Sunteți pe pagina 1din 20

Journal of Cleaner Production 230 (2019) 1376e1395

Contents lists available at ScienceDirect

Journal of Cleaner Production


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

Variable fleet size and mix VRP with fleet heterogeneity in Integrated
Solid Waste Management
Hossein Asefi a, Shahrooz Shahparvari b, *, Prem Chhetri b, Samsung Lim c
a
Ocean College, Zhejiang University, Zhoushan 316021, China
b
School of Business IT and Logistics, RMIT University, Melbourne, VIC, 3000, Australia
c
School of Civil and Environmental Engineering, The University of New South Wales, Sydney, NSW, 2052, Australia

a r t i c l e i n f o a b s t r a c t

Article history: The Integrated Solid Waste Management (ISWM) is a recent effective tool to manage with the growing
Received 24 August 2018 challenge of Municipal Solid Waste (MSW). The ISWM integrates all the system components (i.e. transfer,
Received in revised form treatment, recycling and disposal of wastes) to enhance the sustainable waste management whilst
17 April 2019
reducing operational costs. In this paper, we investigate an integrated framework of the Fleet Size and
Accepted 19 April 2019
Available online 15 May 2019
Mix Vehicle Routing Problem (VRP) to optimize the cost-effective ISWM system. A novel bi-objective
Mixed-Integer Linear Programming (MILP) model is developed to concurrently minimize the trans-
portation cost in the entire waste management system and total deviation from the fair load allocation to
Keywords:
Vehicle routing problem
transfer stations. A complete ISWM system with all interdependent facilities and multiple technologies,
Fleet size and mix VRP is developed to address a tri-echelon Fleet Size and Mix VRP with a heterogeneous fleet of vehicles under
Municipal solid waste multiple technologies and waste compatibility constraints. The model was solved for both the Pre-
Integrated solid waste management emptive and Non-Preemptive conditions using Lexicographic and Goal Programming optimization ap-
Lexicographic optimization proaches. The model was tested on a case of ISWM in the Southern part of Tehran, Iran.
Goal programming © 2019 Elsevier Ltd. All rights reserved.

1. Introduction by many local governments to effectively deal with the challenges


of municipal solid waste management. The argument is to develop
An efficient and sustainable disposal of Municipal Solid Waste a holistic system that integrates all components of MSW, including
(MSW) is a major challenge for local authorities (Badran and El- transfer, treatment, recycling and disposal transportation, pro-
Haggar, 2006). The growing volume and diversity of MSW in cessing and disposal operations (McDougall et al., 2008). An
large cities attest to the urgent need to develop operational and effective MSW management system is “complete” if all the
strategic plans to achieve sustainable goals and to mitigate the mentioned system components are integrated or otherwise
potential risk of service failure (Tan et al., 2014). Dilapidated “incomplete”.
transportation infrastructure, budgetary constraints and anachro- Given the need to implement lean strategy and economic
nistic IT systems make it harder for municipal authorities to rationalization of public service delivery, a reduction in operating
develop a seamless and well-integrated MSW system. It is partic- costs of MSW management systems is increasingly becoming one
ularly problematic in cities in the developing countries, which have of the major objectives for local municipal authorities (Chang et al.,
experienced rapid population growth, industrial transformation 2011). To develop cost-effective ISWM systems, Operations
and increased urbanization. Research (OR) techniques are widely applied to optimize waste
Daily solid waste generation is estimated to reach to 31,000 t by flows, waste allocation and transportation routing such as Ghiani
2020 (Tan et al., 2014). This puts municipal authorities under et al. (2014) and Bing et al. (2016). There have been several
immense pressure to reduce operational costs; whilst maintaining studies that developed solutions for the optimal distribution and
sustainable waste management practices. Hence, an Integrated disposal of MSW. For example, Badran and El-Haggar (2006) pro-
Solid Waste Management (ISWM) was implemented in recent years posed a Mixed-Integer Linear Programming (MILP) model to
minimize the economic cost of the collection, transportation, and
processing of MSW in Port Said, Egypt. Eiselt (2007) utilized the
* Corresponding author concept of hub location-allocation problem to optimize locations of
E-mail address: shahrooz.shahparvari@rmit.edu.au (S. Shahparvari).

https://doi.org/10.1016/j.jclepro.2019.04.250
0959-6526/© 2019 Elsevier Ltd. All rights reserved.
H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1377

landfills without integrating it with location-allocation of treat- echelon complete ISWM with specific modeling conditions and
ment and recycling plants. Dai et al. (2011) proposed a MILP model complexities at each echelon. In this study, a novel bi-objective
to plan capacity expansion of waste treatment and allocation cen- MILP model is developed in which the first objective is to mini-
ters while assuming only single type of MSW. Erkut et al. (2008) mize the total cost of transportation for both wastes and residues
developed an advanced multi-objective MILP model to solve the by concurrently optimizing the fleet size (number and fixed-costs
Location-Routing Problem (LRP) for with key functions of an ISWM of vehicles at each type) and identifying the suitable optimal
system (transfer, recycling, treatment (incineration) and disposal). routes with respect to variable costs of the fleet estimated through
Their model was able to incorporate multiple waste processing the distance traveled and load carried.
technologies for recycling and treatment centers with respect to The remainder of this paper is presented in six sections. Section
MSW type compatibility. Yet, this model was unable to integrate 2 presents the main contributions of the study, clarifying how it
waste processing and disposal of hazardous MSW (e.g. household bridges the research gap and elucidating the scope and framework
hazardous waste). of the addressed problem. Section 3 presents the main contribu-
Recently, Asefi et al. (2015b) proposed a MILP model in the tions of the study, clarifying how it bridges the research gap and
framework of LRP to minimize economic cost of ISWM system. elucidating the scope and framework of the addressed problem.
Edalatpour et al. (2018) developed a LP model to maximize profit of Section 4 explains the methodology used to solve the problem.
an ISWM system in the framework of a material flow optimization Section 5 discusses the numerical results. Managerial insights are
model (locations/technologies of the system facilities are pre- provided in Section 6. Finally, Section 7 provides conclusions and
determined and fixed). Edalatpour et al. (2018) developed a single- points to possible avenues of study for future researchers.
objective model to optimize the MSW processing network for MSW
categories (i.e. just dry and wet), non-diverse waste processing 2. Literature review
technologies and waste types. However, the model distinguishes
waste in a simplified categorization (i.e. just dry and wet) and Municipal solid waste, commonly referred as garbage, trash or
considers non-diverse waste processing technologies. Habibi et al. rubbish, is generated by residential and commercial usage. The
(2017), on the other hand, developed a multi-objective optimiza- management of solid waste is often the responsibility of municipal
tion model for location and capacity allocation for the ISWM in or local government authorities. Solid wastes can also be biode-
Tehran, Iran. Vidovi c et al. (2016) further advanced this under- gradable (i.e. food), recyclable (i.e. glass, paper), inert waste (e.g.
standing by generating a MILP model to maximize the economic construction waste, dirt, debris), electrical and electronic waste,
benefit of a two-echelon recycling network, which consists of hazardous waste (e.g. chemicals, paints), toxic waste (e.g. pesticide)
collection points, transfer stations, and end-users. Later, studies by and biomedical waste (expired pharmaceutical drugs). The man-
Harijani et al. (2017) and Asefi and Lim (2017a) have integrated agement of solid waste often consists of four major components:
qualitative criteria of ISWM system to a MILP model to maximize recycling, composting, disposal and waste-to-energy via incinera-
the economic output of the system while minimizing its non- tion. Waste management involves collection, waste handling, and
economic destructive impacts (e.g. social rejection) (Asefi and separation, storage, processing at the source, segregation, pro-
Lim, 2017a). However, the social pillar of sustainability has cessing, and transformation, transport and transfer. Given the
addressed as a constraint in the model by Harijani et al. (2017) greater diversity of waste types, the process complexity and regu-
while the model proposed by Asefi and Lim (2017a) addressed lation and compliance, the efficient and sustainable management of
the social pillar as a distinct objective to be optimized by the MSW remain a major challenge for most municipal authorities.
mathematical model. The challenges associated with MSW management in terms of
Despite the growing body of knowledge in the field of ISWM (1) capacity-allocation and (2) waste flow optimization in ISWM
optimization models, the attempts to optimize and/or consider the are mostly addressed in the frameworks of location-allocation and
fleet of waste transportation vehicles in the ISWM system were Location Routing Problem (LRP) that generate the overall routing
rare. Given the enormous waste transportation costs involved in plan for the ISWM system. Hence, the issues of (1) detailed vehicles
the ISWM system (Silva et al., 2017), it is important to develop new routing plan (consideration of vehicles in the system and finding
approaches and methods to economize the logistics network optimal routes plan vehicle by vehicle) and (2) fleet size optimi-
through optimizing waste collection and disposal using a fleet of zation (optimal number and types of required vehicles) have
different waste handling vehicles. In addition, the separation and remained the key focus in the optimization of the ISWM in this LRP
disposal of hazardous MSW in the ISWM system is largely studies. Some most recent instances are reviewed below.
neglected in the models developed during early 2000s (Asefi et al., The Vehicle Routing Problem (VRP), introduced Dantzig and
2015b). Ramser (1959), has been applied to combinatorial optimization
In addition to commonly probed decision-making questions in problems to design the optimal routes for a fleet of vehicles
ISWM optimization studies (i.e. location, waste allocation, and servicing a set of known customers (Golden et al., 2008). In vehicle
process network), the available models investigating the optimi- routing optimization problems, the key issue is to optimally allo-
zation of the number of waste transportation vehicles in the cate delivery orders to a fleet of vehicles (Repoussis and Tarantilis,
network are partially addressing the pressing issues underpinning 2010), specifically when the fleet of vehicles is heterogeneous. This
the search for end-to-end solutions for municipal waste manage- is because of the variability in their fixed and operational re-
ment. Arguably, a lack of fully integrated MSW system and waste quirements including capacity, equipment, speed and cost (Taillard,
categorization problem (e.g. just recyclables and non-recyclables) 1999). In addition, a heterogeneous fleet of waste transportation
have affected the practical application of these models when vehicles requires effective decision-making by incorporating real-
applied to complex real-world situations The limitations of the world constraints of loading and carrying different types of waste
methods employed by previous studies are also acknowledged by and residues. These waste types often need special handling care or
Santiban~ ez-Aguilar et al. (2013) for the west-central part of Mexico, equipment or are not to be mixed by loading in the same vehicle
Tan et al. (2014) for Iskandar city, Malaysia, and Xi et al. (2010) for (e.g. hazardous waste and flammable waste materials). They need
the case of city of Beijing, China. to be separately handled.
This study, therefore, aims to develop a new model for the Fleet The VRP with a variable fleet size and heterogeneity seeks to
Size and Mix VRP with heterogeneous vehicles tailored for a tri- minimize the total transportation cost by concurrently minimizing
1378 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

the fleet size (i.e. the number of vehicles and/or fixed cost of ve- unloading facility with the system components consist of a set of
hicles) and the cost of variable vehicles. This is widely known as the depots, stops (customers), and dump sites (transfer stations or
fleet size and mix VRP (FSMVRP). That is, the cost of using a mix of landfill facilities) within the operational area (e.g. catchment areas).
heterogenous vehicles in the fleet must also be economized, in For more details in Waste Collection Vehicle Routing Problem, the
addition to the cost associated with the fleet size (Golden et al., readers may refer to Kim et al. (2006) an explanation of waste
1984). FSMVRP, therefore, enables to determine the optimum collection VRP with time windows in real life problems; Buhrkal
fleet composition and size, and design of depots for returning ve- et al. (2012): the applications of waste collection VRP in the city
hicles and routes to serve customers with a heterogeneous fleet of logistics context and applied solution approaches; and Lin et al.
vehicles with varying capacities and fixed costs (Repoussis and (2014): a survey of green VRP including past and future trends
Tarantilis, 2010). The study by Rabbani et al. (2018) addressed an waste collection VRP in this context.
industrial (hazardous) waste LRP with a heterogeneous fleet of The model can be considered as a Fleet Size and Mix VRP with a
waste collection/transportation vehicles to minimize the cost of heterogeneous fleet and multiple depots (i.e. transfer stations) in
transportation as the economic objective in a multi-objective MILP which each vehicle is allocated to a particular depot and has to
model. return to the origin depot after serving the visited nodes. In the
However, ignoring transfer stations in the network, their pro- elaborated tri-echelon network framework, transfer stations (de-
posed model cannot be assumed as a practical case for a complete pots) serve as depots and waste sorting stations where waste are
ISWM system. Also, the mentioned model assumed waste pro- balled in sorted packs prior to being loaded to higher-capacity
cessing technological diversification only for treatment centers and vehicles.
the recycling centers considered to operate with singular technol- The second objective minimizes the total deviations of fair
ogy. The vehicles were not also allocated to specific depots and waste allocation to transfer stations to optimize the balance of
assumed to be free to return to any depots in their presented model. loads between transfer stations and minimize the lost capacity of
Moreover, proposing a two-stage approach in the mentioned study the transfer station. The developed bi-objective MILP model is
(i.e. routing and transportation), their proposed model addressed formulated and solved using CPLEX solver and implemented for a
vehicles routing for only transportation of waste (from depots to real-life case in the city of Tehran, Iran.
customers and then to recycling and treatment centers). That is, Previous studies addressed the problem in the frame of VRP (not
transportation of residues from waste processing centers (i.e. FSMVRP), yet there are research gaps that need to be addressed.
treatment and recycling centers) to disposal centers was not
included in the VRP framework and vehicles routes were not e The challenges associated with pre-setting the number of ve-
determined for this significant part(s) of the ISWM system. In hicles (at each type) to identify a set of routes for each of
addition, while the mentioned study addressed the problem in the vehicles.
frame of VRP (not the Fleet Size and Mix VRP), optimization of the e The potential effects of a pre-determined number of vehicles
fleet size was not addressed and it is assumed in their proposed routes to ascertain the optimal solutions to tackle real-world
approach that the number vehicles (at each type) is pre- problems.
determined so that the decision-maker finds a set of routes for e The lack of integration of transfer stations and the waste pro-
each of vehicles while the model is bound to utilize all the vehicles. cessing technology for treatment centers to represent a com-
The mentioned dependency of the resulted routes to a pre- plete ISWM system.
determined number of vehicles could highly affect the optimality e The exclusion of transportation of residues from waste pro-
of the final solutions and hence, the effectiveness of the mentioned cessing centers (i.e. treatment and recycling centers) to disposal
two-stage approach for real-world applications. Generalizing the centers in the ISWM system.
above shortcomings listed from minor to major, this is emphasized
on the necessity of addressing an efficient cost-effective Fleet Size The novelty of this proposed model addresses the above-
and Mix VRP for the ISWM system. mentioned research gaps in the current body of literature by
Jabbarzadeh et al. (2016) proposed a model to find optimal lo- concurrently optimizing of fleet size and vehicle routing for a
cations for the waste management system components while complete ISWM system considering a fair load allocation to the
minimizing economic (i.e. fuel consumption) and environmental transfer stations in a tailored tri-echelon network.
(i.e. greenhouse gas emissions) costs by finding the optimum un-
derlying routes and waste streams among the components. Yadav 3. Problem formulation
et al. (2016) addressed a problem of transfer station site selection
in Nashik, India. Their proposed approach used a mathematical 3.1. Problem framework
model to minimize transportation costs while finding suitable lo-
cations for opening transfer stations. Later, Silva et al. (2017) The problem framework is developed as a new variant of the
developed a MILP model to determine locations and capacities of Fleet Size and Mix VRP for the ISWM. This new framework is
single-type treatment centers which treat single-type waste from designed to deal with real-world complexities and operating con-
dairy farms. Recently, Sharif et al. (2018) proposed a model to straints. These includeconsideration of multiple types of MSW,
design a waste processing network and a method to find the multiple waste processing and disposal technologies, waste-
optimal outsourcing prices to optimize the economic profit of a technology compatibility, heterogeneous fleet of vehicles, waste-
municipality and maximizes the economic profit of the bidders waste (i.e. waste-vehicle) compatibility, vehicles and facilities ca-
participated in outsourcing. Most recently, Aydemir-Karadag (2018) pacity limitations and the complete ISWM system with compo-
proposed a mixed integer model for the hazardous waste LRP to nents of transfer stations, treatment-, recycling- and disposal
locate the system facilities while optimizing the transfer of haz- centers. Multiple technologies of treatment (t2T), recycling (r2R)
ardous waste and residues among the system facilities from the and disposal (d2D) are assumed for these centers with respect to
perspective of environmental protection. multiple types of MSW/residues and the corresponding waste/
The scope of this study therefore extends the Waste Collection residue-technology compatibility. Transfer stations serve as de-
Vehicle Routing Problem in which the collection phase covers pots or transfer stations. Each vehicle is allocated to a particular
waste collection vehicles collecting waste from bins to deliver in an depot (i.e. transfer stations).
H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1379

A heterogeneous fleet of vehicles requires a provide the compatible technologies and the proportions of recycled/recovered
decision-maker to consider use of different vehicles technologies materials of recycling centers (bw,r) and mass reduction at treat-
(class, capacity, etc.) and waste-waste (i.e. waste-vehicle) compat- ment centers (zw,t) are released out of the ISWM network. This
ibility (Rabbani et al., 2018). This modeling capability to cope with processing of waste generates multiple types of residues (w2q)
incompatible wastes and consider special care and equipment for such as hazardous and non-hazardous materials at the waste pro-
transporting different types of wastes including hazardous waste cessing centers in different amounts ðew;i Þ. The fleet of residue
(Boyer et al., 2013) is therefore needed. That is, transportation of transferring trucks (qU) start their routes from their origin depot
different types of waste need to be transported in specialized ve- and load the compatible ðn00w;q Þ residue from treatment/recycling
hicles compatible with the vehicle load provides the effectiveness centers and then unload it at the disposal centers with compatible
of the model to prevent the risky interactions between particular ðg4w;d Þ technologies. All vehicles (at each technology/type) must
sorts of wastes which may cause gas emissions and flame genera- then return to their allocated (origin) transfer stations after
tion (Rabbani et al., 2018). While in the Fleet Size and Mix VRP it is unloading waste/residues.
assumed that an infinite number of vehicles at each type is avail- The ISWM system is presented as a tri-echelon transportation
able, each type of vehicle has its specific characteristics including network to separately illustrate the processing and transportation
loading capacity and waste type compatibility. The proposed model of waste (Fig. 2). While a traditional VRP or Fleet Size and Mix VRP
therefore consider a heterogeneous fleet of vehicles involving ve- consist of three main segments of Depot(s), Customers (to be visited)
hicles with three technologies adopted from Habibi et al. (2017): and Destination(s) (storage plants or factories), the new variant of
waste collection vehicles (V), waste transportation semi-trailers (S), the Fleet Size and Mix VRP proposed in this study has different
residue transferring trucks (U). These vehicles also differ in waste- combination of these segments at each echelon. For instance, De-
vehicle compatibility. pot(s) and Destination(s) are the same set at the Echelon I; whilst
The schematic view of the problem network formulated in this transfer stations play both the roles of Depot(s) and Customers at
study is shown in Fig. 1. With the assumption of segregated waste Echelon II. This which cause its particular complexity.
collection, volumes of MSW (bo,i) in multiple types (o2O) are The model aims to find the optimal allocation of waste/residues
accumulated at waste generation nodes (e.g. garbage, recyclable to the system components. The problem concurrently minimizes
and hazardous). As shown in Fig. 1, the lighter vehicle class waste the fleet size, transportation cost and deviations of optimal load
collection vehicles (qV) start their route from the allocated depots as allocation to transfer stations. The total cost of transportation in the
transfer stations, collect the compatible (no,q) waste from genera- entire system is minimized by finding the minimum number of
tion nodes and then unload it to their origin transfer stations. At required vehicles and the corresponding least cost vehicle routes
transfer stations, the collected wastes are sorted in different with respect to the fixed cost (e.g. cost of possessing) of vehicles
amounts into more specified sub-types ðw2WÞ based on their and variable (running) transportation cost associated with the
properties and compatibility attributes using different types of distance travelled and the load shipped.
waste processing technologies and disposal centers available in the The key assumptions in the modeling include:
system at different amounts (pw,i). The larger vehicle class fleet of
semi-trailer waste transportation vehicles (qS) as larger vehicle - The amount of waste generations at each type is known for each
class start their route from their origin transfer stations, load the of the generation nodes.
compatible ðn0w;q Þ sorted waste from transfer stations and unload to - The capacities of all the system components are limited and
compatible recycling/treatment/disposal centers with compatible defined.
technologies (g1w;t , g2w;r , g3w;d ). The wastes are processed at the waste - Each vehicle is allocated to one depot at a time.
processing centers (i.e. recycling and treatment centers), with

Fig. 1. The ISWM system.


1380 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

of the vehicles and variable (running) costs of the vehicles as


measured by the load and distance traveled of the vehicles.
- Minimizing the total deviation of the fair load allocation to
transfer stations in order to optimize the balance of workload
between the transfer stations and minimize their lost capacities.
The fair (balanced) allocation of waste to a transfer station is
estimated as the ratio of the capacity of the transfer station to
the total capacity of the transfer stations in the ISWM system.

The data gathered in this study indicates that some transfer


stations are underutilized whilst others perform to their full ca-
pacity. However, treatment centers operate to their fare load as
they work under certain technologies. The issue did not observe for
the other system components (e.g. treatment centers) as and allo-
cations of waste/residue to these facilities are almost fare because
of the need of particular waste/residue types for the certain types of
technologies. Hence, the second objective is formulated to fairly
allocate the capacity to the transfer stations. The proposed math-
ematical model therefore generates the following outputs:

- The load and type of waste/residue to be allocated/processed at


each system's component and technology
- The fleet (number and type) of vehicles to be employed in the
ISWM system
- The fleet (number and type) of vehicles required at each depot
- The routes of compatible vehicles to collect and transfer waste/
residues in the entire system (i.e. routes among the generation
nodes and all the system components)
- The load (amount and type) of vehicles on each segment of the
routes

The objective functions and constraints are formulated as


follows.

3.2.1. Objective functions


Fig. 2. Tri-echelon logistics network and transportation system in ISWM: a schematic
demonstration.
Equation (1) minimizes (i) the number and fixed costs of vehi-
cles at each type, (ii) the running cost of vehicles associated with
- There exist multiple depots (transfer stations) as a multi-depot the load and distance traveled. Equation (1) involves six terms or
problem. three pairs where each pair consecutively associates to an echelon
- There are multiple types of treatment technologies (t2T), in consecutively (i.e. Echelons I-III). For each pair, the first term
recycling (r2R), and disposal technologies (d2D), with one of counts for the fixed cost and the second term computes the variable
them active at each treatment center. cost of the corresponding echelon. In doing so, the first term in
- Each type of waste sorted at transfer stations is compatible to Equation (1) counts the fixed cost of each vehicle of Echelon I (i.e. v)
q
exactly one of the treatment/recycling/disposal technology. if the vehicle used where binary variable xi;j shows utilization of
q
- Each type of residue generated at transfer stations is compatible vehicle q (xi;j ¼1) if the vehicle starts from the origin(s) of the
to exactly one of the disposal technologies. echelon (K) to destination(s) in the echelon (G). The second term in
- The amount of each waste/residue type accumulated at each Equation (1) computes the variable cost or transportation cost of
customer node (see customers at each echelon in Fig. 2) does not Echelon I for vehicles of the echelon (i.e. v) which departs from the
exceed the capacity of the fleet. echelon origins (i.e. G for Echelon I) to the echelon customers/
- Each customer node is served once by a compatible vehicle. destinations (i.e. G and K for Echelon I). The procedure explained
- The partial waste/residue collection is not allowed. above is similarly applied for Echelon II (i.e. second pair of terms in
- Vehicles at each type have the same capacity. Equation (1)) and Echelon III (i.e. third pair of terms in Equation
- Transportation cost is directly proportional to the distance (1)).
traveled and load shipped and vehicle types. The second objective formulated in Equation (2) to minimize the
total deviation of fair load allocation to transfer stations to mini-
mize lost capacities of transfer stations and provide the load bal-
3.2. Mathematical model ance and fair allocation of waste to transfer stations. Equation (2)
represents deviations of the ideal load ratios of waste processing
In this study, we developed a novel bi-objective mixed-integer (i.e. fair balance of loads) at transfer stations from actual loads of
optimization model to address the complexity of the ISWM Fleet waste processing at transfer stations on the basis of their capacities.
Size and Mix VRP. The model optimizes the ISWM system with In doing so, the total lost capacities of transfer stations are summed
respect to two objectives: in Equation (2). That is, the ideal (i.e. fair balanced load) of a transfer
station is assumed as the ratio of the transfer station's capacity to
- Minimizing the total cost of transportation in the ISWM system the total available capacity of transfer stations in the system (i.e.
imposed by the fixed costs (e.g. cost of purchase or possessing) summation of capacities of the transfer stations). The ratio of actual
H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1381

load of waste processing at a transfer station is also calculated as


the total amount of waste (in all types) processed at the transfer
q
stations to the total amounts of the waste in the system (at all xi;j  zq  0 ci; j2G; cq2V2Q (5)
generation nodes).

P XX q
X X X q q q
X X X q
MinimizeF1 ðxÞ ¼ fq xi;j þ ci;j xi;j ui þ fq xi;j
q2V2Q i2Gj2K∪G q2V2Q q2S2Q
i2K j2G
P P P X X Xi2Kqj2M∪H∪L
þ cqi;j xqi;j uqi þ fq
xi;j (1)
i2K j2K∪M∪H∪L q2S2Q q2U2Q i2K j2M∪H
P P P q q q
þ ci;j xi;j ui
i2M∪H j2M∪H∪L q2U2Q

X X q
xi;j  ‫א‬K  1 cq2S2Q (6)
i2K j2K∪M∪H∪L
 P K 
 yj;o 
X pKj 
X X q
MinimizeF2 ðxÞ ¼ o2O
P K  P P  (2) xi;j ¼ zq cq2S2Q (7)
 p b 0
j2K  i o;i  i2K j2M∪H∪L
i2Ko2Oi 2G
0

X X q
xi;j  1 cq2U2Q (8)
i2K j2M∪H

3.2.2. Route structure: starting from depots X X


P P q xqi;j ¼ zq cq2U2Q (9)
The simplified equation term of i2DepotðsÞ j2CustomerðsÞ xi;j ¼ 1
i2K j2M∪H
is frequently utilized in similar studies (Rabbani et al., 2018) to
formulate utilization of vehicles and starting routes from depots.
q
However, using the term in the mentioned form (i.e. equality to 1) xi;j  zq  0 ci; j2M∪H; cq2U2Q (10)
forces to utilize every vehicle (as predetermined in the vehicle set)
resulting to the questionability of optimization the fleet size. This
shortcoming is effectively addressed in the proposed model by
stating a set of constraints (i.e. Eqs. (3)e(10)) to replace it. 3.2.3. Route structure: traditional VRP
Equation (3) ensures that a vehicle in Echelon I can start its route Equation (11) ensures if a vehicle in Echelon I entered to a
from at most one depot (i.e. K) to serve a customer (i.e G). This customer, must be out to either the next customer or the final
constraint enforces vehicle to be allocated at most one depot. destination. The constraint (11) is a traditional constraint of VRP to
Equation (4) shows that if a vehicle in Echelon I starts from an ensure a vehicle cannot finish its route before returning to its final
origin (depot) to serve a customer, then its corresponding variable destination or its origin (i.e. depot). Equations (12) and (13)
showing the vehicle usage (zq) equals to 1; otherwise 0. Equation describe the same constraint as Equation (11) for Echelons II and
(5) ensures that a vehicle in Echelon I cannot transfer between two III with respect to the corresponding origin, customer and desti-
customers unless the vehicle started its route from an origin nation nodes of the echelons.
(depot). That is, the starting point for a vehicle in Echelon I cannot X q
X q
be a customer. Equation (5) ensures that transferring between two xi;j  xj;i0  0 cj2G; cq2V2Q (11)
customers of Echelon I (i.e. G) by a vehicle of the echelon can be i2K∪G i0 2G∪K
conducted only if the vehicle is an assigned vehicle (i.e. zq ¼ 1). In X X
q q
other words, the equation ensures that a vehicle in Echelon I cannot xi;j  xj;i0  0 cj2K; cq2S2Q (12)
transfer between two customers unless the vehicle had started its i2K i0 2K∪M∪H∪L
route from an origin (depot). That is, the starting point for a vehicle
X X
in Echelon I cannot be a customer. Equations (6) and (7) describe xqi;j  xqj;i0 ¼ 0 cj2M∪H; cq2U2Q (13)
the same constraints as Equations (3) and (4) for Echelon II with i2K∪M∪H i0 2M∪H∪L
respect to the vehicles, the corresponding origin, customer and
destination nodes of the echelon. Equations (8)e(10) describe the
same constraints as Equations (3)e(5) for Echelon III with respect
to the vehicles, the corresponding origin, customer and destination 3.2.4. Route structure: serving customers (loading waste and
nodes of the echelon. residue)
Equation (14) ensures that for loading waste types, a customer
XX q
xi;j  1 cq2V2Q (3) node in Echelon I (i.e. G) must have one input compatible vehicle
i2K j2G per type of its accumulated waste which from either a previously
visited customer (i.e. G) or an origin node (i.e. K). Equations (15) and
XX q (16) describe the same constraint as Equation (14) for Echelons II
xi;j ¼ zq cq2V2Q (4)
and III with respect to the corresponding origin, customer and
i2K j2G
destination nodes of the echelons. Hereafter, when it is stated that
the same constraint applies to other echelons means that using the
1382 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

same explanation, new equations for the other echelons is yielded X X 


with consideration of the echelon vehicles, origins, customers. uqi  uqj þ no;q jqo xqi;j  no;q jqo
o2O o2O
X X q 
xi;j no;q ¼ 1 cj2G; co2O (14)  bo;j ci; j2G; cq2V2Q (22)
i2K∪G q2V2Q
X q
X
X X bo;i no;q  ui  no;q jqo ci2G; cq2V2Q (23)
xqi;j n0w;q ¼1 ci2K; cw2W (15) o2O o2O
j2K∪M∪H∪L q2S2Q
XX q q
X X xi;j bo;j no;q  uj cj2G; cq2V2Q (24)
xi;j n00w;q
q
¼1 ci2M∪H; cw2q (16) i2K o2O
j2M∪H∪L q2U2Q
!
X X  
uqj  no;q jqo þ xqi;j bo;j  jqo cj2G; cq2V2Q
o2O i2K
3.2.5. Route structure: vehicles allocation and returning to depots (25)
Constraint (17) enforces that if a vehicle in Echelon I starts from
a depot to visit a customer, then the vehicle belongs to the depot X X 
(i.e. origin depot) and cannot be served by other depots so that uqi  uqj þ n0w;q jqw xqi;j  n0w;q jqw
must back to the origin depot after visiting the customer nodes. w2W w2W

Equations (18) and (20) present the same constraint as equation  pw;j ci; j2K; cq2S2Q
(14) for Echelons II and III respectively. Constraint (19) forces that if
a vehicle in Echelon II departs from an origin (i.e. K) to a customer (26)
(i.e. M∪H) or a final destination (i.e. L), then must back to the origin. X X
Equation (21) presents the same constraint as Equation (19) for pw;i n0w;q  uqi  n0w;q jqw ci2K; cq2S2Q (27)
Echelon III. w2W w2W
X q q
X q XX
xi;j ¼ Z i ¼ xj;i ci2K; cq2V2Q (17) xi;j pw;j n0w;q  uj
q q
cj2K; cq2S2Q (28)
j2G j2G i2K w2W

X q
X q q X X 
xj;i  xi0 ;j ¼ Z j uqi  uqj þ n00w;q jqw xqi;j  n00w;q jqw
i2K∪M∪H∪L i0 2K w2q w2q
X q 
¼ xi00 ;j cj2K; cq2S2Q  ew;j ci; j2M∪H; cq2U2Q
i00 2M∪H∪L
(29)
(18)
X X
X ew;i n00w;q  ui  n00w;q jqw ci2M∪H;
q
q q cq2U2Q (30)
xi;j  xj;i0 ci2K; cj2M∪H∪L; cq2S2Q (19) w2q w2q
i0 2K
XX
X q q
X q xqi;j ew;i n00w;q  uqj cj2M∪H; cq2U2Q (31)
xi;j ¼ Z i ¼ xi0 ;j ci2K; cq2U2Q (20) i2K w2q
j2M∪H i0 2L

X X X q q
q
xi;j 
q
xi0 ;i ci2L; cj2K; cq2U2Q (21) uqj  n00w;q jqw þ xi;j ew;j  jw
i0 2M∪H w2q i2K
!
 cj2M∪H; cq2U2Q (32)

3.2.6. Route structure: evading sub-tours and vehicle loading


capacity
Equation (22) presents that a vehicle in Echelon I serve the next
customer to load its waste if the total load of vehicle after visiting
the customer does not exceed the remained capacity of the vehicle. 3.2.7. Route structure: unloading waste
Equation (23) ensures that if a customer visited by a vehicle in Constraints (33)e(35) enforce unloading waste at treatment,
Echelon I, the total demand (i.e. waste) of the customer will be recycling and disposal centers with compatible technologies.
loaded while the waste type loaded is compatible with the vehicle Unloading waste to a customer/destination node with a compatible
and does not exceed the total capacity of the vehicle. This equation technology is forced for treatment, recycling and disposal facilities
also ensures that partial collection load of nodes is not possible. by Equations (33)e(35), respectively. This set of equations guar-
Equations (24) and (25) ensures that the total demand of customers antee that a vehicle can unload its load to the corresponding
in Echelon I must be satisfied using the fleet of the system (i.e. destination only when the vehicle carries the compatible type of
compatible vehicles). Equations (26)e(28) state the same con- waste (compatibility with the vehicle) which is compatible to the
straints as Equations (22)e(24) for Echelon II. Equations (29)e(32) existing technology of waste processing at the destination.
state the same constraints as Equations (22)e(24) for Echelon III. XX
Constraints (22)e(32) state for vehicles’ load capacity limita-
q
xi;j  n0w;q g1w;t aM
t;j ci2K; cj2M; cq2S2Q (33)
w2W t2T
tions and traditional VRP constraints for evading sub tours (Rabbani
et al., 2018).
H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1383

X X by compatible vehicles (compatibility of carrying waste type with


q
xi;j  n0w;q g2w;r aHr;j ci2K; cj2H; cq2S2Q (34) vehicle type).
w2W r2R
X X
X X yKj;o ¼ xqi;j uqi no;q cj2K; co2O (41)
q
xi;j  n0w;q g3w;d aLd;j ci2K; cj2L; cq2S2Q (35) q2V2Q i2G
w2W d2D
X X
xi;j ui n0w;q qf cj2M;
q q
yM
j;w ¼ cw2W (42)
q2S2Q i2K

3.2.8. Route structure: unloading residue X X


Given the same explanation as those for Equations (33)e(35), yH
j;w ¼ xqi;j uqi n0w;q cj2H; cw2W (43)
constraint (36) enforces unloading residue at disposal centers with q2S2Q i2K
compatible technologies.
X X
XX y1L xi;j ui n0w;q cj2L; cw2W
q q
q
xi;j  n00w;q g4w;d aLd;j ci2M∪H; cj2L; cq2U2Q j;w ¼ (44)
q2S2Q i2K
w2qd2D
(36) X X 00
y2L
j;w ¼ xqi;j uqi nw;q cj2L; cw2q (45)
q2U2Q i2M∪H

3.2.9. Demands at transfer stations


Constraint (37) states the amounts of different waste types
sorted at transfer stations for sending to waste processing and
disposal centers. 3.2.12. Supplying all demands
X Constraint (46) guarantees to supply all demand (generated
pw;j ¼ aw;j yKj;o cj2K; cw2W (37) waste) of generation nodes to transfer stations. Constraint (47)
o2O guarantees to supply all demand (sorted waste) of transfer sta-
tions to waste processing and disposal centers. Constraint (48)
guarantees to supply all demand (sorted residues) of treatment
and recycling centers to disposal centers.
3.2.10. Demands at waste processing centers
Constraints (38) and (39) state the amounts of residues gener- XX XX
bo;i ¼ yKj;o (46)
ated at treatment and recycling centers respectively. Constraint
i2Go2O j2K o2O
(38) computes the total residues generated at treatment centers on
the basis of ratios of mass reduction of input waste for different XX X X X X XX
types of wastes after different treatment processes performed by pw;i ¼ yM
j;w þ yH
j;w þ y1L
j;w (47)
different treatment technologies. Similarly, constraint (39) results w2W i2K j2M w2W j2H w2W j2L w2W

in the total amount of residues generated at recycling facilities on X X XX


the basis of the input waste to recycling centers and ratios of mass ew;i ¼ y2L
j;w (48)
reduction for different types of wastes after different recycling w2qi2M∪H j2L w2q
processes performed by different recycling technologies.
Constraint (40) states the amounts of different residue types
generated at treatment and recycling centers for sending to
disposal centers.
3.2.13. Facilities capacity limitations
XX  
cM
i ¼ yM M
i;w at;i 1  zw;t ci2M (38) Constraints (49)e(52) show the capacity limitations of the sys-
w2W t2T tem facilities. That is, the amount of total waste/residue processing
at a system facility cannot exceed the capacity of the facility. The
X X  
cHi ¼ y1L H total amounts of processed waste at a system's facility (i.e.
i;w ar;i 1  bw;r ci2H (39)
w2W r2R computed in constraints (41)e(45)) cannot exceed the available
capacities (for different types of input waste) of the system's fa-
X X cilities. In doing so constraints (49)e(52) applied to Transfer sta-
ew;i ¼ aM M M
t;i mw ci þ aH H H
r;i mw ci ci2M∪H (40)
tions, Treatment centers, Recycling centers, and disposal centers.
t2T r2R
X
yKj;o  pKj cj2K (49)
o2O
3.2.11. Facilities workloads (allocated waste/residue to system
X X
components) yM M
aM
j;w  pj t;j cj2M (50)
Constraints (41)e(45) show amounts of sorted, treated, recycled w2W t2T
and disposed waste/residue at the corresponding waste compo-
nents respectively. That is, the actual amounts of total waste/res- X X
yH H
j;w  pj aH
r;j cj2H (51)
idue processed at the system facilities are determined by Equations w2W r2R
(41)e(45). In doing so, each of these five equations counts the
processed waste for the corresponding facility (constraint (41) for X X X
y1L
j;w þ y2L L
j;w  pj aLd;j cj2L (52)
Transfer station, (42) for Treatment centers, …, (45) for Disposal
w2W w2q d2D
centers), on the basis of input compatible waste (compatibility with
available waste processing technology) transferred to these centers
1384 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

3.2.14. Variable types shown in Algorithm 1.


The non-negativity and binary variables of the model are pro-
Algorithm 1. The sequential lexicographic optimization algo-
vided in Constraints (53) and (54).
rithm - *Adopted from (Ocampo-Martinez et al., 2008)
þ
uqi ; pw;i ; ew;i ; yKi;o ; yM H 1L 2L M H ML HL
i;w ; yi;w ; yi;w ; yi;w ; ci ; ci ; ci;j ; ci;j 2R (53)
1: set the order of objectives F1>/>Fk
2: find z1 where z1 ¼ minF1 ðxÞ
xqi;j ; zq ; Z qi 2f0; 1g (54) 3: for i ¼ 2 to k 

4: find zi where zi ¼ min½Fi ðxÞFj ðxÞ  zj ; j ¼ 1; …; i  1
The nonlinear terms in objective functions and Equations (28), 5: end for 
(31) and (42)e(45) are modified by defining some auxiliary vari- 
6: The lexicographic optima is x where x 2fx 2SFj ðxÞ  zj ;
ables and constraints to form a linear mixed-integer programming j ¼ 1; …; kg.
model. The linearization approach is presented in Appendix II.
Stating the method in a two-level optimization order (Sun et al.,
1999), the addressed two-objective problem is solved in the
4. Solution approach e Multiobjective optimization
following order. Here, the first level optimization is z1 ¼
min½F1 ðxÞ : x 2S, where z1 is the optimum objective value. This
The proposed multi-objective optimization problem is shown as 1
Problem 4 where x is the vector of decision variables, F1(x), F2(x) are
optimization
    obtains the optimal decision set S ¼
level
x2S : F1 x ¼ z1 . The next level minimization is then obtains
the objective functions, and S is the feasible region for the Con-
straints (3)e(54). h i
z2 ¼ min F2 ðxÞ : x 2 S1 ;
minF1 ðxÞ
minF2 ðxÞ h i
s:t:
(I) S2 ¼ x 2 S1 : F2 ðxÞ ¼ z2 :
x 2 S;
Performing the sequential optimization method (See Algorithm
In this section, two well-known multi-objective optimization 1), the proposed model is reformulated into Problem II when the
methods (i.e. Lexicographic and Goal Programming) adopted to first objective is the dominant objective.
solve the problem (i.e. Problem I) are briefly presented. The
addressed problem are solved under two different conditions minF2 ðxÞ
(scenarios): Preemptive and Non-Preemptive objectives. The Pre- s:t:
emptive objectives define the condition that there is a strict F1 ðxÞ  z1 (II)
dominance order between the objectives ðuF i0 Þ  ðuF j0 Þ. In this and
case, a sequence of solves is implemented to optimize each objec- constraints ð3Þ  ð54Þ:
tive (Ignizio, 1976). The Non-Preemptive objectives represent the The Hierarchical Lexicographic Optimization (HLO) is a variant
condition when the priorities of the objectives can be reflected by a of Lexicographic optimization with a constraint relaxation
set of linear weights (ui). That is, in this condition, the decision- approach by slightly increasing the Right Hand Side (RHS) of the
maker can apply his/her preferences (i.e. the priorities of the ob- constrained objective (fj(x)) by a ratio of its optimal value (zj ). That
jectives) using a set of scalar weights. is, the HLO allows the higher dominance objectives to deviate ε%
from their optimums (Marler and Arora, 2004). Accordingly, the
4.1. Preemptive objectives HLO method is utilized by reformulating the proposed model into
 
Problem III (when wF 01  ðwF 02 Þ where 0 < εj  100.
To solve the proposed multi-objective model (i.e. Problem I), the
Sequential Lexicographic Optimization (SLO), Hierarchical Lexico- minF2 ðxÞ
graphic Optimization (HLO) and the Goal Programming (GP) are
s:t:
adopted.  ε 
F1 ðxÞ  1 þ 1 z1 (III)
100
4.1.1. Lexicographic optimization: SLO and HLO
and
The lexicographic method is a sequential approach in which the
decision-maker considers that the k objectives in the model follow constraints ð3Þ  ð54Þ:
a priority (dominance) order. The solution obtained by the lexico-
graphic method is a particular kind of Pareto-optimal solution
while a priority (importance) order for the objectives are consid-
ered (Ocampo-Martinez et al., 2008). 4.1.2. Goal Programming
Applying the lexicographic method to the multi-objective Goal Programming is a solution approach to optimize problems
problem (Problem I), a sequential order of single objective con- with multiple and conflicting objectives possibly measured in
strained optimization problems is solved. That is, the highest pri- nonhomogeneous units (Ignizio, 1985). The Goal Programming
ority objective function (e.g. F1) is first minimized subject to the method considers the objectives concurrently and minimizes the
original constraints. Then, the second objective function (e.g. F2) is total deviations of the objectives from their optimal values simul-
minimized while in addition to the original constraints, a new taneously. Accordingly, the Goal Programming is adopted in this
constraint (i.e. binding the first objective function to its optimal research to address minimizing the two objectives concurrently
value from the first step) is added to ensure that the highest priority when they are assumed at the same importance. The method
objective function preserves its optimal value. In general, when a proceeds by formulating an objective Fi ðxÞ into a Goal by using an
multi-objective minimization problem consists of k objective Aspiration Level Di. That is, the Goals are formulated into the form
function ordered from F1 to Fk, the sequential lexicographic opti- Fi ðxÞ  Di where  is a vector of relational operators f ; ¼ ;  g. In
mization method is implemented in a sequential algorithm as doing so, the positive and negative deviations from (i.e. slacks) from
H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1385

the Aspiration Level are defined as dþ 


i  0 and di  0 to be added to Non-Preemptive objectives, the Weighted Goal Programming
the Goals terms. The method seeks to minimize the appropriate (WGP) is adopted to optimize the problem with respect to a com-
slacks (e.g. dþ
i in minimizing cost-based objectives) to obtain the bination of linear objective weights.
Goals. Accordingly, the proposed model is formulated in the form as
presented in Problem IV while it should be noted that we use the
Utopia point (i.e. the results by single-objective optimization for
the two objectives: D1 ¼ z1 and D2 ¼ z2 ) as the Aspiration Levels
and the objective function is scaled by multiplying dþ i with the 4.2.1. Weighted Goal Programming (WGP)
corresponding objective optimum value (i.e. 1/ zi ). The Weighted Goal Programming (WGP) is adapted to handle
þ þ
the two objectives while seeking to minimize the total deviations
mind1 þ d2 from the goals. The WGP takes the two goals into consideration
s:t: simultaneously using an achievement function in which the total
F1 ðxÞ  dþ 
1 þ d1 ¼ D1 weighted deviation from the goals is minimized. The method
F2 ðxÞ  dþ 2 þ d
2 ¼ D2
(IV) provides the decision-maker with the capability of reflecting his/
and her preferences regarding the relative importance of the goals into
constraints ð3Þ  ð54Þ; the model. Due to involvement of the two different goals with
dþ  þ 
1 ; d1 ; d2 ; d2  0: different measurement units, a kind of incommensurability appears
in the formulation of the goals. To tackle this issue, a scheme of
normalization needs to be adopted. Hence, we use a mostly-applied
normalization method by factoring the values of utopia (Karsak
4.2. Non-Preemptive objectives et al., 2003). Accordingly, using the WGP, the proposed problem
is reformulated into Problem V where ui denotes the weight coef-
P
Considering the two objectives of the addressed problem as the ficient for the goal i, and i ui ¼ 1.

Fig. 3. Map of the case study area- South of Tehran, Iran.


1386 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

Divisions (MDs) numbered 12, 15, 16, 19 and 20. Accordingly, the
dþ dþ waste generation nodes in the study are labeled as g1, g2, g3, g4 and
Minu1 2
þ u2 2
z2 z2 g5 representing MD12, MD15, MD16, MD19 and MD20 (see Fig. 7).
s:t: The waste generation nodes are supplying by three transfer stations
þ 
named Harandi, Azadegan, Shahre-rey. Respectively, the transfer
F1 ðxÞ  d1 þ d1 ¼ D1 stations (depots) are denoted by k1, k2, and k3. The waste pro-
þ  (V) cessing facilities serving the nodes involves a material recovery
F2 ðxÞ  d2 þ d2 ¼ D2
facility (h1) compatible with dry-recyclable, a compost facility (h2)
and compatible with wet-recyclable, a treatment center with inciner-
constraints ð3Þ  ð54Þ:; ation technology (m1), a treatment center with physical-chemical
technology (m2), a general disposal site (l1) and a sanitary land-
dþ  þ 
1 ; d1 ; d2 ; d2  0: fill cell (l2) compatible for hazardous waste/residue where all the
waste processing and disposal facilities are centralized in Aradkuh
(i.e. a center with 500 ha area located 35 km away from south of
5. Empirical results Tehran) (Asefi and Lim, 2017a). The southern part of Tehran and the
corresponding ISWM system is schematically displayed in Fig. 3.
The accuracy of the proposed model and the effectiveness of the Adopting from (Asefi and Lim, 2017a), it is assumed that waste at
solution approaches are testified by implementing the model for a generation nodes accumulated in three types of garbage (o1),
numerical experiment. The case study is chosen as the recently recyclable (o2) and hazardous (o3) and further sorted at transfer
conducted case in (Edalatpour et al., 2018) for the southern part of stations to w1 (compatible with l1), w2 (compatible with l2), w3
Tehran, Iran. The authority responsible for managing municipal (compatible with h1), w4 (compatible with h2), w5 (compatible
waste in Tehran is Tehran Waste Management Organization with m1) and w6 (compatible with m2). The waste processing
(TWMO). The southern part of Tehran covers five Municipality centers (treatment and recycling facilities) generates two types of

Fig. 4. Map of vehicle's with technology V.


H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1387

residues: non-hazardous residue ðw1Þ and hazardous residue ðw2Þ. 2013). To take into account other related costs of transportation
The problem is solved for one cycle/shift assuming two shifts a day (e.g. insurance, vehicle depreciation, and driver wage), the constant
and two cycles per shift. The capacity of facilities and amount of factor of 2 is multiplied by the unit cost of transportation (Boyer
waste generation at southern Tehran derived from (Asefi and Lim, et al., 2013). Table 1 summarizes the data and sources used in
2017a) and adjusted for one cycle while the proportions of this study.
different waste types are adopted from the reported values in The addressed problem is solved using CPLEX 12.3 on a com-
(Abduli et al., 2015). The total proportion of mass reduction (release puter with Core i7 2.40 GHz processor and 16 GB RAM memory. To
out of network) at recycling centers is assumed as 80% and mass initiate the solving, the fleet of waste management system is
reduction by incineration and physical-chemical treatment are assumed at total number of 60 vehicles with 20 vehicles for each
assumed as 80% and 20% respectively (Samanlioglu, 2013). vehicle technology V: v01 , …,v20 ; S: s01 , …,s20 and U: u01 , …u20 ). The
The loading capacities of trucks and semi-trailer vehicles are obtained results by solving the model with respect to each objec-
considered as 63, 72 and 53 t for technologies v, s, and u respec- tive function singularly are reported in Table 1. As can be seen in
tively (Habibi et al., 2017). For fixed cost of vehicles, we consider the Table 2, the utopia and nadir points are found at (2,505,154.507,
purchase cost and estimated by interviewing TWMO organization 0:0001095273y0) and (3,635,621.537, 0.259) respectively.
members as $100,000, $130,000 and $70,000 for technologies V, S Applying the Lexicographic optimization method using the HLO
and U respectively. To calculate the cost of transportation, actual approach, the optimal solutions by the method found at F HLO F1[F2 ¼
distances for each pair of nodes are derived from Google Map API. ð2; 505; 918:255; 0:021Þ and F HLO
F2[F1 ¼ ð2; 905; 576:515;

The average fuel consumption for a truck is assumed 0.3 L per km 0:0001095273 y0) for  u F1 [
0 u0 F2 (i.e. minF2 F1 ¼ z1 ) and
(Samanlioglu, 2013) and the average cost of fuel is considered as 0 
u F2 [u F1 (i.e. minF1 F2 ¼ z2 ) respectively. Assuming u0 F1 ≡u0 F2 ,
0 

$0.4 per litre in the study region. To consider the special care and the solution found at F GPF1≡F2 ¼ ð2; 705; 495:78; 0:00079 y0Þ  by
 þ
vehicle technology/equipment associated with shipping hazardous applying the Goal Programming method when solving mindþ 1 d2 ¼

waste/residue, the unit cost of transportation for this material type d2 . The solutions obtained by the two multi-objective optimization
is assumed to be 43% more than other waste types (Samanlioglu, methods under the condition that the objectives are Preemptive are

Fig. 5. Map of vehicle's with technology S.


1388 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

shown in Fig. 8 along with the results by single-objective linear weights are presented in Table 5 and plotted in Fig. 10a.
optimizations. Among the presented six solutions (S13eS18), it can be seen that
The overall performance of the system for single and multi- three solutions are non-dominated producing the efficient frontier
objective objective optimizations is summarized in Table 3. It can shown in Fig. 10b. This issue affirms that while the WGP is capable
be seen that optimizing the transportation cost singularly (S1 ) to obtain non-dominated (efficient) Pareto optimal solutions, it may
resulted in being idle for one transfer stations (i.e. k3 which played also yield weakly Pareto or inefficient solutions along with Pareto
only as a depot) with load ratios of 0% while the other two transfer optimal solutions (Marler and Arora, 2004). A Pareto-optimal so-
stations forces to work with loads higher than their expected lution is a solution which cannot be improved in one objective (e.g.
balanced load. On the other hand, while minimizing the balance F1) without a deterioration in the other objective (e.g. F2). On the
deviation singularly ðS2 Þ resulted to almost fair distribution of load other hand, while the lexicographic method evades producing
to transfer stations, the transportation cost increased by almost 45% inefficient solutions, its drawback is the computational expanse of
compared to z1 . The multi-objective optimization (i.e. S3 , S4 and S5 ) having to solve a sequence of problems (Marler and Arora, 2004).
achieved minimized costs when no transfer station is idle (i.e. The definitions of the above solution types are presented in
utilizing all the three transfer stations at almost fair ratios). Appendix III.
The results obtained by the HLO method when ε1 varied in a Implementing the model successfully, the corresponding results
dispersed range from the value resulting in the optimum of F1(x) to to solution S3 (i.e. F HLO
F1[F2 ) as an instance of the obtained solutions
the nadir (i.e. 1e40%) are presented in Table 4 and Fig. 9a The Pareto are reported in Tables 6 and 7 to provide more detailed outcomes
frontier obtained by the method is also shown in Fig. 9b. The (see Figs. 4,5 and 6). The solution resulted to a reasonable fare
conflicting behavior of the two objectives is also evident in this allocation of waste to the transfer stations (i.e. the total deviation of
case. That is, improvement in one objective cannot be achieved almost 2.1% with no idle transfer station) while minimize the fleet
without deterioration in another objective. size (10, 12 and 2 vehicles for vehicles V, S and U) and found the
Considering the condition Non-Preemptive objectives, the re- corresponding least cost routes by serving as more customer as
sults obtained by the WGP corresponding to a set of dispersed possible at each route.

Fig. 6. Map of vehicle's with technology U.


H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1389

Fig. 7. Map of waste collection to the g points.

Table 1
Data and sources.

No Data Source

1 Case study area (infrastructure and network) Edalatpour et al. (2018)


2 Waste at generation nodes and types Asefi and Lim (2017a)
3 The capacity of facilities and the amount of waste generation Asefi and Lim (2017a)
4 The proportions of different waste types Abduli et al. (2015)
5 The loading capacities of trucks and semi-trailer vehicles Habibi et al. (2017)
6 The unit cost of transportation for hazardous material type Samanlioglu (2013)
7 The fixed cost of vehicles Interview (TWMO)
8 The total proportion of mass reductions Samanlioglu (2013)
9 The distance of transportation Google Maps API
10 The average fuel consumption rate Samanlioglu (2013)
11 Other related costs of transportation Boyer et al. (2013)

6. Managerial implications

This study offers several practical insights that can inform


Table 2 evidence-based decision-making to help reduce operational costs,
Solutions for objective functions.
enhance sustainable practices and improve the efficiency of solid
minF1(x) minF2(x) waste management system. The findings of the developed model
F1(x) z1 ¼ $2; 505; 154:507 $ 3,635,621.537 can help identify deficiencies and supply chain bottlenecks within
F2(x) 0.259 z2 ¼ 0:0001095273y0 the current ISWM system. The model enables the authority to
1390 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

Fig. 8. : Multi-Objective optimization: Preemptive objectives.

Table 3
Performance of the ISWM system for different optimizations.

Solution Method Condition (Scenario) Number of vehicles Load at k1 (%) Load at k2 (%) Load at k3 (%) Total Total CPU
No. used VeS e U - Total Expected- Actual Expected- Actual Expected-Actual deviation cost Time (s)
($M)

S1 Single-objective ðz1 Þ Min F1 ðxÞ 8e12e 2 22 27.7e39.8 59.4e60.2 12.9e0 25.9% 2.51 725.68
S2 Single-objective ðz2 Þ Min F2 ðxÞ 14e15 e 4 33 27.7e27.7 59.4e59.4 12.9e12.9 0% 3.64 497.92
S3 Lexicographic Multi- u0F1 [ uF2 :Min
0 8e12 e 2 22 27.7e26.6 59.4e59.7 12.9e13.7 2.1% 2.51 1000.21
objective: HLO ðF HLO
F1[F2 Þ F2 ðxÞF1 ðxÞ  z1 ε1 ¼
1
S4 Goal Programming Multi- u0F1 ≡u0F2 :Min dþ þ
1 þ d2 13e13 e 2 28 27.7e27.7 59.4e59.4 12.9e12.9 0.1% 2.81 837.42
objective: GP ðF GP
F1≡F2 Þ
S5 Lexicographic Multi- u0F1 ≪u0F2 :Min 12e12 e 2 26 27.7e27.7 59.4e59.4 12.9e12.9 0% 2.91 1000.21
objective: HLO ðF HLO
F1≪F2 Þ F1 ðxÞF2 ðxÞ  z2 ε2 ¼
7

Table 4
Solutions by HLO.

Solution No. ε1 F1(x): Total cost ($M) F2(x): Total deviation (104) Pareto Optimality

S4 1 2.505918255 210 Efficient (non-dominated Pareto)


S5 6 2.606067999 20 Efficient (non-dominated Pareto)
S6 11 2.776022191 7.817583 Efficient (non-dominated Pareto)
S7 15 2.845858342 7.817583 Weakly efficient
S8 20 3.005994678 7.275256 Efficient (non-dominated Pareto)
S9 25 3.075893687 7.2753 Weakly efficient
S10 30 3.235753097 7.275257 Weakly efficient
S11 35 3.376235668 0.5781702 Efficient (non-dominated Pareto)
S12 40 3.505523381 0.5781394 Efficient (non-dominated Pareto)

estimate the under or over capacity of fleet to manage by solid fleet size and allocation; capacity allocation and vehicle routes
waste types in each echelon. That is, the municipal authorities can under different technological regimes. These outputs in turn will
then develop a system to isolate and manage different types of allow vehicles of different capacities and compatibilities to be
waste (eg. recyclable, hazardous, garbage) that must be collected assigned to the proximal facility, alleviate idling of vehicles and
and processed within the ISWM system. reduce fuel costs on inefficient routes.
The model, furthermore, provides the municipal authorities
with an effective decision support tool to inform optimal decisions
7. Conclusion
on fleet size, fleet allocation, capacity allocation and vehicle routes
across the entire ISWM system. An effective implementation of the
This study developed the Fleet Size and Mix VRP to optimize the
model will help develop a sustainable yet cost-effective waste
ISWM consisting of all the system components and to model the
management system. Municipal authorities will be able to identify
entire chain of waste/residue management (i.e. transfer stations,
the best and least costly sites for facilities; ascertain the optimal
treatment-, recycling- and disposal) with multiple waste
H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1391

Fig. 9. Comparison of Preemptive and non-dominated solution for HLO.

Table 5
Solutions by WGP.

Solution No. Weights (u1,u2) F1(x):Total cost ($M) F2(x): Total deviation (104) Pareto Optimality

S13 (0.1, 0.9) 2.705628281 7.79236 Inefficient


S14 (0.3, 0.7) 2.705508867 7.275237 Weakly efficient
S15 (0.4, 0.6) 2.705391734 20 Efficient (non-dominated Pareto)
S16 (0.7, 0.3) 2.705429742 7.817615 Efficient (non-dominated Pareto)
S17 (0.8, 0.2) 2.705457855 10 Inefficient
S18 (0.9, 0.1) 2.705503217 7.275257 Efficient (non-dominated Pareto)

Fig. 10. Comparison of Preemptive and non-dominated solution for WGP.

Table 6 minimize the total deviation of fair load allocation to transfer sta-
Optimal solution by sequential lexicographic method (S3): fleet size. tions to optimize the workload balance among them.
Depot Number of Vehicles The study concludes that the Goal Programming yielded ineffi-
cient solutions; whilst the Lexicographic approach achieved non-
V S U Total
dominated solutions but evaded producing inefficient solutions.
k1 2 3 2 7
On the other hand, the Lexicographic approach necessitates higher
k2 5 5 0 10
k3 1 4 0 5 computation requirement to implement a sequence of problems.
Total 8 12 2 22 The ISWM model was tested using the case study in the southern
part of Tehran, Iran. The model obtained the optimal: (i) number
and type of vehicles; (ii) vehicle allocation to transfer stations
processing/disposal technologies. Optimal capacity to the ISWM (depots); (iii) vehicles routes and (iv) capacity allocation to the
system components was allocated by concurrently optimizing the system components and technologies. The results show that single-
fleet size, vehicle routes and fare load allocation to transfer stations objective optimization optimized one of the objectives in compar-
through the development of a bi-objective optimization model. ison to the multi-objective optimization which has achieved the
The ISWM system was designed as a tri-echelon network and optimal solution with almost fare load allocation to transfer sta-
solved in the framework of the Fleet Size and Mix VRP with het- tions and minimized total transportation cost.
erogeneous fleet of vehicles and multiple depots. This was con- This developed model is methodological innovative as it con-
ducted by determining optimal fleet size, allocated capacity and siders the real-world constraints and complexities representing
vehicle routes in order to (i) minimize the total transportation cost diversified MSW types, varying use of waste-processing disposal
(fixed and variable cost) of waste/residue management; (ii) technologies, waste-technology compatibility, heterogeneous fleet,
waste-waste compatibility, and facility and vehicle capacity
1392 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

Table 7
Optimal solution by sequential lexicographic method (S3): routes.

Vehicle Origin Depot Load type Route Route (final) load

Technology V
v01 k1 o1 k1 /g4 /g1 /k1 60.64
v03 k2 o1 k2 /g5 /g3 /k2 60.52
v04 k1 o2 k1 /g1 /k1 42.38
v05 k2 o2 k2 /g5 /k2 38.63
v14 k2 o2 k2 /g3 /g4 /k2 53.63
v18 k2 o3 k2 /g4 /g5 /g1 /g3 /g2 /k2 19.34
v19 k3 o1 k3 /g2 /k3 52.88
v20 k2 o2 k2 /g2 /k2 58.75
Technology S
s01 k2 w1 k2 /l1 /k2 57.717
s04 k2 w2 k2 /k3 /l2 /k2 14.187
s07 k2 w3 k2 /h1 /k2 69.621
s09 k3 w3 k3 /k1 /h1 /k3 46.77
s11 k3 w4 k3 /k2 /h2 /k3 70.937
s12 k1 w4 k1 /h2 /k1 25.755
s13 k3 w5 k3 /m1 /k3 5.288
s14 k2 w5 k2 /k1 /m1 /k2 33.389
s17 k2 w6 k2 /k1 /m2 /k2 16.694
s18 k3 w6 k3 /m2 /k3 2.644
s19 k1 w1 k1 /k3 /l1 /k1 38.975
s20 k1 w2 k1 /l2 /k1 5.151
Technology U
u01 k1 w1 k1 /h1 /h2 /m1 /m2 /l1 /k1 40.611
u11 k1 w2 k1 /h1 /m1 /m2 /h2 /l2 /k1 25.14

limitations. The ISWM model also provides decision support tools revolving around reduction, recovery, reuse, and recycling should
to help local councils to identify options to minimize waste and be deployed to increase the resource efficiency of waste processing
maximize product output. In addition, the performance of waste and disposal and a series of actions implemented to target the zero
treatment technologies is evaluated which will provide bench- waste production as a long-term goal.
marking economic viability against sustainable waste management The scope of the ISWM model will be further extended to
goals and best practices to achieve the cleaner production strategy. incorporate environmental indicators such as carbon emissions
The results from this study will help the Tehran City Council in across different decoupling nodes of the waste/residue manage-
achieving the strategic goals by providing empirical evidence to ment cycle (i.e. transfer-treatment-recycling-disposal) with multi-
support planning toward proper and Healthy collection of wastes ple waste processing/disposal technologies. Future research will
from production origins, quick and immune transportation of them improve the developed ISWM model by adopting heuristic
to the waste processing and disposal, making use of private sector methods to generate solution approaches for larger cases where
capacities. The optimal fleet size, allocated capacity and vehicle computation requirements by exact methods are less stringent.
routes will enable the local government to reduce transport and Uncertainty associated with variables such as amounts of waste
disposal costs of waste and improve better utilization of transfer generation could also be integrated to enhance model's reliability
stations and landfill sites. The cost variability computed in this and applicability. Lastly, relaxing constrained assumptions in the
study for trucks, semi-trailers and residue transferring trucks in- model (e.g. partial waste collection) could enhance the effective-
forms the municipal government or private operators to economize ness of the developed model in dealing with multiple scenarios.
the ISWM system. This will allow the industry to reduce transfer
costs, improve the service levels while operating on a full cost re-
covery principle. This saving from better utilization of assets can Appendix A. Mathematical model e Nomenclature
then potentially reduce the household based garbage collection
charge, which is found to be a financial burden for deprived com- The problem is defined on a graph F¼(N,A) where N is the set of
munities in Iran. However, a range of waste reduction strategies n nodes in the network and A is the arc set as A ¼ ði;jÞ : 0  i, j  n,
isj.

Sets and indices


g2 G waste generation nodes
k2 K transfer station nodes
m2 treatment center nodes
M
h2 H recycling center nodes
l2 L disposal center nodes
q2 Q fleet of vehicles \ Q ¼ V ∪S∪U vehicles technologies
v2 V fleet of collection vehicles
s2 S fleet of semi-trailer transportation vehicles
u2 U fleet of semi-trailer residue transferring vehicles
o2 O waste types accumulated at generation nodes
waste types sorted at transfer station nodes
H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1393

(continued )

w2
W
w2 q residue types generated at treatment and recycling centers
t2 T treatment technologies
r2 R recycling technologies
d2 D disposal technologies

Parameters
fq fixed cost of vehicle q2Q
bo;i amount of waste type o2O accumulated at generation node i2G
cqi;j transportation cost of one unit waste/residue between nodes i and j by vehicle q2Q
no;q 1 if waste type o2O is compatible with collection vehicle q 2V2Q , 0 otherwise
n0w;q 1 if waste type w2W is compatible with transportation vehicle q 2S2Q
n00w;q 1 if residue type w2q is compatible with residue transferring vehicle q 2U2Q
aM
t;i
1 if treatment technology t2T exists at treatment center i2M; or 0 otherwise
aH
r;i
1 if recycling technology r2R exists at recycling center i2H; or 0 otherwise
aLd;i 1 if disposal technology d2D exists at disposal center i2L; or 0 otherwise
li;j traveling distance between nodes i and j2N
g1w;t 1 if waste type w2W is compatible with treatment technology t2T; or 0 otherwise
g2w;r 1 if waste type w2W is compatible with recycling technology r2R; or 0 otherwise
g3w;d 1 if waste type w2W is compatible with disposal technology d2D; or 0 otherwise
g4w;d 1 if residue type w2q is compatible with disposal technology d2D; or 0 otherwise
jqo maximum allowable load for vehicle q which is compatible with waste type o
jqw maximum allowable load for vehicle q which is compatible with waste type w
jqw maximum allowable load for vehicle q which is compatible with residue type w
aw;i ratio of sorting waste type w2W at transfer station i2K
bw;r proportion of total recycling of waste type w2W by recycling technology r2R
zw;t ratio of mass reduction of waste type w2W after treatment by treatment technology t2T
mMw
proportion of generated residues of a treatment center which is type of w2q
mHw proportion of generated residues of a recycling center which is type of w2q
‫א‬K number of transfer stations
pKi capacity of transfer station i2K
pM capacity of treatment center i2M
i
pHi capacity of recycling center i2H
pLi capacity of disposal center i2L

Decision variables
q
xi;j 1 if node i visited just after node j by vehicle 2Q ; or 0 otherwise
pw;i amount of waste type w2W sorted at transfer station i2K
ew;i amount of residue type w2q generated at treatment/recycling center i 2M∪H
zq 1 if vehicle q2Q is used; or 0 otherwise
Z qi 1 if vehicle q2Q is used and allocated to transfer station i2K; or 0 otherwise
yKi;o amount of waste type o2O processed at transfer station i2K
yM
i;w
amount of waste type w2W processed at treatment center i2M
yH
i;w
amount of waste type w2W recycled at recycling center i2H
y1L amount of waste type w2W disposed at disposal center i2L
i;w
y2L amount of residue type w2q disposed at disposal center i2L
i;w
cMi
total amount of residue produced at treatment center i2M
cHi total amount of residue produced at recycling center i2H
cML
i;j
amount of residue of treatment center i2M to send to disposal center j2L
cHL
i;j
amount of residues of recycling center i2H to send to disposal center j2L
uqi load of vehicle q2Q after visiting node i

Appendix B. Mathematical model e Linearization of


nonlinear terms xuqi;j  uqi ci2G; cj2G∪K; cq2V2Q (B.2)

 
xui;j  ui  1  xi;j M0
q q q
ci2G; cj2G∪K; cq2V2Q
Linearization I. : Linearization of xqi;j *uqi for the first echelon
(B.3)
Converting xqi;j uqi to xuqi;j in Equation (41) and objective function;
adding the following constraints to the model (Rabbani et al., 2018).
M0 > ∞
q
xui;j  0; ci2G; cj2G∪K; cq2V2Q (B.4)

xui;j  M 0 xi;j
q q
ci2G; cj2G∪K; cq2V2Q (B.1)
q q
Linearization II. : Linearization of xi;j ui for the second echelon
1394 H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395

q q q
Converting xi;j ui to xui;j in Equations (42)e(44) and objective
xei;j;w  M 0 xi;j
q q
function; adding the following constraints to the model. ci2K; cj2M∪H; cq2U2Q ; cw2q
(B.17)
xui;j  M0 xi;j
q q
ci2K; cj2K∪M∪H∪L; cq2S2Q (B.5)
q
xei;j;w  ew;j ci2K; cj2M∪H; cq2U2Q ; cw2q
q q
xui;j  ui ci2K; cj2K∪M∪H∪L; cq2S2Q (B.6) (B.18)

 xeqi;j;w  ew;j  1
q q
xui;j  ui  1 
 xi;j M 0 ci2K;
q
 cj2M∪H; cq2U2Q ; cw2q
 xqi;j M0 ci2K; cj2K∪M∪H∪L; cq2S2Q (B.7)
(B.19)

xei;j;w 0; M 0 > ∞ ci2K; cj2M∪H; cq2U2Q ; cw2q


q
M0 > ∞
q
xui;j  0; ci2K; cj2K∪M∪H∪L; cq2S2Q
(B.8) (B.20)

q q
Linearization III. Linearization of xi;j *ui for the third echelon Linearization VI. : Linearization of Absolute function in Equation
q q q (2): second objective
Converting xi;j ui to xui;j in Equation (45) and objective function;
adding the following constraints to the model. Due to limit on utilizing the ABS function in linear program-
ming, the second objective function is redefined as below.
xui;j  M0 xi;j
q q
ci2M∪H; cj2M∪H∪L; cq2U2Q (B.9) ,
X
pKj pKi ¼ sEi : Expected load at transfer station i2K
i2K
q q
xui;j  ui ci2M∪H; 2dcj2M∪H∪L; cq2U2Q (B.10) (B.21)

 ,
q q X XX
xui;j  ui  1 yKj;o bo;i0 ¼ sAi : Actual load at transfer station i2K
 o2O o2Oi0 2G
 xqi;j M0 ci2M∪H; cj2M∪H∪L; cq2U2Q (B.11) (B.22)

xuqi;j  0; M0 > ∞ ci2M∪H; cj2M∪H∪L; cq2U2Q sAi  sEi ¼ sDi


(B.12) : Devition of fair load allocation at transfer station i2K
(B.23)
To form ABS (sD
i ), two new positive variables are defined as:
q
Linearization IV. Linearization of xi;j *pw;j
Converting xqi;j pw;j to xpqi;j;w in Equation (28) and adding the
sPlusDi  0 ci2K (B.24)
following constraints to the model.
sNegDi  0 ci2K (B.25)
xpqi;j;w  M 0 xqi;j ci2K; cj2K; cq2S2Q ; cw2W
where
(B.13)
sDi ¼ sPlusDi  sNegDi ci2K (B.26)
q  
xpi;j;w  pw;j ci2K; cj2K; cq2S2Q ; cw2W (B.14) Finally, sD 
i in Equation (2) is replaced by (sPlusi þ sNegi ) to
D D

form the
 second objective
 function in the form of MinimizeF2 ðxÞ ¼
P  D
j2K sPlusj þ sNegj .
D

xpqi;j;w  pw;j  1
 Appendix C. Pareto optimality definitions
 xi;j M0 ci2K;
q
cj2K; cq2S2Q ; cw2W (B.15)
While a single solution can represent the optimality in a single-
objective problem, in multi-objective optimization problems there
M0 > ∞
q
xpi;j;w  0; ci2K; cj2K; cq2S2Q ; cw2W is not a single-global solution to reflect optimality of a solution.
(B.16) Consequently, a set of points need to be determined which fit a
predetermined definition for an optimum. The Optimality concept
in multi-objective optimization is presented by Pareto Optimality
in a framework as below (Marler and Arora, 2004). A multi-
Linearization V. : Linearization of xqi;j ew;j .
objective optimization problem stated as minf ðxÞ ¼ ff1 ðxÞ; f2 ðxÞ; …;
Converting xqi;j ew;j to xeqi;j;w in Equation (31) and adding the fk ðxÞg s.t. x2X and have k (k  2) competing objective functions (f1 :
following constraints to the model. Rn /R) which are to be minimized simultaneously.
H. Asefi et al. / Journal of Cleaner Production 230 (2019) 1376e1395 1395

Definition 1. A decision vector, x 2X, is efficient (Pareto optimal) in solid waste management: a survey of strategic and tactical issues. Comput.
if there does not exist another vector x2X, xsx , such that f ðxÞ  Oper. Res. 44, 22e32.
Golden, B., Assad, A., Levy, L., Gheysens, F., 1984. The fleet size and mix vehicle
f ðx Þ, and fi ðxÞ  fi ðx Þ for at least one function. Hence, x 2 X is routing problem. Comput. Oper. Res. 11, 49e66.
efficient and f ðx Þ is non-dominated. Golden, B.L., Raghavan, S., Wasil, E.A., 2008. The Vehicle Routing Problem: Latest
Advances and New Challenges, vol. 43. Springer Science and Business Media.
Definition 2. A decision vector, x 2X, is weakly efficient (weakly Habibi, F., Asadi, E., Sadjadi, S.J., Barzinpour, F., 2017. A multi-objective robust
Pareto optimal) if there does not exist another vector, x2 X, xs x , optimization model for site-selection and capacity allocation of municipal solid
waste facilities: a case study in Tehran. J. Clean. Prod. 166, 816e834.
such that f ðxÞ  f ðx Þ. Hence, x 2X is weakly efficient and f ðx Þ is Harijani, A.M., Mansour, S., Karimi, B., Lee, C.-G., 2017. Multi-period sustainable and
weakly non-dominated. integrated recycling network for municipal solid wasteea case study in Tehran.
J. Clean. Prod. 151, 96e108.
Definition 3. A solution, x 2X, is supported if it is Pareto optimal Ignizio, J.P., 1976. Goal Programming and Extensions. Lexington Books.
and there exist some real positive numbers (weights) u1 ; u2 ; …; uk Ignizio, J.P., 1985. Introduction to Linear Goal Programming. Sage Publications.
Jabbarzadeh, A., Darbaniyan, F., Jabalameli, M.S., 2016. A multi-objective model for
so that the solution is optimal with respect to the linear combi-
P location of transfer stations: case study in waste management system of Tehran.
nation of (weighted sum problem): minf ki¼1 ui fi ðxÞg s.t. x2 X with J. Ind. Syst. Eng. 9, 109e125.
coefficients u1 ; u2 ; …; uk . Otherwise the solution is non-supported. Karsak, E.E., Sozer, S., Alptekin, S.E., 2003. Product planning in quality function
deployment using a combined analytic network process and goal programming
approach. Comput. Ind. Eng. 44, 171e190.
Kim, B.-I., Kim, S., Sahoo, S., 2006. Waste collection vehicle routing problem with
Appendix D. Supplementary data
time windows. Comput. Oper. Res. 33, 3624e3642.
Lin, C., Choy, K.L., Ho, G.T., Chung, S.H., Lam, H., 2014. Survey of green vehicle
Supplementary data to this article can be found online at routing problem: past and future trends. Expert Syst. Appl. 41, 1118e1138.
https://doi.org/10.1016/j.jclepro.2019.04.250. Marler, R.T., Arora, J.S., 2004. Survey of multi-objective optimization methods for
engineering. Struct. Multidiscip. Optim. 26, 369e395.
McDougall, F.R., White, P.R., Franke, M., Hindle, P., 2008. Integrated Solid Waste
References Management: a Life Cycle Inventory. John Wiley and Sons.
Ocampo-Martinez, C., Ingimundarson, A., Puig, V., Quevedo, J., 2008. Objective
Abduli, M., Akbarpour Shirazi, M., Omidvar, B., Samieifard, R., 2015. A survey of prioritization using lexicographic minimizers for MPC of sewer networks. IEEE
municipal solid waste generation in 22 regions of Tehran with solid waste Trans. Control Syst. Technol. 16, 113e121.
reduction approach. Tolooebehdasht 14, 23e33. Rabbani, M., Heidari, R., Farrokhi-Asl, H., Rahimi, N., 2018. Using metaheuristic al-
Asefi, H., Lim, S., 2017a. A novel multi-dimensional modeling approach to integrated gorithms to solve a multi-objective industrial hazardous waste location-routing
municipal solid waste management. J. Clean. Prod. 166, 1131e1143. problem considering incompatible waste types. J. Clean. Prod. 170, 227e241.
Asefi, H., Lim, S., Maghrebi, M., 2015b. A proof-of-concept framework of municipal Repoussis, P.P., Tarantilis, C.D., 2010. Solving the fleet size and mix vehicle routing
solid waste location routing problem. In: Australasian Transport Research problem with time windows via adaptive memory programming. Transport.
Forum (ATRF), 37th, 2015 (Sydney, New South Wales, Australia). Res. C Emerg. Technol. 18, 695e712.
Aydemir-Karadag, A., 2018. A profit-oriented mathematical model for hazardous Samanlioglu, F., 2013. A multi-objective mathematical model for the industrial
waste locating-routing problem. J. Clean. Prod. 202, 213e225. hazardous waste location-routing problem. Eur. J. Oper. Res. 226, 332e340.
Badran, M., El-Haggar, S., 2006. Optimization of municipal solid waste management Santiban ~ ez-Aguilar, J.E., Ponce-Ortega, J.M., Gonza lez-Campos, J.B., Serna-
in port \SaideEgypt. Waste Manag. 26, 534e545. Gonz alez, M., El-Halwagi, M.M., 2013. Optimal planning for the sustainable
Bing, X., Bloemhof, J.M., Ramos, T.R.P., Barbosa-Povoa, A.P., Wong, C.Y., van der utilization of municipal solid waste. Waste Manag. 33, 2607e2622.
Vorst, J.G., 2016. Research challenges in municipal solid waste logistics man- Sharif, N.S., Pishvaee, M.S., Aliahmadi, A., Jabbarzadeh, A., 2018. A bi-level pro-
agement. Waste Manag. 48, 584e592. gramming approach to joint network design and pricing problem in the
Boyer, O., Sai Hong, T., Pedram, A., Mohd Yusuff, R.B., Zulkifli, N., 2013. municipal solid waste management system: a case study. Resour. Conserv.
A mathematical model for the industrial hazardous waste location-routing Recycl. 131, 17e40.
problem. J. Appl. Math. 2013, 435272 https://doi.org/10.1155/2013/435272, 10 Silva, S., Alçada-Almeida, L., Dias, L.C., 2017. Multiobjective programming for sizing
pages. and locating biogas plants: a model and an application in a region of Portugal.
Buhrkal, K., Larsen, A., Ropke, S., 2012. The waste collection vehicle routing problem Comput. Oper. Res. 83, 189e198.
with time windows in a city logistics context. Procedia-Soc. Behav. Sci. 39, Sun, C., Ritchie, S.G., Tsai, K., Jayakrishnan, R., 1999. Use of vehicle signature analysis
241e254. and lexicographic optimization for vehicle reidentification on freeways.
Chang, N.-B., Pires, A., Martinho, G., 2011. Empowering systems analysis for solid Transport. Res. C Emerg. Technol. 7, 167e185.

Taillard, E.D., 1999. A heuristic column generation method for the heterogeneous
waste management: challenges, trends, and perspectives. Crit. Rev. Environ. Sci.
Technol. 41, 1449e1530. fleet VRP. Oper. Res. 33, 1e14.
Dai, C., Li, Y., Huang, G., 2011. A two-stage support-vector-regression optimization Tan, S.T., Lee, C.T., Hashim, H., Ho, W.S., Lim, J.S., 2014. Optimal process network for
model for municipal solid waste managementea case study of Beijing, China. municipal solid waste management in Iskandar Malaysia. J. Clean. Prod. 71,
J. Environ. Manag. 92, 3023e3037. 48e58.
Dantzig, G.B., Ramser, J.H., 1959. The truck dispatching problem. Manag. Sci. 6, Vidovi c, M., Ratkovi
c, B., Bjeli
c, N., Popovi
c, D., 2016. A two-echelon location-routing
80e91. model for designing recycling logistics networks with profit: MILP and heuristic
Edalatpour, M., Mirzapour Al-e hashem, S., Karimi, B., Bahli, B., 2018. Investigation approach. Expert Syst. Appl. 51, 34e48.
on a novel sustainable model for waste management in megacities: a case study Xi, B., Su, J., Huang, G.H., Qin, X.-S., Jiang, Y., Huo, S., Ji, D., Yao, B., 2010. An integrated
in Tehran municipality. Sustain. Cities Soc. 36, 286e301. optimization approach and multi-criteria decision analysis for supporting the
Eiselt, H.A., 2007. Locating landfillsdoptimization vs. reality. Eur. J. Oper. Res. 179, waste-management system of the city of Beijing, China. Eng. Appl. Artif. Intell.
1040e1049. 23, 620e631.
Erkut, E., Karagiannidis, A., Perkoulidis, G., Tjandra, S.A., 2008. A multicriteria fa- Yadav, V., Karmakar, S., Dikshit, A., Vanjari, S., 2016. A feasibility study for the lo-
cility location model for municipal solid waste management in north Greece. cations of waste transfer stations in urban centers: a case study on the city of
Eur. J. Oper. Res. 187, 1402e1421. Nashik, India. J. Clean. Prod. 126, 191e205.
Ghiani, G., Lagana , D., Manni, E., Musmanno, R., Vigo, D., 2014. Operations research

S-ar putea să vă placă și