for improving a long-establishedterminal’s performance: A simulation study of a Turkish container terminal. Flexible Services andManufacturing Journal 25(4): 503–527.
Kytöjokia, J., Nuortiob, T., Bräysya, O. and Gendreauc, M. (2007) An efficient variable neighborhoodsearch heuristic for very large scale vehicle routing problems. Computers & Operations Research34(9): 2743–2757.
Mladenović, N. and Hansen, P. (1997) Variable neighborhood search. Computers & OperationsResearch 24: 1097–1100.
Mourao, M.C., Pato, M.V. and Paixao, A.C. (2001) Ship assignment with hub and spoke constraints.Maritime Policy & Management 29(2): 135–150.
Notteboom, T.E. (2004) Container shipping and ports: An overview. Review of Network Economics3(2): 86–106.
Ocean Shipping Consultants. (2011) South Europe and Mediterranean Container Markets to 2025.Surrey, UK: Haskoning UK Ltd.
Parragh, S., Doerner, K. and Hartl, R. (2008) A survey on pickup and delivery problems. Journal fürBetriebswirtschaft 58(2): 21–51.
Polat, O., Kulak, O. and Günther, H.-O. (2012) An adaptive neighborhood search approach forVRPSPDTL. In: H.-O. Günther, K.-H. Kim and H. Kopfer (eds.) The 2nd International Conferenceon Logistics and Maritime Systems 22–24 August, Bremen, Germany, pp. 429–437.
Psaraftis, H. and Kontovas, C. (2013) Speed models for energy-efficient maritime transportation:A taxomomy and survey. Transportation Research Part C: Emerging Technologies 26: 331–351.
Ridolfi, G. (1999) Containerisation in the Mediterranean: Between global ocean routeways and feederservices. GeoJournal 48(1): 29–34.
Robinson, R. (1998) Asian hub/feeder nets: The dynamics of restructuring. Maritime Policy &Management 25(1): 21–40.
Ronen, D. (1983) Cargo ship routing and scheduling: Survey of models and problems. EuropeanJournal of Operational Research 12(2): 119–126.
Ronen, D. (1993) Ship scheduling: The last decade. European Journal of Operational Research 71(3):325–333.
Rudić, D. and Hlača, B. (2005) Feeder service to promote revitalization of the container transport inthe Adriatic ports. Naše More: Journal of Marine Science and Technology 52(1–2): 37–44.
Sambracos, E., Paravantis, J.A., Tarantilis, C.D. and Kiranoudis, C.T. (2004) Dispatching of smallcontainers via coastal freight liners: The case of the Aegean Sea. European Journal of OperationalResearch 152(2): 365–381.
Sigurt, M., Ulstein, N., Nygreen, B. and Ryan, D. (2005) Ship scheduling with recurring visits and visitseparation requirements. In: G. Desaulniers, J. Desrosiers and M. Solomon (eds.) ColumnGeneration. New York, USA: Springer, pp. 225–245.
Stenger, A., Vigo, D., Enz, S. and Schwind, M. (2012) An adaptive variable neighborhood searchalgorithm for a vehicle routing problem arising in small package shipping. Transportation Science47(1): 64–80.
Stopford, M. (2009) Maritime Economics, 3rd edn. New York: Routledge.Sun, X. and Li, N. (2006) A new immune genetic algorithm for the large scale ship routing problems.
Paper presented at the International Conference on Management of Logistics and Supply ChainChang Sha/Chine, Sydney, Australia.
UNCTAD. (2012) United Nations Conference on Trade and Development Review of MaritimeTransport. New York and Geneva: United Nations.
Varbanova, A. (2011a) Current issues in operational planning of general cargo transportation oncontainer feeder lines in the Black Sea region. The International Virtual Journal for Science,Techniques and Innovations for the Industry ‘Machines, Technologies, Materials’ 2011(3): 35–38.
Varbanova, A. (2011b) Theoretical analysis of cost efficiency of general cargo transportationon container feeder lines in the Black Sea region. The International Virtual Journal for
Windeck, V. (2013) A Liner Shipping Network Design: Routing and Scheduling Considering Environ-mental Influences. Wiesbaden, Germany: Produktion und Logistik. Springer Gabler.
Yang, Z., Chen, K. and Notteboom, T. (2012) Optimal design of container liner services: Interactionswith the transport demand in ports. Maritime Economics & Logistics 14(4): 409–434.
Appendix A
Table A1: Demand parameters for relevant container terminalsa
Terminalnumber
Terminal Portnumberb
Region Supplyc Demandd Handlingb
A Port Said A Black Sea — — 50B Candarli (Izmir) B Black Sea — — 501 Burgas 1 Black Sea 8 9 16.72 Varna 2 Black Sea 30 35 253 Constanta 1 3 Black Sea 34 82 504 Constanta 2 3 Black Sea 100 105 16.75 Illiychevsk 4 Black Sea 63 90 33.36 Odessa 5 Black Sea 120 160 16.77 Novorossiysk 1 6 Black Sea 166 110 258 Novorossiysk 2 6 Black Sea 108 65 12.59 Poti 7 Black Sea 26 112 11.110 Batumi 8 Black Sea 8 8 2511 Trabzon 9 Black Sea 14 29 11.112 Haydarpasa (Istanbul) 10 Sea of Marmara 26 58 33.313 Ambarli 1 (Istanbul) 11 Sea of Marmara 179 234 33.314 Ambarli 2 (Istanbul) 11 Sea of Marmara 120 166 33.315 Ambarli 3 (Istanbul) 11 Sea of Marmara 67 105 2516 Gebze 1 (Izmit) 12 Sea of Marmara 36 63 33.317 Gebze 2 (Izmit) 12 Sea of Marmara 35 77 2518 Gemlik 1 (Bursa) 13 Sea of Marmara 31 57 33.319 Gemlik 2 (Bursa) 13 Sea of Marmara 58 65 2520 Gemlik 3 (Bursa) 13 Sea of Marmara 17 22 33.321 Aliaga 1 (Izmir) 14 Aegean Sea 34 59 5022 Aliaga 2 (Izmir) 14 Aegean Sea 20 33 33.323 İzmir 15 Aegean Sea 86 165 2524 Thessaloniki 16 Aegean Sea 37 56 33.325 Piraeus 1 17 Aegean Sea 51 111 33.326 Piraeus 2 17 Aegean Sea 102 189 2527 Antalya 18 East Mediterranean sea 22 50 5028 Mersin 19 East Mediterranean sea 90 187 2529 Limassol 20 East Mediterranean sea 13 78 5030 Lattakia 21 East Mediterranean sea 40 75 33.331 Beirut 22 East Mediterranean sea 56 85 2532 Haifa 23 East Mediterranean sea 107 142 2533 Ashdod 24 East Mediterranean sea 123 125 25
34 Alexandria 1 25 East Mediterranean sea 38 130 33.335 Alexandria 2 25 East Mediterranean sea 26 52 33.336 Damietta 26 East Mediterranean sea 60 132 25
aDistances between ports calculated by use of the Netpas Distance software.bPort code of terminal as shown in Figure 5.cTerminal’s daily container supply in TEU.dTerminal’s daily container demand in TEU.eTerminal container handling efficiency in TEU per hour.
