ww.sciencedirect.com

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 0

Available online at w

journal homepage: www.elsevier .com/locate/watres

Multi-objective evolutionary optimization forgreywater reuse in municipal sewer systems

Roni Penn, Eran Friedler, Avi Ostfeld*

Faculty of Civil and Environmental Engineering, Technion e IIT, Haifa 32000, Israel

a r t i c l e i n f o

Article history:

Received 1 March 2013

Received in revised form

11 June 2013

Accepted 10 July 2013

Available online 20 July 2013

Keywords:

Greywater

Sewer system

Multi-objective

Management

Operation

NSGA-II

* Corresponding author. Tel.: þ972 4 8292782E-mail address: ostfeld@tx.technion.ac.il

0043-1354/$ e see front matter ª 2013 Elsevhttp://dx.doi.org/10.1016/j.watres.2013.07.012

a b s t r a c t

Sustainable design and implementation of greywater reuse (GWR) has to achieve an op-

timum compromise between costs and potable water demand reduction. Studies show that

GWR is an efficient tool for reducing potable water demand. This study presents a multi-

objective optimization model for estimating the optimal distribution of different types of

GWR homes in an existing municipal sewer system. Six types of GWR homes were

examined. The model constrains the momentary wastewater (WW) velocity in the sewer

pipes (which is responsible for solids movement). The objective functions in the optimi-

zation model are the total WW flow at the outlet of the neighborhoods sewer system and

the cost of the on-site GWR treatment system. The optimization routing was achieved by

an evolutionary multi-objective optimization coupled with hydrodynamic simulations of a

representative sewer system of a neighborhood located at the coast of Israel. The two non-

dominated best solutions selected were the ones having either the smallest WW flow

discharged at the outlet of the neighborhood sewer system or the lowest daily cost. In both

solutions most of the GWR types chosen were the types resulting with the smallest water

usage. This lead to only a small difference between the two best solutions, regarding the

diurnal patterns of the WW flows at the outlet of the neighborhood sewer system. How-

ever, in the upstream link a substantial difference was depicted between the diurnal

patterns. This difference occurred since to the upstream links only few homes, imple-

menting the same type of GWR, discharge their WW, and in each solution a different type

of GWR was implemented in these upstream homes. To the best of our knowledge this is

the first multi-objective optimization model aimed at quantitatively trading off the cost of

local/onsite GW spatially distributed reuse treatments, and the total amount of WW flow

discharged into the municipal sewer system under unsteady flow conditions.

ª 2013 Elsevier Ltd. All rights reserved.

1. Introduction services, recreation and sporting activities, and firefighting.

Inmany countries, the urban sector is the largest consumer of

potable water. In Israel it consumes some 700e800 � 106 m3/

year (for agriculture irrigationmostly treatedWWeffluents are

used). Domestic/residential consumers consume about 70% of

the municipal water demand, while the rest is consumed for

uses such as: tourism, offices, education, commerce, health

; fax: þ972 4 8228898.(A. Ostfeld).ier Ltd. All rights reserved

Therefore, reducing domestic water demand by on-site water

reuse has the potential to play a significant role in alleviating

the stress from existing water sources, in reducing the urgent

need for exploring new (and usually costly) water resources,

and to increase the sustainability of urban water usage.

The use of evolutionary optimization techniques for multi

e objective optimization is not new in the field of urban

.

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 05912

drainage/sewer systems. In a comprehensive review of state

of the art for genetic algorithms (GA) methods and their

application in the field of water resources planning and

management (including sewer systems) carried out by

Nicklow et al. (2010) it was shown that evolutionary compu-

tation can be a flexible and powerful tool when used appro-

priately in this field. They further stated that it will continue to

evolve in the future due to several challenges. The efficiency

in using GA for multi-objective optimization of integrated

sewer systems (integrating some of the following: the sewer

system, the wastewater treatment plant (WWTP), the

receiving water bodies, pollution e load and water quality

model) was shown by Boomgaard et al. (2004), Fu et al., 2008,

Muschalla 2008, Rathnayake and Tanyimboh 2011, Rauch

and Harremoes 1999 and others. The optimization of sewer

networks in terms of speed and reliability was found to be

more efficient by using an adaptive GA, in which the constrain

handling method was adapted (Haghighi and Bakhshipour,

2012). Pan and Kao 2009 optimized sewer system design by

developing a GA based approach combined with a nonlinear

cost optimization model that was approximated and trans-

formed into a quadratic programing. Atef et al., 2012 pre-

sented an algorithm to allocate budgetary resources for

condition assessment of water and sewer networks.Ward and

Savic 2012 applied a multi-objective GA optimization model,

coupled with an enhanced critical risk of failure methodology

for sewer rehabilitation. 1D2D coupled model (1D subsurface

and 2D surface flow models) was linked with NSGA-II (an

evolutionary algorithm) for multi-objective optimization of

cost-benefit of urban floodmanagement (Delelegn at al., 2011).

The computational efficiency of this method was proved to be

acceptable for optimization. As shown above, optimization of

urban sewer systems by GA multi-objective optimization, is

broadly found in the literature. However modeling greywater

reuse (GWR) together with evolutionary optimization as a tool

for evaluating the optimum distribution of GWR homes, has

not been found in the literature.

Greywater (GW) is generally defined as domestic sewage

excluding the wastewater (WW) stream generated by toilets.

Kitchen (kitchen sink and dishwasher) wastewater is defined

as dark GW; sometimes washing machine WW is included in

this definition too. WW streams generated by the bath,

shower and washbasin are defined as light GW, and WW

generated from toilets (WC) is defined as blackwater.

To prevent hygienic and health risks, and to minimize

negative aesthetic effects, treatment of GW is necessary prior

to reuse (Diaper et al., 2001; Dixon et al., 1999). Friedler (2004)

has shown that, as the demand for GW within the urban

environment (i.e., for toilet flushing and garden irrigation) is

significantly lower than its production, it is not necessary to

recycle all GW streams, but rather to focus on the less polluted

light GW, and to discharge the more polluted dark GW

together with the blackwater stream to the urban sewer

system.

Sustainable design and implementation of GW reuse

(GWR) has to achieve an optimum compromise between costs

and potable water demand reduction. Studies show that GW

reuse (GWR) is an efficient tool for reducing potable water

demand. Using onsite light GWR for toilet flushing can reduce

daily household water consumption by 26%. Using the excess

amount of the light GW for garden irrigation can further

reduce the daily water demand to an overall reduction of 41%

(Penn et al., 2012). Further, integrating residential wells,

rainwater tanks and GW systems can result in significant

water savings at the household scale (Hunt et al., 2011; Rozos

et al., 2010; Rozos and Makropoulos, 2012). However, rainfall

harvesting depends on stochastic phenomena (i.e., climatic

conditions, including rainfall and temperature) whose varia-

tion introduces long-term uncertainties in the systems’ per-

formance (Rozos et al., 2010). In a research carried out by

Rozos and Makropoulos (2012) the reliability of water-aware

technologies (e.g., rainwater harvesting schemes and sus-

tainable drainage systems) is proven to decrease with urban

density. In this study the water saving tools focused on are

low-flush toilets and different types of GWR. Friedler and

Hadari (2006) demonstrated that under certain circum-

stances onsite GWR for toilet flushing can be economically

worthy even to the consumer itself. It depends on the treat-

ment technology chosen, on the size of the served population

and on the price of water. With GWR it might be possible to

postpone the enlargement of existing sewer systems, to

construct new sewers with smaller pipe diameter, and lower

energy consumption for sewage pumping (Friedler and

Hadari, 2006; Penn et al., 2013).

Several studies dealing with multi-objective optimization

between potable water demand reduction and costs can be

found in the literature, however the water saving schemes

usually include rainwater harvesting combined with GWR,

whereas in Israel rainwater harvesting is not a feasible solu-

tion. Some of these studies are further briefly discussed.

Oldford and Filion (2012) show, by multi-objective optimiza-

tion, a trade-off between decreasing water demand and

upgrading of operational cost. The decision variables in their

model were the diameters of new water mains, the price of

water, and the decision to offer rebates for various low-flow

fixtures or appliances. This was demonstrated on a five-

node network. Brock et al. (2010) used multiple objective

optimization, by a genetic algorithm, to identify which of the

following options, or combination of them, is the optimal so-

lution for water saving: rainwater, stormwater and GWR sys-

tems. This was done by determining evaluation criteria that

reflect the cost and environmental impacts and hence sus-

tainability of these systems. Rozos et al. (2010) assessed sus-

tainable design and implementation of two water saving

schemes by multi-objective optimization between costs

(including energy) and benefits (potable water demand

reduction). The first scheme was rainwater harvesting. In this

scheme harvested rainwater, stored in a local tank, was used

for toilet flushing, washing mashing and outside uses. The

reset of the appliances were supplied with potable water from

water mains. The second scheme was a combination of rain-

water harvesting and local GWR. In this scheme GW from the

shower bath and wash basin was treated locally and stored

together with the harvested rainwater. The influence of po-

tential changes in climatic conditions (oceanic, Mediterra-

nean, and desert) to the scheme’s efficiency was also taken

into account. Their results indicate that rainwater harvesting

alone can achieve significant reduction in domestic potable

water consumption. However, in this scheme the systems’

performance can suffer from long term uncertainties since

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 0 5913

the water source depends on climate conditions. They further

show that schemes that are efficient in their use of local GW

are less susceptible to changes in climatic conditions, due to

the invariant nature of GW as a water source. Another

mathematical tool for assessing optimum strategies for the

realization of sustainable urban water management (storm-

water drainage and domestic WW reuse) was developed by

Kaufmann et al. (2007).

As a result of reduction of flows released to sewers,

blockage rates could increase (Parkinson et al., 2005;

Drinkwater et al., 2008; Marleni et al., 2012). Lower flow ve-

locities may result in increased sedimentation, which could

then be followed by the formation of H2S and release of mal-

odors. This phenomenon represents a substantial problem,

particularly in combined sewers which convey water from

two sources, namely WW and storm water. The latter is un-

affected by GWR, and hence is expected to cause lower effects.

Moreover, Bertrand et al. (2008) report positive effects of GWR

on combined sewers by reducing floods and overflows. How-

ever, since the pipes of a combined sewer system are larger

than the pipes of a separate system, when flows are reduced

the risk of sedimentation increases, leading to deposits in the

sewer. This problem can increase in regions with steep pop-

ulation decrease (e.g. former east Germany, see for example:

Barjenbruch, 2003; Schlor et al., 2009; Nowack et al., 2010).

However, it has been suggested that this problem should not

be substantial in many (or the majority of) existing municipal

separate sewers (as in the case study) of mega-cities, since

they are maintained close to or even above their design ca-

pacity, and since the specific water demand (and consequent

WW generation) in many regions increases (e.g. Freni et al.,

2012). Further, in a research conducted by Penn et al. (2013)

it was shown that, in a specific case study of a separate

sewer system, as a result of GWR no excess blockages are

expected to occur.

There is an infinite number of ways to spread different

GWR types along neighborhood homes considering the loca-

tion of the homes reusing GW and the GWR type (these types

will be further elaborated below). Apart from the desire to

achieve maximum water saving (that can be achieved from a

maximumnumber of GWRhomes), there is a significant effect

to the location of the GWR homes on the function of the sewer

system. In upstream links of the sewer system WW flows are

intermittent and lower than in downstream links, where

additional houses discharge their WW, and higher and

steadier WW flows can be found. Hence, a GWR home located

upstream of the sewer system will have a different effect on

the WW flows, and hence on solids movement in the sewer

system, than a GWR home located downstream. This study

presents a multi-objective optimization model for estimating

the optimum distribution of different types of GWR homes

and toilet flush volumes in an existing municipal sewer sys-

tem. The model takes into consideration the momentary WW

velocity in the sewer pipes (which is responsible for solids

movement), the total WW flow at the outlet of the neighbor-

hoods sewer system, and the cost of the on-site GWR treat-

ment system.

Municipal sewers function under unsteady, non-uniform

flow conditions, where the sub-daily instantaneous flows

responsible for the transport (or sedimentation) of solids.

Hence, the model was developed on a sub-daily scale. This

was achieved by multi-objective optimization and dynamic

simulation of the sewer system as further described below.

To the best of our knowledge this is the firstmulti-objective

optimization model aimed at quantitatively trading off the

cost of local/onsite GW spatially distributed reuse treatments,

and the total amount of WW flow discharged into the

municipal sewer system under unsteady flow conditions.

2. Methedology

This work presents an optimization model for estimating the

optimum distribution of different GWR types in houses in an

existing municipal sewer system. It combines the use of a

sewer simulation model and a multi-objective optimization

algorithm.

The simulation model used was SIMBA6 (ifak, 2009).

SIMBA6 is an integrated simulation software for sewer sys-

tems, wastewater treatment plants (WWTPs) and river water

quality. The hydrodynamic modeling modules of SIMBA6 are

based on an extended version of the US-EPA SWMM

(Rossman, 2004) model, which incorporates the full dynamic

solution of the Saint Venant differential equations.

The multi-objective optimization algorithm utilized was

based on NSGA-II which is amulti-objective genetic algorithm

developed by Deb et al. (2002).

2.1. Basic work structure

The multi-objective optimization algorithm, NSGA-II (Deb

et al., 2002), was used for finding the non-dominated solu-

tions for the optimum type of GWR (will be described below)

discharged from different locations, in the sewer system. The

decision variables incorporated a vector of length 34, repre-

senting the number of nodes in the sewer system into which

WW is discharged (further description of these nodes will be

given below, 3.1). Each location in the vector contained the

type of GW discharged to the corresponding node. Two

objective functions were formulated. The first was for mini-

mum WW flow at the outlet of the sewer system (i.e., mini-

mizing water usage, thus maximizing GWR). The second

objective function was for minimum cost, which included

operation and maintenance costs of the GWR treatment unit.

The cost for each GWR type, $/m3, and the quantity treated,

are given in Table 1. Unit costs were taken from Friedler (2008).

Although the minimum momentary velocity required for the

movement of solids is 0.6 m/s (Walski et al., 2009), for more

flexibility in the optimization, the model constrain was set to

0.45 m/s in most pipes. It should be noted that in reality when

GWR is implemented, in some pipes the maximum velocity,

throughout the day, might be lower than the minimum

required for solid movements. If such situations occur,

changes in design should be made, and pipe flushing should

be considered.

The SIMBA simulation model was used for obtaining the

total diurnal WW flow at the outlet of the sewer network and

for calculating the momentary WW flow and velocity

throughout the day in each pipe segment in the sewer system.

Table 1 e Types of greywater reuse (GWR).

GWR type Toilet flush volumes (L) GWR unit cost ($/m3) Quantity treated (L/PE/dd)

Full flush Half flush

1 No GWRa 9 6 NAe NA

2 GWR WCb 9 6 1.1 37.7

3 GWR WC þ IRRc 9 6 0.8 57.3

4 No GWRa 6 3 NA NA

5 GWR WCb 6 3 1.5 21

6 GWR WC þ IRRc 6 3 0.8 57.3

a No GWR e no greywater (GW) reuse. Wastewater from all in house sources runs to the sewer.

b GWR WC (toilet) e light GW is treated and used for toilet flushing, excess flows discharged (without treatment) to the sewer system as

overflow.

c GWR WC þ IRR (irrigation) e same as#2 but excess light GW (overflow) is reused for garden irrigation after treatment.

d L/PE/d e liter/person/day.

e NA ¼ not applicable.

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 05914

2.1.1. Multi-objective optimization through NSGA-IIA flowchart of the solutionmethodology is presented in Fig. 1.

Initially 10 different distributions of GWR types were

randomly selected (i.e., initial solution population). The pop-

ulation was then multiplied through crossover combinations.

The crossover combinations were performed by first choosing

randomly which two distributions to combine, further by a

uniform distribution (probability of 1/34) it was decided where

the crossover connection will be made. For each distribution

(from the 20), in each place in the vector, a decisionwhether to

make a mutation or not was determined according to Ber-

noulli distribution (probability of 1/34 for implementing a

mutation). If the result was to have a mutation (i.e., to change

the type of GWR in that place), then the decision to which type

to modify was made according to a uniform distribution

(probability of 1/5).

Fig. 1 e Methodology flowchart.

At this step, for each of the 20 distributions, the simulation

in SIMBAwas performed for three days. The results of the last

day for each distribution were saved and, taking into account

the constrains, the objection functions were calculated. If the

maximum momentary diurnal velocity did not exceed 0.45 a

fine was added to the objective cost function. The fine was

high enough for the solution not to be considered as an opti-

mum solution and not to be chosen for the next generation (as

will further be described). The fine was arbitrary chosen to be

the velocity multiplied by ten to the power of ten.

Ranking and selecting the best 10 solutions was preformed

according to the Elitist non-dominated sorting genetic algo-

rithm (Deb et al., 2002). Clearly these solutions included the

non-dominated results. From this stage the algorithm restar-

ted, for a total of thirty generations. Each generation took

around 80 min on a PC AMD Athlon (tm) 64 � 2 Dual, Core

Processor 5600 þ 2.91 GHz, 3e2.5 GB of RAM, thus a single

model run lasted w1.7 days. At the end of a run the non-

dominated solutions were saved. This procedure was

repeated for 23 trials. The results that will later be presented

are of the non-dominated solutions from all the 23 trials.

2.1.2. Sewer system simulation modelAsmentioned above, the SIMBA simulation system (ifak, 2009)

was used for this work. In this study, inflows to the sewer

network are calculated separately and fed into the SWMM

block within the SIMBA simulator as inputs. The GWR type

discharged to each node was obtained by the multi-objective

simulations, as described above. The numbers of residents

discharging their WW into each node were entered into the

INFLOW block used in this SIMBA model. Then the appro-

priate diurnal WW discharge corresponding to a one person

discharge pattern were multiplied by the number of residents

corresponding to that node. These inflows were fed into the

SIMBA-SWMM block, which represents the complete sewer

network and gives detailed hydrodynamicmodeling, based on

the full solution of the Saint Venant equations. It incorporates

an extension (ifak 2009, Schutze, 2008) of the SWMM program

of the US EPA (Rossman, 2004). The output of the model (i.e.,

momentary diurnal velocities in the sewer pipes and the total

WW flow at the outlet of the sewer system), was returned to

themulti- objective algorithm as described above and in Fig. 1.

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 0 5915

Full metadata (see attached Program_metadata.doc) and

all SIMBA and MATLAB codes used in this study are attached

as Supplementary Material.

3. Case study

3.1. The sewer system

The methodology is demonstrated on a representative sewer

system of a densely populated neighborhood, located in cen-

tral Israel, near the coast (flat terrain). In the sewer system

(Fig. 2) there are 154 nodes and 153 links [additional full in-

formation on the system geometry and layout is available (i.e.,

pipe material, diameter, longitudinal slope, location, types of

manholes, etc.) at the supplementary material file: full_-

format_modified.inp, see Program_metadata.doc file].

To simplify the modeling process, it was assumed that

buildings are connected to every second node, and adjacent

buildings were aggregated into 34 clusters, each discharging

wastewater into a single node. The nodes, into which a cluster

of buildings discharged their wastewater, were chosen arbi-

trarily (Fig. 2, blue circles). The blue numbers in Fig. 2 are the

number of buildings in the corresponding cluster. The total

length of the system is about 6 km. For simulating effects on

downstream segments of the sewer system [i.e., segments

adjacent to the municipal waste water treatment plant

(WWTP)], a long downstream segment (5 km) was artificially

added to the existing system (links 85 and 154). The number of

Fig. 2 e System layout (follo

residents in theneighborhoodwas estimated at 15,000, living in

68 buildings (each 20e22 stories high, accommodating 220 res-

idents).Theneighborhood isservedbyaseparate sewersystem.

3.2. Types of GWR examined

Six types of GWR were examined (Table 1). These were based

on three types of houses: I) A non-recycling house (No GWR);

II) A house reusing light GW for toilet flushing (GWR WC); and

III) A house reusing light GW for toilet flushing and garden

irrigation (GWRWCþ IIR). Based on these types of houses, two

types of toilet flush volumes were considered: I) where toilet

flush volumes were 9 and 6 L for full and half flushes,

respectively (i.e., the existing situation in Israel), referring to

GWR types 1, 2 and 3 (Table 1). II) Where toilet flush volumes

were reduced to 6 and 3 L for full and half flushes, respectively,

referring to GWR types 4, 5 and 6 (Table 1). It is further

assumed that at each cluster of houses, all residents dis-

charged the same type of GWR.

3.3. Domestic wastewater discharge

The domestic WW discharge input to the model was the cor-

responding diurnal hydrograph of wastewater discharged to

the sewer by individual persons during weekdays (L/PE/day)

(i.e., type 1, 2, 3, 4, 5 or 6 GWR, see Table 1 and Fig. 3). The

diurnal flow patterns refer to WW discharges, during week-

days (weekends were not modeled), from WW generating

household appliances (i.e., kitchen sink, wash basin, bath,

wing Penn et al., 2013).

Fig. 3 e Diurnal hydrographs of the wastewater (WW) discharged to the sewer by one. person for the six greywater reuse

(GWR) scenarios (see also Table 1). L/PE/10min [ liter per one person per 10 min; No GWR e no greywater (GW) reuse.

Wastewater from all in house sources runs to the sewer; GWR WC e light GW is treated and used for toilet flushing, excess

flows discharged (without treatment) to the sewer system as overflow; GWR WC IRR e same as GWR WC but excess light

GW (overflow) is reused for garden irrigation after treatment; 9/6 stands for toilet flush volumes of 9 and 6 L for full and half

flush respectively; 6/3 stands for toilet flush volumes of 6 and 3 L for full and half flush, respectively.

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 05916

shower and washing machines) which were derived from a

10 min interval dataset obtained from previous work per-

formed in single houses in England (Butler et al., 1995). Data on

domestic toilet usage was taken from Friedler et al. (1996a,

1996b). Derivation of these hydrographs is further described

in Penn et al. (2012) and Penn et al. (2013).

Fig. 4 e Model Pareto front incorporating all 23 non-dominated

solutions are surrounded by rectangles (i.e., solutions #1 and #

wastewater flow at the outlet of the sewer system (see Fig. 2); 7

4. Results and discussion

Fig. 4 presents the daily operation andmaintenance costs, and

the total diurnal flow discharged to the sewer system of all the

non-dominated solutions from the 23 full NSGA-II trails. The

model trial repetitions [the non-dominated best two

2)]. #1 e solution number 1; 1116 (m3/day) e total daily

26 ($/day) e total daily cost of solution #1.

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 0 5917

resulted Pareto front incorporates four non-dominated solu-

tions. The best two solutions are surrounded by rectangles,

and defined as solutions 1 (#1) and 2 (#2), respectively. Those

were selected as best solutions, as they have either the

smallest WW flow discharged at the outlet of the neighbor-

hood sewer system (#1), or the lowest daily cost (#2). The total

WW flow at the outlet of the neighborhood sewer system was

1116 and 1139 m3/day for solutions #1 and #2, respectively.

The daily costs were 726 and 696 $/day for solutions #1 and #2,

respectively. As one can see there is less difference between

the solutions (from the 23 rounds) regarding the total daily

WW flow discharged to the sewer compared with the daily

costs. The largest difference between the maximum and

minimum daily WW flow was found to be 29% from the

minimum, compared with the difference in daily costs, which

was found to be of 12% from the minimum.

In Fig. 5 the sewer layout with the type of GWR discharged

from the different clusters of buildings for solutions 1 and 2,

respectively, is described. In most of the nodes the types of

GWR chosen were the types resulting with the smallest water

usage (e.g., type 6, GWR for toilet flushing and garden irri-

gation, with toilet flush volumes of 6 and 3 L for full and half

flush, respectively, appears 20 times in both solutions, Table

2) and type 3, same as type 6 only with higher toilet flush

volumes of 9 and 6 L for full and half flush, respectively (6

times in solution 1 and 10 times in solution 2, Table 2). This

Fig. 5 e Detailed results for optimal

indicates that if there were no velocity constrains at all, it

might have been that the best solutions would include more

clusters with the types of GWR resulting in the maximum

water saving (i.e., types 6 and 3) a situation like this can occur

for example in a sewer system with steeper slopes or smaller

pipe diameters. A situation with higher velocities can also be

found in the current sewer system if additional houses will

be connected to it, and hence its enlargement will be post-

poned. As more GW is being reused (and treated) the opera-

tion and maintenance cost of the GW treatment unit increase

(i.e., as more GW is being treated and reused, treatment cost

for one cubic meter decreases but the overall cost increases,

Table 1).

Hardly any differences were depicted between the diurnal

momentary WW flow at the outlet of the neighborhood sewer

system of the two best solutions (Fig. 6B). This was due to the

fact that there was not much difference between the total

daily WW flows of the two best solutions. However, in the

upstream link (i.e., link 36) since in each solution a different

type of GWR was implemented (i.e., #1 e type 4, and #2 e type

6) a substantial difference can be seen between the diurnal

patterns of the WW flows of both solutions (Fig. 6A).

As expected, since in solution 2more GW is being reused in

the cluster of buildings discharging their WW into link 36 (in

solution 1 there is no GWR), the momentary WW flows are

lower than those of solution 1.

solutions #1 and #2 (see Fig. 4).

Fig. 6 e Diurnal wastewater flow in upstream and downstream links for solutions #1 and #2, respectively (see Figs. 2 and 4).

Table 2 e Distribution type of cluster building discharging GWR in solutions 1 and 2 (see Figs. 5 and 6).

Solution Type of GWR

1 2 3 4 5 6

No GWRa9/6

d GWR WCb9/6 GWR WC IRRc

9/6 No GWR6/3e GWR WC6/3 GWR WC IRR6/3

1 0 2 6 5 1 20

2 1 0 10 3 0 20

a No GWR e no greywater (GW) reuse. Wastewater from all in house sources runs to the sewer.

b GWR WC e light GW is treated and used for toilet flushing, excess flows discharged (without treatment) to the sewer system as overflow.

c GWR WC IRR e same as#2 but excess light GW (overflow) is reused for garden irrigation after treatment.

d In types 1, 2 and 3 toilet flush volumes are 9 and 6 L for full and half flush, respectively.

e In types 4, 5 and 6 toilet flush volumes are 6 and 3 L for full and half flush, respectively.

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 05918

5. Conclusions

In this study an optimization model for estimating the opti-

mum distribution of different GWR types in clusters of

buildings in an existing municipal sewer system, was devel-

oped and demonstrated. The model combines the usage of a

sewer simulation model with a multi- objective evolutionary

optimization algorithm.

A higher difference between the non-dominated solutions

was observed regarding the daily cost comparedwith the total

diurnal wastewater discharged at the outlet of the sewer sys-

tem. This was emphasized with the two best solutions (non-

dominated solutions having either the smallest WW flow dis-

charged at the outlet of the sewer system or the lowest daily

costs),wherehardly anydifferencewasobserved regarding the

diurnal pattern of the WW flow at the outlet of the sewer sys-

tem.However, in theupstreamlinks, sincedifferentGWRtypes

were implemented, a higher difference was observed.

From the results presented in this study it can be

concluded that in a sewer system having higher velocities (as

a result of e.g. higher slopes, smaller diameters or additional

houses connected to the same sewer system) the optimum

compromise between on-site GWR treatment system cost of

maintenance and operation, and potable water demand

reduction, could draw to most houses implementing GWR

types resulting in maximum water savings.

Research extensions of this work should take into consid-

eration other constraints such as water usage costs, costs of

changes in the design of the sewer network (e.g., pipe slopes,

diameters, etc.), and different parameters for judging sewer

blockages occurrences (e.g. critical water depth or shear

stress, Butler et al., 2003; Arthur et al., 2008; Marleni et al.,

2012). Further, stochastic characteristics of individual per-

sons WW discharges can provide a more reliable input to the

simulation model. Moreover, water quality, which was not

referred to in this study, is an important factor that should be

taken into consideration. It can be incorporated via the

objective functions (e.g., as minimum pollutants concentra-

tions at the outlet of the sewer system entering the municipal

wastewater treatment plant). Or it could be embedded as a

constraint (e.g., minimumandmaximumpollutant loads and/

or concentrations).

Another aspect that should be considered is the reduction

in simulation time. In this study 23 trials were conducted,

each lasting w1.7 days. Reduction in simulation time can

enable more simulation runs and hence achievement of more

results, which could potentially produce a larger Pareto-front.

Reduction of the computational time is especially important if

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 0 5919

the model is intended to be extended for possible real-time

operation of greywater sewer systems.

Acknowledgments

This researchwas partially supported by a grant fromMinistry

of Science & Technology of the State of Israel and FZK FOR-

SCHUNGSZENTRUM KARLSRUHE and by Israel Water

Authority.

The authors wish to thank Dr. Manfred Schutze from ifak,

Institut fur Automation und Kommunikation, Magdeburg,

Germany, for his generous help regarding the problems

encountered in the hydrodynamic simulations in SIMBA.

Appendix A. Supplementary data

Supplementary data related to this article can be found at

http://dx.doi.org/10.1016/j.watres.2013.07.012.

r e f e r e n c e s

Arthur, S., Crow, H., Pedezert, L., 2008. Understanding blockageformation in combined sewer networks. Proceedings of theInstitution of Civil Engineers. Water Management 161 (WM4),215e221.

Atef, A., Osman, H., Moselhi, O., 2012. Multi-objective geneticalgorithm to allocate budgetary resources for conditionassessment of water and sewer networks. Canadian Journal ofCivil Engineering 39 (9), 978e992.

Barjenbruch, M., 2003. Prevention of odour emergence in sewagenetworks. Water Science and Technology 47 (7e8), 357e363.

Bertrand, N., Jefferson, B., Jeffrey, P., 2008. Cross sectoral andscale-up impacts of greywater recycling technologies oncatchment hydrological flows. Water Science and Technology57 (5), 741e746.

Boomgaard, M.E., Langeveld, J.G., Clemens, F.H.L.R., 2004.Wastewater system optimisation using genetic algorithms. In:Paper Presented at the Bridging the Gap: Meeting the World’sWater and Environmental Resources Challenges. Proceedingsof the World Water and Environmental Resources Congress,vol. 2001.

Butler, D., Friedler, E., Gatt, K., 1995. Characterising the quantity &quality of domestic wastewater inflows. Water Science andTechnology 31 (7), 13e24.

Butler, D., Davies, J.W., Jefferies, C., Schutze, M., 2003. Gross solidtransport in sewers. Proceedings of the Institution of CivilEngineers: Water and Maritime Engineering 156 (2), 175e183.

Brock, C., Colsey, A., McGregor, E., Russo, S., Dandy, G., Maier, H.,2010. Multi-objective optimization of integrated water reusesystems at a cluster scale. In: Water Distribution SystemAnalysis e WDSA2010, Tucson, AZ, USA, Sept. 12e15.

Deb, K., Pratap, A., Agarwal, S., Meyarivan, T., 2002. A fast andelitist multiobjective genetic Algorithm: NSGA e II. IEEETransactions on Evolutionary Computation 6 (2), 182e197.

Delelegn, S.W., Pathirana, A., Gersonius, B., Adeogun, A.G.,Vairavamoorthy, K., 2011. Multi-objective optimisation ofcost-benefit of urban flood management using a 1D2D coupledmodel. Water Science and Technology 63 (5), 1053e1059.

Diaper, C., Dixon, A., Butler, D., Fewkes, A., Parsons, S.A.,Stephenson, T., Strathern, M., Strutt, J., 2001. Small scale

water recycling systems e risk assessment and modelling.Water Science and Technology 43 (10), 83e90.

Dixon, A.M., Butler, D., Fewkes, A., 1999. Guidelines for greywaterre-use: health issues. Water and Environment Journal 13 (5),322e326.

Drinkwater, A., Chambers, B., Waylen, C., 2008. Less Water toWaste: Impact of Reductions in Water Demand onWastewater Collection and Treatment Systems, p. 85. WRcReport to UK Environment Agency.

Freni, G., Mannina, G., Viviani, G., 2012. The role of modelinguncertainty in the estimation of climate and socioeconomicimpact on river water quality. Journal of Water ResourcesPlanning and Management 138 (5), 479e490.

Friedler, E., 2004. Quality of individual domestic greywaterstreams and its implication on on-site treatment and reusepossibilities. Environmental Technology 25 (9), 997e1008.

Friedler, E., 2008. The water saving potential and the socio-economic feasibility of greywater reuse within the urbansector e Israel as a case study. International Journal ofEnvironmental Studies 65 (1), 57e69.

Friedler, E., Brown, D.M., Butler, D., 1996a. A study of WC derivedsewer solids. Water Science and Technology 33 (9), 17e24.

Friedler, E., Butler, D., Brown, D.M., 1996b. Domestic WC usagepatterns. Building and Environment 31 (4), 385e392.

Friedler,E.,Hadari,M.,2006.Economic feasibilityofon-sitegreywaterreuse in multi storey buildings. Desalination 190, 221e234.

Fu, G., Butler, D., Khu, S., 2008. Multiple objective optimal controlof integrated urban wastewater systems. EnvironmentalModelling and Software 23 (2), 225e234.

Haghighi, A., Bakhshipour, A.E., 2012. Optimization of sewernetworks using an adaptive genetic algorithm. WaterResources Management 26 (12), 3441e3456.

Hunt, J., Anda, M., Ho, G., 2011. Water balance modelling ofalternate water sources at the household scale. Water Scienceand Technology 63 (9), 1873e1879.

ifak, 2009. SIMBA6 e Simulation of Wastewater Systems,Reference and Tutorial. ifak e Institut fur Automation undKommunikationtechnik e. V. Magdeburg, Germany.

Kaufmann, I., Meyer, T., Klasch, M., Schmitt, T.G.,Hamacher, H.W., 2007. Implementation of sustainablesanitation in existing urban areas: long-term strategies for anoptimised solution. Water Science and Technology. 56 (5),115e124.

Muschalla, D., 2008. Optimization of integrated urban wastewatersystems using multi-objective evolution strategies. UrbanWater Journal 5 (1), 59e67.

Marleni, N., Gray, S., Sharma, A., Burn, S., Muttil, N., 2012. Impactof water source management practices in residential areas onsewer networks e a review. Water Science and Technology 65(4), 624e642.

Nicklow, J., Reed, P., Savic, D., Dessalegne, T., Harrell, L., Chan-Hilton, A., Karamouz, M., Minsker, B., Ostfeld, A., Singh, A.,Zechman, E., 2010. State of the art for genetic algorithms andbeyond in water resources planning andmanagement. Journalof Water Resources Planning and Management 136 (4),412e432.

Nowack, M., John, S., Tranckner, J., Gunther, E., 2010.Demographic change as driver of wastewater fees in urbandrainage systems e a comparison of demography, watersaving, maintenance cost, operating cost and industry effects.GWF. Wasser e Abwasser 151 (11), 1076e1085.

Oldford, A., Filion, Y., 2012. A new optimization framework thatincludes water conservation strategies to reduce demand inwater distribution networks. In: World Environmental andWater Resources Congress 2012: Crossing Boundaries May20e24, Albuquerque, New Mexico.

Pan, T., Kao, J., 2009. GA-QP model to optimize sewer systemdesign. Journal of Environmental Engineering 135 (1), 17e24.

wat e r r e s e a r c h 4 7 ( 2 0 1 3 ) 5 9 1 1e5 9 2 05920

Parkinson, J., Schutze, M., Butler, D., 2005. Modelling the impactsof domestic water conservation on the sustainability of theurban sewerage system. Journal of the Chartered Institute ofWater and Environment Management (CIWEM) 19 (1), 49e56.

Penn, R., Hadari, M., Friedler, E., 2012. Evaluation of the effects ofgreywater reuse on domestic wastewater quality andquantity. Urban Water Journal 9 (3), 137e148.

Penn, R., Schutze, M., Friedler, E., 2013. Modelling the effects ofon-site greywater reuse and water-efficient toilets onmunicipal sewer systems. Journal of EnvironmentalManagement 114, 72e83.

Rathnayake, U.S., Tanyimboh, T.T., 2011. Optimal control ofcombined sewer systems using SWMM 5.0. WIT Transactionson the Built Environment 122, 87e96.

Rauch, W., Harremoes, P., 1999. Genetic algorithms in real timecontrol applied to minimize transient pollution from urbanwastewater systems. Water Research 33 (5), 1265e1277.

Rossman, L.A., 2004. Stormwater Management Model. User’sManual. SWMM5.0. Environment Protection Agency.November 2004.

Rozos, E., Makropoulos, C., 2012. Assessing the combined benefitsof water recycling technologies by modelling the total urbanwater cycle. Urban Water Journal 9 (1), 1e10.

Rozos, E., Makropoulos, C., Butler, D., 2010. Design robustness oflocal water recycling schemes. Journal of Water ResourcesPlanning and Management 136 (5), 531e538.

Schutze, M., 2008. SIMBA e Ein Simulator furKanalnetzsteuerung. In: DWA-Seminar “Abflusssteuerung inKanalnetzen e Kostengunstige Reduzierung der Entlastungenaus Mischsystemen”. Heidelberg, 27.10.2008.

Schlor, H., Hake, J.F., Kuckshinrichs, W., 2009. Demographics as anew challenge for sustainable development in the Germanwastewater sector. International Journal of EnvironmentalTechnology and Management 10 (3e4), 327e352.

Walski, T., Edwards, B., Helfer, E., Whitman, B.E., 2009. Transportof large solids in sewer pipes. Water Environment Research 81(7), 709e714.

Ward, B., Savi�c, D.A., 2012. A multi-objective optimisation modelfor sewer rehabilitation considering critical risk of failure.Water Science and Technology 66 (11), 2410e2417.