Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Journal of Cleaner Production 278 (2021) 123445

Contents lists available at ScienceDirect

Journal of Cleaner Production


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

Location of municipal waste containers: Trade-off between criteria



Vlastimír Nevrlý a, *, Radovan Somplak b, Veronika Smejkalova
 a, Toma
s Lipovský a,
a
Josef Jadrný
a  2896/2, 616 69, Brno, Czech
Institute of Process Engineering, Faculty of Mechanical Engineering, Brno University of Technology e VUT Brno, Technicka
Republic
b
Sustainable Process Integration Laboratory e SPIL, NETME Centre, Faculty of Mechanical Engineering, Brno University of Technology e VUT Brno,
Technicka  2896/2, 616 69, Brno, Czech Republic

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

Article history: There is a growing trend in waste management, which is associated with an increasing number of
Received 15 January 2020 separately collected waste fractions. This is supported by novel directives of EU, where Circular Economy
Received in revised form Package was issued. It presents waste separation and recycling goals, which encourages municipal ad-
21 July 2020
ministrators to support a suitable infrastructure of waste collection points. Such an infrastructure grid
Accepted 23 July 2020
should be designed efficiently. There is always a limited number of waste collection points caused by
^ as de
Handling Editor: Cecilia Maria Villas Bo economic reasons. The paper aims to propose a solution to the allocation of waste collection points. It
Almeida uses different parameters as the decision criteria. These criteria include the volume-weighted walking
distance, the number of collection points, the service time of collection vehicle and the total cost of
Keywords: purchasing containers. The problem is formulated through mixed-integer linear programming, and
Waste bin and container various models are defined as single and multi-objective. The approach is demonstrated in a case study
Waste collection point for a municipality in the Czech Republic for plastic waste. The comparison of different approaches is
Service time presented by a series of simulations for all defined models. The individual criteria can go against each
Walking distance
other in the objective function. The best results are proposed by the model, which minimises the
Multi-objective
deterioration from optimal values of all criteria. The purchase cost for plastic waste in tested municipality
Location problem
is 23% lower compared with next tested multi-objective model using min-max principle. The better
purchase cost is at the expense of walking distance. The difference in walking distance for the two
assessed models is 12% per capita. The proposed method serves to stakeholders and municipal admin-
istrators to make decisions about the number of containers to purchase and their allocation within an
analysed territory for new or already collected waste fractions.
© 2020 Elsevier Ltd. All rights reserved.

1. Introduction there is so-called zero waste goal. Source separation is highly


promoted (the waste is separated in three different bins located
People all over the world are encouraged to increase the rate of outside citizens households) and even fined when there is poor
waste separation (Karak et al., 2012) so that the recycling rates can waste separation practice. The city also focuses on reduction of
grow as well. These two terms are very often and repeatedly greenhouse gas emissions (Wilson et al., 2015). In some lower in-
misunderstood and confused. The increase in both rates is forced by come countries and developing countries, the important part of the
directives in the European Union (Directive (EU) 2018/851) and waste management is represented by informal recycling. For
support measures and campaigns in other countries or cities. In example, in Lusaka, Zambia, the informal sector is responsible for
Japan, for example, they have a law to increase recycling rates of more than 30% of waste collection (Godfrey et al., 2018). High rate is
waste from the production, distribution and consumption of food. also seen in Buenos Aires, Argentina, or some Indian, Brazilian and
The law defines the conditions and gives responsibility for food- Bolivian cities. The development and improvement of waste
related businesses (Jeff et al., 2017). In Seattle, Washington, U.S., collection services is crucial to implement and enforce the sus-
tainable waste management (Wilson et al., 2015).
Legislative changes are also reflected locally. The upcoming
* Corresponding author. amendment to the Act on waste includes goals given by the EU as
E-mail address: Vlastimir.Nevrly@vutbr.cz (V. Nevrlý). well as other supporting tools to promote waste separation. One of

https://doi.org/10.1016/j.jclepro.2020.123445
0959-6526/© 2020 Elsevier Ltd. All rights reserved.
2 V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445

them is the pay as you throw (PAYT) system, which is designed to The critical objective of the optimisation is primarily the eco-
motivate the population to reduce the production of mixed nomic aspect of the location and allocation tasks for treatment
municipal waste. PAYT strategies are becoming widely applied in facilities in waste management. Recently, an environmental crite-
solid waste management systems, especially in the US and in the rion has been addressed as well. An extensive review of these types
EU (Elia et al., 2015). Support for waste separation is not only the of problems was reported by Barbosa-Po voa et al. (2018). It was
domain of developed countries (Hemmati et al., 2019). Promoting concluded that sustainability was mainly tackled by assessing
waste separation in developing countries necessarily requires economic and environmental aspects, leaving behind the social
building new infrastructure. It is evident that waste management is aspect. Another case is the design of collection containers where
evolving worldwide, and building adequate infrastructure is the walking distance from waste producers generally is considered
essential for different stages of development. as a criterion. This fact shows that GIS-based (Geographic Infor-
Newly separated waste fractions are a direct consequence of mation System) computational tools are widely used (Lo  pez
strict legislative regulations in EU. Waste oils and fats are the next Alvarez et al., 2008). The respective article dealt with the collec-
separately collected items. There is an obligation to facilitate the tion route design for paper and cardboard waste from small busi-
possibility to separate such waste by municipalities from 2020 nesses, suggesting new locations and numbers of containers
(Amendment no. 210/2018 Coll.). Another example is textile, whose needed in streets for both commercial and non-commercial waste.
collection is going to be regulated by legislation from 2025 The distance to collection point was used as the criterion. Similar
(Directive (EU) 2018/851). An appropriate infrastructure of waste topic was discussed by Zamorano et al. (2009) who, in addition to
containers is a prerequisite to meet the set goals, as it significantly walking distance, also addressed the load of the containers. The
increases recycling rates (Knickmeyer, 2020). Nowadays, the trends article considered more municipal waste fractions with limited
in the waste collection are very often labelled as “smart”. Bong et al. maximum available distance using GIS-based infrastructure.
(2018) implemented such direction in agricultural waste manage- Despite the dynamic developments in the waste management
ment. Wireless sensor networks with smart bins were introduced sector, new articles are similar in elementary ideas. They focus
by Ramson and Moni (2017). These improvements relate to cost primarily on walking distance and use networks based on GIS
efficiency, which can be optimised on different levels of manage- software to design grids of collection points. A non-traditional
ment, mainly connected to the operation of the waste collection waste collection method was proposed by Hidalgo et al. (2018)
process. Collection routes influence the cost and also environ- and used vacuum systems that are suitable for hard-to-reach
mental impact (Fan et al., 2018) because the vehicle fleet is still areas. Due to the smaller number of large-capacity containers, the
most commonly internal combustion engine-based. The whole system could be easily optimised for placement and operation. The
waste collection activity generates a considerable amount of economic balance sheet of the vacuum system was presented and
emissions, and human labour is often neglected in this system (Di compared with traditional waste collection systems. Besides, two
Maria et al., 2019). When assessing the risks related to future case studies from Spain and Greece were published. In the case of
cleaner production, the role of a stakeholder is significant as well the Greek one, also a map with the locations of collection points
(Ngan et al., 2018). The introduced approach has shown positive was presented. Vacuum waste collection systems achieved high
results due to task integration. The suitability of the vehicle fleet is reductions in CO2 emissions (Hidalgo et al., 2018). However, they
also very important (Asefi et al., 2019), which was proven by the did not provide details about the mathematical models and metrics
case study from Tehran, where the fleet size is examined in order to used. Waste collection systems were evaluated by Rodrigues et al.
minimise waste collection costs. (2016a) for Portugal. The second part of the mentioned study
Knowledge of the current waste bin load may significantly in- focused more on container types and volumes (Rodrigues et al.,
crease the collection efficiency (Yusof et al., 2018). Extensive 2016b). Waste management systems with the highest volumes
research on location routing problems produced tools that can be and amounts of waste collected per time were selected as the most
used to optimise waste management systems (Prodhon and Prins, efficient.
2014). The combination of bin allocation and consequent routing Several other studies focused on the multi-scenario tasks.
was defined by Hemmelmayr et al. (2014). The locations of the Ferronato et al. (2020) considered the location of collection points
collection sites, their volumes, quantities and numbers, and for different municipal solid waste fractions (5 scenarios in total).
collection routes were determined by mixing exact and heuristic The criteria were again the walking distance with the maximum
approaches. It was found that the locations of the collection points allowed value using geographic data. A Bolivian developing city
and the routing were closely related. The grid of collection points was assessed, but no details regarding the utilised mathematical
plays an important role in waste management and mathematical model were provided. Ferronato et al. concluded that the integra-
programming can be used to find the optimal placement of waste tion of informal waste pickers can increase recycling rate and
containers over a given locality. The overall cost of waste collection reduce waste management expenses. Other scenarios were used by
is influenced by tourism (Greco et al., 2018). This was demonstrated Gallardo et al. (2015). Different types of waste repositories defined
by studies of 68 municipalities in Italy. This aspect should be the total number of collection points using GIS (5 scenarios in total).
considered when planning the waste collection or the container In addition, the uneven production of municipal waste fractions
grid in the case of a town where tourism is non-negligible. during the year was considered.
Walking distances to the bins have a significant impact on the Morato et al. (2019) presented an interesting alternative to
quality of separation and needs to be addressed (Rousta et al., deploying waste repositories in Bolivia. A framework for locating
2015). The participation of households was measured in Sweden biomass collection points was developed using an iterative model
after two interventions: (a) shorter distance to the drop-off point in a GIS environment. The model used the spatial distribution of the
and (b) easy access to correct sorting information. The population biomass and the road network maps with preference to zones with
welfare point of view should also be reflected in optimal location high biomass availability and a short distance to the roads. The
tasks as it is closely related to the willingness of people to separate. density of biomass production played an important role. In addi-
A suitable waste container infrastructure must be implemented tion, territorial-type constraints (e.g. National Protected Areas or
with regards to the economic possibilities while trying to maximise lakes) were taken into account. The task also suggested 7 locations
separation rates. The next section concentrates more on such ap- for facilities for biomass energy utilisation and assignment of
proaches previously published or applied. biomass producers.
V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445 3

Until now, GIS applications have been used to select the loca-  decision on the collection frequency for the collection points;
tions of collection points. Shi et al. (2020) added the costs of  compliance with the minimum utilisation rates of collection
opening a collection point while considering the walking distance. points e total and individual.
It was formulated as a max-min problem, where the aim was to
maximise the amount of separated municipal waste while mini-
mising the costs associated with collection points. The results 1.2. Novelty and contribution
concerning electrical waste were supplemented by estimating the
Pareto Frontier (the resulting non-linear optimisation model was The issue of the optimal placement of containers has already
solved using a meta-heuristic algorithm). been addressed, as the introduction part has shown. Previously
Waste collection areas were introduced in the context of the presented models differ in individual approaches, nevertheless,
walking distance and the number of collection points by Bautista there is potential for further development in this area. The
et al. (2006). An interesting feature of this article was the calcula- following requirements AeC have been established for the pre-
tion based on the developed heuristic algorithm. sented model.
Vu et al. (2018) proposed a two-phase task, where in the first
step, the locations of waste collection points were decided on and 1.2.1. Holistic approach
in the second step, the collection plan was optimised. The criterion The issue of container or bin location is often GIS-based
for the first phase was only the walking distance. Scenarios for a (Boskovic and Jovicic, 2015), with optimisation using map data.
different number of collection containers and maximum distance Another approach is to optimise selected criteria both for the
were considered, but no mathematical model was proposed. A single-objective and multi-objective model. Hemmelmayr et al.
similar problem was solved with heuristic algorithms by Ghiani (2014) introduced model for bins location with the simultaneous
et al. (2014). An optimal location of collection points was previ- planning of collection routes, while the total cost was minimised.
ously formulated as a mixed integer linear programming (MILP) Rossit et al. (2020) created a multi-objective model taking into
model (Ghiani et al., 2012), while also the capacities of individual account installation cost, collection frequency and average walking
collectors were selected. In the context of daily waste production, distance. The model provided is a quality basis for decision-making
clustering aggregation has been used to define areas. The results in waste management. Minimisation of the multi-objective prob-
revealed savings compared to the actual situation in tens of lem presented by Rossit et al. (2020) worked based on results of
percent. single-objective optimisation considering individual criteria. Pos-
The ultimate extension of the task lies in the multi-criteria sibility for further development was found in the area of criteria
implementation. Tralh~ ao et al. (2010) considered the walking dis- links and their combination within the multi-objective formula-
tance in 3 different objective functions and the investment costs for tion. The reason is that some commonly used criteria move the
a collection point. For the purpose-built distance-related functions: objective function in the same direction, e.g. the number of
the first one minimises the average distance from the dwellings to collection points and installation cost. On the other hand, some
the respective multi-compartment container; the remaining two criteria are against each other, e.g. the number of collection points
objectives address the “pull” and “push” characteristics of the de- and walking distance. This paper (sec. 3.1e3.3), deals with criteria
cision problem, one by minimising the number of dwellings too characteristics, especially their interrelations. The influence of the
close to any container, and the other by minimising the number of compilation of the objective function (single, dual, multi) is
dwellings too far from the multi-compartment container. This described in the case study.
article was followed by Coutinho-Rodrigues et al. (2012), who used
an objective function including the investment costs associated 1.2.2. Real applicability
with the collection points and the aspect of satisfaction of the The task of bins location is applied when a new infrastructure
population (waste producers). needs to be created. Another motivation may be the modification of
the existing infrastructure, where it is appropriate to assess its ef-
1.1. Approach and objectives ficiency compared to the current system. Erfani et al. (2017) sug-
gested GIS-based approach for optimal bins location. Erfani et al.
The goal was to define dependencies among economic, social, (2017) report that the reduction of total travelled distances can
and practical points of view through a set of different criteria. Based be 53% in the comparison of the current system for the presented
on such evaluation, proper decisions and plans could be formu- case study in the City of Mashhad. It is appropriate to consider the
lated. The annual investment costs were assessed via the pur- specific characteristics of the modelled area. In the case of GIS
chasing price of waste containers. The optimisation task was models, it is an approach for setting a suitable location. Specific
designed to include separation rate, collection frequency and waste characteristics of waste producers with regard to specific waste
container types within constraints and objectives. It was defined as production are not considered (Paul and Krishna, 2017). The aim of
a multi-objective MILP problem. The following criteria were the new model in this paper is to take into account the individual
considered: production of waste at address points.

 walking distance e the link to the willingness of people to 1.2.3. Capacity utilization of bins
separate recyclable fractions of municipal waste; Models usually consider the capacity coverage of the whole
 number of collection points e related to the administration and modelled area (Aremu and Sule, 2012). Nevertheless, the distribu-
the cost of collection; tion of waste production across the territory is not uniform. The
 purchase price of containers e investments cost; amount of waste produced is affected mainly by population density,
 collection service duration e time aspect, cost related to the utilisation of bins is closely connected with point B. Real applica-
servicing of the collection points. bility. The bins capacity is considered in the model presented by
Rossit et al. (2020), and the waste production of individual pro-
The task comprised the following optimisation challenges: ducers is reflected.
In the point of view of requirements AeC mentioned above, the
 decision on the placements of the collection containers; most corresponding paper presented before is by Rossit et al.
4 V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445

(2020). Both, the model by Rossit et al. (2020) and newly presented area. In the presented approach, the definition was more precise;
model suggest a multi-objective approach with cost, social and the area was considered as the municipality and its address points
other criteria. The main contribution of the presented model were potential locations of the collection points. The optimality
compared with Rossit et al. (2020) is in the analysis of individual could be seen from many perspectives; thus, the problem was
objectives and their combination for the multi-objective problem. defined as a multi-objective MILP. These criteria could correspond
The largest part of the text is devoted to this issue (sec. 3.1e3.3). with the number of collection points/installed containers, walking
This paper follows the idea presented by Nevrlý et al. (2019). The distances, subsequent overall collection time, etc. An important
allocation of the container capacities was analysed with respect to parameter was also the utilisation rate of such installed capacities.
the total and individual utilisation rates of the collection points. The These conditions and objectives were formulated using the re-
even trade-off between the volume-weighted walking distances quirements of the respective municipality. The problem is sche-
and the numbers of collection points neglected all remaining sce- matically depicted in Fig. 2, where the two steps describe the pre-
narios and corresponding detailed analysis. The objective of the processing phase (waste production estimation, address point
mentioned paper (Nevrlý et al., 2019) was the optimal setting of the definition, distance calculation, and edge elimination) and
utilisation rates and the evaluation of the resulting computational container allocation optimisation.
complexity. These results could then be used in a more compre-
hensive task. The hypothesis was that a proper combination of
different criteria would enhance the performance of the individual 2.1. Parameters and pre-processing
collection points.
From the point of view of the mayor, the proposed approach Waste production was one of the critical parameters featured in
could be applied as the analysis of the possible modifications of the almost all the waste management tasks. It had to be estimated on
existing collection point and container infrastructure or the addi- different territorial levels from the location to the routing prob-
tion of new waste fractions. Due to a limited budget, it was lems. Regarding the task of container assignment, the data were
necessary to maximise the deployment efficiency, but not to exceed needed for the address points which were used as baseline units for
the capacity utilization of the vehicle fleet so that new collection the calculation. Waste production was estimated using the average
vehicles did not have to be purchased. An important aspect was the number of inhabitants at a specific address point. The weekly
walking distance, which was a key indicator of population production was assumed to be 8.79 kg/cap. The calculation was
satisfaction. carried out according to equation (1) for the analysed waste type,
The reviewed tasks concentrated solely on the operational where i represents the address points and pi the corresponding
points of view. However, the targeted future limits, possibility of waste productions.
separation, or combination of criteria were not adequately ana-  
populationi ,weekly production cap,L
lysed. A new contribution of the presented approach lay in the pi ¼ (1)
definition of compromise between the mentioned criteria. This was
bulk density week
reached through different model formulations while considering In practice, various average walking distances could be seen
certain constraints and previously analysed bounds. The applica- depending on waste types. The containers for mixed municipal
tion of multi-objective approach enabled the specification of the waste were located directly in or nearby the buildings. This
minimum and maximum values of the individual criteria. The depended on the dwelling type and density of the population at a
procedure followed the diagram in Fig. 1. The novel approach specific area. Waste fraction such as plastic, glass, and paper were
considered criteria as stated in sec. 1.1 while new constraints on the often situated at the same collection point. This provided a certain
possible collection frequencies was addressed as well. guide for collection points’ location e connection of some fractions
The calculations resulted in the definition of the trade-off for could be defined. For example, the results pertaining to the most
various combinations which lead us to suggest socially and often collected waste fractions could be used for others since a
economically acceptable decisions. Multiple criteria were consid- stand-alone container for separately collected fractions was unde-
ered to show the problem from multiple angles and allow the sirable. This idea built on the use of an iterative calculation
possibility of managerial decisions. Two benefits were defined, one approach.
showing the standard minimisation of the appropriately weighted Price analysis for different types of containers was carried out
sum of all criteria, and the other (which was not published) (volume-based). Also, the average lifespan of the collection con-
showing the deterioration from optimal values in all criteria on the tainers and their average collection service time were estimated
smallest possible level. from the data obtained from the servicing companies, see Table 1.
Containers of other volumes were also used in practice, but they
2. Methods and problem description were waste type-specific. It is clear from the comparison of annual
costs that the use of large containers (1100 L) is disadvantageous
In general, the discussed problem can be defined as an optimal from the point of view of purchase costs, while 240 L containers can
assignment of the waste collection containers within a particular be up to twice as cheap. Another critical parameter was the bulk

Fig. 1. The steps of the proposed procedure.


V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445 5

Fig. 2. Two-step illustration of the presented approach.

Table 1
Main characteristics of collection containers.

Capacity [L] Purchase cost [EUR] Lifespan [y] Collection service time [s] Annual cost [EUR/a] Unit annual cost [EUR/(a,m3)]

120 35 8 35 4.375 36.46


240 47 8 35 5.875 24.48
1100 266 5 90 53.2 48.36

density of waste. It differed among municipalities, local areas, and 1) Volume-weighted walking distance e the distance between an
dwelling types. It could be determined based on the number of address point and a candidate container location multiples by
inhabitants living in buildings. In the presented task, a simplifica- the amount of transported waste.
tion was introduced in the form of the bulk density being inde- 2) The number of the allocated collection points e significantly
pendent of dwelling and container type. influences the total number of containers.
For a further case study, plastic waste was selected for the 3) The total purchase cost of the entire collection grid e it was
demonstration purposes and the bulk density 60 kg/m3 is assumed defined by the purchasing cost of the containers and normalised
inside the container. From the perspective of separated waste types, per year.
this was the most discussed fraction of municipal solid waste, 4) Collection service duration e the total of the collection times for
which was normally separated in larger quantities such as glass, the different container types.
paper, or metal. Plastics types vary widely, and it could be expected
that the efficiency of the system would be increased by allocating These criteria were further combined in several ways to identify
specific types of plastics for sorting, such as PET. For the other their mutual dependencies. The multi-objective approaches that
fractions, the current management of separated waste was signif- were used followed the principles of the weighted-sum method
icantly simpler. (scalarization method) and the ε-constraint method. All the sym-
To allow various capacities of the collection points, basic bols used in the following equations and constraints are described
container volumes from Table 2 were combined. The following in Table 3.
capacities were considered for all potential address points: 0 L, Equation (2) defines the total volume-weighted walking dis-
120 L, 240 L, 480 L, 720 L, 1100 L, and 2200 L. This also corresponded tance, which was calculated between the address points and the
to the setting of the collection frequency. It could be influenced by candidate collection points. Equation (3) defines the number of
the waste production at the respective collection point or via a allocated collection points. Equation (4) defines the purchase cost
maximum number of containers therein. In the presented of the intended collection points or containers that were normal-
approach, it was given by the population density (as listed in ised by the corresponding lifetime periods. Equation (5) defines the
Table 2) based on the typical threshold for the definition of the total collection service time per the allocated containers. Equation
dwelling types. (6) describes the weighted sum method for two chosen criteria m
and n. Constraint (7) limits the relative change from the optimal
2.2. Definition of equations value for all criteria. Equation (8) defines the overall relative change
from the optimal value for all criteria.
Two different approaches of multi-objective optimisation were X
used in the development of the models. Criteria that were z1 ¼ d j xj (2)
compared and limited by some values were formulated according j2J
to the following definitions.
XX
z2 ¼ hi;k qk (3)
Table 2 i2I k2K
Frequency of collection based on the population at address points.
XX
Number of inhabitants Considered collection frequencies z3 ¼ hi;k ri;k (4)
0e8 once every 4 weeks once every 2 weeks i2I k2K
8e173 once every 2 weeks once a week
6 V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445

Table 3
The sets, parameters, and variables explained.

Type Symbol Description

Sets i2 I Nodes corresponding to all address and collection points


j2 J Edges defining connected nodes
k2 K Set of the possible collection point capacities and their frequency of collection
m; n2 Set of criteria {1 ¼ volume weighted walking distance, 2 ¼ number of collection points, 3 ¼ purchase cost, 4 ¼ service time}
S
Parameters pi Waste productions at the address points [L/week]
Ai;j Incidence matrix defining relations between the nodes e calculated in the pre-processing phase from a fully connected matrix
Bi;j Submatrix of Ai;j where the values 1 were omitted
dj Edge lengths [m]
ci;k Matrix defining the possible capacity and frequency values for each node i [L]
ri;k The annual costs for the candidate locations and their capacities [EUR/a]
gi;k Collection service times for the candidate locations and their capacities [s]
fi;k The collection frequencies for the candidate locations and their capacities
UT Minimum required total utilisation rate
UiI Minimum required individual utilisation rates for each node i
qk Auxiliary parameter which serves for counting collection points.
wi An auxiliary binary parameter restricting the set of the potential container locations
a Criteria weighting parameter
Variables xj Waste flow through edge j [L]
ti Waste accumulated at node i [L]
hi;k Binary variables indicating the installation of a specific capacity and container collection frequency at node i
z1 Volume weighted walking distance [km,L/week]
z2 Number of the collection points
z3 The purchase cost of the collection containers [EUR/a]
z4 Collection service duration for the containers [s]
z5 Convex combination of two criteria weighted by a and normalised by optimal, single-criterion values e represents a relative change from optimum
z6 Limit values for the relative changes from the optimum of all criteria
z7 Total relative change from the optimal, single-criteria values
Scalar z*m The optimal value of a single-objective model

X
XX ai;j xj þ pi  ti ¼ 0 ci2I; (9)
z4 ¼ hi;k gi;k fi;k (5) j2J
i2I k2K
X
ti  ci;k fi;k hi;k wi ci2I; (10)
k2K
 
z5 ¼ a , zm z* þ ð1  aÞ,zn z* (6)
m n XX X
ci;k fi;k hi;k U T  ti ; (11)
i2I k2K i2I

z6  zm z*  1 cm2S; (7) X
m ti  ci;k fi;k hi;k UiI ci2I; (12)
k2K

X 
 X
z7 ¼ zm z*  1 (8) pi  bi;j xj ci2I; (13)
m j2J
m2S

To meet the boundary conditions, the following constraints X


were needed. Network flow balance was defined by Equation (9). It hi;k ¼ 1 ci2I; (14)
k2K
connected the waste production, waste transfer, and waste accu-
mulation at specific collection points. The volume of accumulated
waste had to be lesser or equal to the allocated capacity, as stated in hi;k 2 f0; 1g ci2I; ck2K; (15)
Equation (10). The total and individual utilisation rates were
bounded by the given values as shown by Equations (11) and (12). xj  0 cj2J; (16)
The outflow from each node was restricted to the waste production
at this address point by Equation (13). Such restriction was needed ti  0 ci2I: (17)
due to the formulation of the model. This flow model construction
was chosen to simplify the task and ensured its solvability in a
reasonable wall-clock time. Equations (14) and (15) ensured that
only one capacity was selected for each node (please note that the 2.3. MILP formulation of models
set of container capacities also included zero capacity). Constraints
(16) and (17) governed the non-negativity of continuous variables The criteria, equations, and constraints described in the previ-
e waste transfers between the address points and the collection ous section were used to formulate several models with different
points, and the volumes of the accumulated waste. objective functions. These included models for a single-objective
V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445 7

and three approaches for multi-objective formulation.


Model I

minzm
s.t.
Equations (2)e(5), (9)e(17).
The first model was defined as a single-criterion one for the
minimisation of multiple tasks. These were calculated in order to
obtain optimal values for normalisation, z*1 ; z*2 ; z*3 ; and z*4 , that were
further used in multi-objective models.
Model II

minz5
s.t.
Equations (2)e(5), (6), (9)e(17).
The second model was a dual-criteria one. The main role was
played by equation (6), which used a convex combination of the
two standardised criteria. Depending on the choice of m and n, the
Fig. 3. Population distribution corresponding to the potential collection points.
problem could take six different forms (for m ¼ n, it would be a
one-criterion problem). The weighting parameter a determined the
preference of the criterion for minimisation (value 0.5 defines both
points. This was done to reduce the computational time, which was
criteria as equally significant). Values z*m and z*n were obtained by significant due to the type of the optimisation task (the algorithm
solving Model I with the corresponding criterion. utilised approx. 36 million binary variables). Only up to 20 closest
Model III candidates from each point were considered in the models, further
also reduced by the maximum walking distance of 200 m (Nevrlý
minz6
et al., 2019). These values stemmed from the previous research
s.t. on the instance size and computational time. The procedure
Equations (2)e(5), (7), (9)e(17). defined a network consisting of approx. 118,000 edges. The “as the
Other types of models were built as multi-objective ones and crow flies” walking distance was used. It was calculated by co-
comprised 4 criteria e Models III and IV. All criteria were consid- ordinates of all the network points. All mathematical models were
ered in these models. The goal was to limit the negative impact of a implemented in GAMS (GAMS Development Corporation, 2013;
multi-criteria task in Model III (i.e., the deterioration of the results General Algebraic Modeling System) and solved using the CPLEX
for one criterion due to efforts to minimise the others). This meant solver 12.6.2.0 (IBM) with the relative optimality gap of 10%.
that a compromise was sought between the individual criteria so Population distribution represented an essential role in this task
that no significant deterioration in comparison with the results (see the map of the analysed area in Fig. 3). The grid of potential
from the one-criterion task occurred. As a result, this deterioration collection points corresponded to the address points shown in the
was minimised for the criterion where deterioration was most map.
prominent.
Model IV
3.1. Single-criterion calculations
minz7
Model I was applied to minimise all the defined criteria, i.e.,
s.t. 1evolume-weighted walking distance, 2ethe number of collec-
Equations (2)e(5), (8), (9)e(17). tion points, 3epurchase cost, and 4eservice time. Specific criteria
Model IV minimised the sum of all deteriorations from the were able to reach different levels of accuracy. The volume-
optimal values in a relative manner. weighted walking distance was calculated using the default
setting, that is, via a dynamic search. The computation was very
3. Case study fast while relative optimality gap dropped below 3%. The branch-

The models and assumptions from section 2 were used in a case


study to analyse the plastic waste for the municipality of Ta bor in
the Czech Republic. It was assumed that Ta bor produced approxi-
mately 8.8 kg/cap annually with the waste density being 60 kg/m3
(based on data from the servicing companies). In Ta bor, there are
around 35,000 inhabitants, which correspond to approx. 6000
address points. The results presented earlier by Nevrlý et al. (2019)
were used in the proposed approach. The parameters of the total
and individual utilisation rates were thoroughly assessed and a
decision was made to consider 70% and 30%, respectively. The
parameter wi was set to 1 for all nodes, which meant that no
subsequent restriction was defined prior to the calculation.
The pre-processing algorithm, which was implemented in
Microsoft Excel using Visual Basic for Applications defined the set
of candidates for the assignment of containers for all the address Fig. 4. Interdependence of the four criteria.
8 V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445

Fig. 5. Distribution of container types for the four criteria.

Fig. 8. Pareto frontier e dependency of the number of collection points on the walking
distance.

and-cut algorithm was used for purchase cost and service time
calculations. These took longer and reached optimality gaps below
10%. Regarding the number of collection points, the calculation
took a long time and the solution was further updated by a special
algorithm.
Minimising in one direction can result in extremely high values
in other criteria as shown in Fig. 4 which compares the four criteria.
The relative change from optimality was calculated and displayed
on the vertical axis. A typical example was the walking distance. If it
was minimised, the inevitable consequence was a large number of
collection points. Conversely, if the number of collection points was
minimised, this resulted in an increase in the walking distance.
Fig. 6. The most often used collection frequency.
The variance of the criteria values was quite different. It could be
noted that with the minimisation of the number of collection
points, costs, or service time, the remaining two criteria increased a
little, but in all three cases, the criterion of volume-weighted
walking distance has grown more than 15 times. On the other
hand, the number of collection points increased more than eight
times when minimising walking distance. Annual costs and service
time increased only about four times. If the growth were not so
significant, this solution would have been satisfactory in all
respects.
Address points with a maximum of 8 inhabitants were pre-
dominant (there were 5236 of them, which accounts for 88%). Fig. 5
shows the distribution of the utilised collection container types.
The next graph (Fig. 6) defines the relation to the collection fre-
quency for different optimisation criteria with frequency selection.
Obviously, the frequency was very low if one cared about the
duration of collection.
Only 29 collection points comprised more than one container
for volume-weighted walking distance. More than 90% of single-
Fig. 7. Average walking distance for single-objective optimisation.
container capacities had a volume of 120 L. The situation was

Table 4
Summary of the results for all criteria.

Minimised criteria

1 2 3 4

Volume-weighted walking distance [km,L/week] 388 6742 6270 7150


Number of collection points 2827 351 581 534
Purchase cost [EUR/a] 12,892 9290 3481 14,879
Service time [s] 36,078 14,506 16,100 9805
Number of containers 2856 670 629 537
Average walking distance [m/cap] 29 93 89 98

Note: Values highlighted in bold represent the optimum for the respective criterion (z*1 ; z*2 ; z*3 ; z*4 ).
1 ¼ volume weighted walking distance, 2 ¼ number of collection points, 3 ¼ purchase cost, 4 ¼ service time.
V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445 9

Fig. 9. Relative changes of the criteria for different values of the weighting parameter.

walking distance was 29 m/cap while for the remaining criteria it


was around 90 m/cap. Residents were furthest from the collection
points when service time was minimised.
An overview of results is listed in Table 4, where specific values
of criteria are compared. It should be emphasized that in each case,
the goal was to minimise a single criterion. The single-container
capacity of the collection points was used for the single-objective
problems in most cases. This was also probably due to the ana-
lysed waste fraction (and its considered production). It can also be
seen from the table that the lowest number of installed collection
containers did not guarantee the lowest number of collection
points and vice versa. The annual purchase cost per capita was only
Fig. 10. The dependences of the criteria in combination with the volume-weighted around 0.1 EUR for criterion 3.
walking distance.

3.2. Dual-objective combinations


similar in the case of purchase cost for the single-container points,
but with a difference in the container volume, which most often This section discusses the results obtained using Model II for all
was 240 L. This was due to the annual unit investment per the combinations of the criteria. Their interdependencies were inves-
container lifespan. Regarding the collection points criterion, the tigated in a series of calculations using different values of a. This
quantity of containers was twice the number of points. The capacity parameter ranged from 0 to 1 with step 0.1. Optimal values from
of 720 L was widespread (68% of the cases). The weekly collection Table 4 were used for normalisation as described previously. There
was rarely used in this criterion. were six possible combinations of criteria but only selected results
The histogram in Fig. 7 shows the average walking distance per will be pointed out.
capita for various minimised criteria. It should be noted that when The Pareto frontier for criteria 1 and 2 and 1 and 3 is shown in
the collection point was installed directly at the address point, the Fig. 8. The dependency is presented for normalised values e
distance was zero. Additionally, 25 m were added to each distance volume-weighted walking distance per capita, the number of
between the address point and the collection point. This corre- collection points per capita, and the annual cost per capita. For a
sponded to the real collection point assigned directly to the denser network of points on the Pareto curve, it would be possible
dwelling. to carry out the calculation with a smaller step of a parameter (it
There was a clear difference in the average walking distances could also be changed adaptively). It is obvious that the de-
among minimised criteria. This fact was expected, as the weighted pendency on the volume-weighted walking distance had a similar
and average walking distances were closely related. The average shape for both compared criteria.

Fig. 11. Comparison of the dual-criteria combinations.


10 V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445

Table 5
Overview of results for the dual-criteria problems.

Criteria

1,2 1,3 1,4 2,3 2,4 3,4

Volume-weighted walking distance [km,L/week] 1067 659 579 6371 6923 6676
Number of collection points 1176 1787 2033 425 388 652
Purchase cost [EUR/a] 7933 8219 1129 3620 1288 4303
Service time [s] 25,803 34,204 25,108 15,917 9767 14,602
Number of containers 1391 1804 2040 644 439 706
Average walking distance [m/cap] 36 32 31 90 96 93

Note: 1 ¼ volume weighted walking distance, 2 ¼ number of collection points, 3 ¼ purchase cost, 4 ¼ service time.

Fig. 12. Comparison of the single- and the multi-objective approaches.

Fig. 14. Illustrative example of collection points’ location (purple points represent the
original locations). (For interpretation of the references to colour in this figure legend,
the reader is referred to the Web version of this article.)

clearly, is not much), while the average walking distance varied


from 36 m to 96 m. Please recall that the “as the crow flies” distance
was considered.
Fig. 13. The total numbers of containers and the corresponding investment costs.
3.3. Multi-objective optimisation

This section discusses and compares the multi-objective for-


A more telling way of representing the results is shown in Fig. 9. mulations of models. There were two models considering all
The triangles define the progression of volume-weighted walking criteria, Model III and Model IV. The first one followed the min-max
distance with the changing parameter a. The distance was very long principle, i.e., minimise the maximum variance from the optimal
when almost ignored compared to the optimal value of the single- values. The second model minimised the sum of deviations from
criterion optimisation. On the other hand, the number of collection the optimal values for individual criteria. The comparison of ana-
points (squares) changes in a relatively linear fashion and does not lysed approaches is shown in Fig. 12. When comparing the models’
reach such large values. Diagonal crosses illustrate the unweighted total relative changes in criteria, the lowest value was achieved
sum of the normalised criteria, while circles denote the weighted using Model IV because it was minimised in the objective function.
sum. It can be seen that the minimum value is reached for a equal to Model II (criteria 1, 2) is included because it showed the best per-
0.5. formance from dual-objective variants. The corresponding total
All combinations were compared with other criteria for the relative changes were equal to 643% (Model III), 586% (Model IV),
volume-weighted walking distance in Fig. 10. One can see that the and 691% (Model II).
number of collection points varied the most. It must also be noted From a practical point of view, Model III seemed to be the best
that the volume-weighted walking distance varied for those points because it did not favour any criterion at the expense of another.
and specific a. Compared to Model IV, it was only about 6% worse overall. Fig. 13
The graph in Fig. 11 summarises the results for all dual-criteria compares the investment costs and the total numbers of collec-
combinations with a being set to 0.5. The graph is followed by an tion containers suggested by the different approaches. The highest
overview of all the values in Table 5. It should be noted that the number corresponds to the single-criterion minimisation of the
average annual investment costs for the containers ranged from volume-weighted walking distance, which is around 2850. This
approx. 0.1 EUR per capita to approx. 0.3 EUR per capita (this, result made it clear that the constant pressure to reduce the
V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445 11

average walking distance leads to the overall unsustainability of the Because any such decision-making has a social impact on the citi-
waste management system. The unsustainability is defined through zens in the form of the sizes of the collection containers and the
the high expenditures e purchase cost and operational cost e collection frequencies, the presented tool could be used to ensure
demanding collection service by vehicles. The lowest number was that the containers are sufficient.
observed in the case of service time minimisation, which corre- Considering only the walking distance in the evaluation of the
sponded to the fact that each container means extra idle time for container grid leads to other important criteria being neglected. In
the service workers. All the criteria except for the walking distance terms of the Ta bor municipality analysed in the case study, there
were within reasonable limits. would have to be 2856 containers available (2827 collection points)
Regarding the total investment costs, the worst values were to minimise this criterion. Compared to the situation in 2018, there
provided by Model I minimising criteria 1 and 4. For criterion 1, it were 184 collection points for plastic waste in Tabor. The difference
was caused by the large number of 1100 L containers (these are between collection points suggestions and reality is enormous, but
investment-ineffective). In the case of criterion 4, the significant government subsidies are sometimes provided to the municipal-
cost related to the number of deployed containers. Excluding Model ities in the Czech Republic to purchase new waste containers. A
I (criterion 3), where costs were minimised, Model III and IV per- more detailed analysis of the area captured in Fig. 14 shows a sig-
formed best. nificant increase in number of collection points, specifically from 3
Typically, 5 to 15 address points were assigned to each collection to 13. However, this change substantially reduces the walking dis-
point. This number, however, depended mainly on the dwelling tance. The original locations of the collection containers have a total
types in the area. An example of the proposed location of the volume of 3300 L (three large containers of 1100 L), compared to
collection points and their corresponding address points is shown the newly proposed volume of 3120 L (13 collection containers with
in Fig. 14. This shows that some address points used more collection a volume of 240 L). In terms of overall capacity, there has been
points, which corresponded to the reality when people choose almost no change. On the contrary, from the point of view of
where to take their waste with respect to the direction of travel average purchase costs per year, it is halved for the proposed so-
from the house or full capacity of some containers. Individual uti- lution in the given area (76 EUR/a compared to 160 EUR/a for
lisation rates of the respective collection points ranged from 67% to original deployment). From the point of view of waste collection,
100%. this is an approximately 60% increase in the distance travelled by
collection vehicles.
4. Results and discussion The lowest number of collection points obtained using the
proposed models was 351 with 670 containers in total. This resul-
Obviously, the volume-weighted walking distance and the ted in a very long average walking distance (93 m) while also other
number of collection points went against each other. When one of criteria were neglected. The multi-objective approaches presented
these criteria increased, the other decreased. The purchase cost much better results, which are highlighted below.
behaved similarly to the number of collection points, but with the It is worth discussing whether the investment costs play such an
difference that the collection containers with the lowest volume to important role when the total price is divided per capita and year.
purchase cost ratio were preferred. Taking into account the service Nevertheless, the presented results reflected just one waste fraction
time of the collection vehicles, the purchase cost went up sharply and, therefore, the actual cost would be much higher when calcu-
due to the large number of 1100 L collection containers. This then lated for all fractions together. Moreover, for example, the purchase
also required the smallest number of collection containers. Each prices of containers for glass waste are much higher than in the case
criterion had its advantages and disadvantages and so it was of plastic waste due to the required properties. Municipal budgets
important to consider all of them when deciding on the locations of are then further burdened with other related waste management
the collection points. Model III, which did not favour any of them, costs. One of the most expensive items is waste collection. These
has proved its strength. However, it generally is up to the mayors to cost the discussed municipality of Ta bor tens of thousands of EUR
decide what the specific objective is and what resources are per year. The collection is only indirectly captured in the form of
available (budget, vehicle fleet, employees) in their municipalities. service time as one of the minimised criteria. However, it is difficult
At present, the Czech collection network consists of over or even impossible to quantify this value directly in terms of money
413,000 publicly available containers for separate collection (in- and it serves as a comparative and indicative measure during the
cludes all waste fractions except from mixed municipal waste). On decision-making process. For decision-makers, results pertaining to
average, there are only 124 inhabitants per a single collection point, the best-performing approaches (Model II minimising criteria 1
which is a good result in comparison with other European coun- and 2, Model III and Model IV) are summarised in Table 6.
tries. In 2018, the average walking distance to separate waste Even though Model IV presented the smallest sum of relative
collection points was 91 m (EKO-KOM, 2018). These parameters changes from the optimal value, Model III did not produce anything
rank the Czech Republic among the European elite. Still, the values significantly larger (approximately 6%). Besides, Model III offered
should be further improved in order to reach the high separation an interesting alternative for municipalities that put great
and recycling targets, as specified among the goals of the recently emphasis on increasing the waste separation and streamlining of
issued EU directives. Since the average walking distance can be the waste management system. This was mainly due to the
calculated via different methods and the method used by EKO-KOM formulation of the objective function where the aim was not to
a.s. (the authorised packaging company taking care of the take-back worsen any criterion to the detriment of others. Only the walking
directives and waste recovery) is not publicly available, the pre- distance deteriorated with this approach (an increase of 7 m/cap
sented results could not be directly compared to the values re- meaning that Model III performs about 12% better). This was pri-
ported by EKO-KOM a.s. marily represented by the citizen’s view, while the other criteria
Using the presented model, it was possible to analyse the eco- corresponded primarily to technical services or the collection
nomic impacts in the case of a change of the current system. In companies, and asymmetry was suppressed. Compared to the
addition, it was appropriate to also include other aspects than just current state, the walking distance was reduced by 43 m/cap while
distance e the only measured parameter by EKO-KOM a.s. The the purchasing cost of all suggested containers reached 0.2 EUR/cap
number of collection points and the associated costs were impor- annually. Even though, the purchase cost is about 23% lower for
tant for the municipalities and the waste collection companies. Model IV, this increase in costs is negligible when calculated per
12 V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445

Table 6
Overview of the results for the multi-criteria problems.

Modelling approach

Model II (Criteria 1,2) Model III Model IV

Volume-weighted walking distance [km,L/week] 1067 1193 1537


Number of collection points 1176 1080 912
Purchase cost [EUR/a] 7933 7084 5432
Service time [s] 25,803 23,396 18,148
Number of containers 1391 1194 955
Average walking distance [m/cap] 36 48 55

Note: 1 ¼ volume weighted walking distance, 2 ¼ number of collection points.

capita. The cost of the already installed containers was not sub- could involve a more precise calculation of the walking distance so
tracted and, thus, the final cost would be much lower. that realistic distances are considered via pedestrian infrastructure
Source separation highly differs among countries around the (streets and footpaths). The difference could be compared, but this
world. It is caused by their economic situation. It can be observed should not influence the results significantly. The collection per
that the larger the country’s GDP, the more developed its waste capita rate could be modelled by some function in which the
management is and the more types of waste are also separated. amount of separated waste would depend on several parameters
While some states follow the path of common collection containers (e.g., the walking distance). Then the mathematical model would be
for multiple types of recyclable materials, others often have a reformulated. However, obtaining such a function is not easy as no
special container for each type of waste (Wilson et al., 2015). With data are available, and the model would also become non-linear.
the gradual development of waste management, it can be expected
that less developed countries will copy the historical development CRediT authorship contribution statement
of the advanced ones (e.g. Japan, Singapore, etc.) (Jeff et al., 2017). In
these countries, the presented approach has a great application, as Vlastimír Nevrlý: Visualization, Conceptualization, Validation,
it is possible to gradually model and adjust the network of collec- 
Methodology. Radovan Sompl k: Conceptualization, Supervision,
a
tion points with regard to the current yield per capita. At the same Methodology. Veronika Smejkalova : Visualization, Data curation,
time, it is possible to estimate the related costs in advance, both for Resources. Toma s Lipovský: Software, Conceptualization. Josef
the purchase and maintenance of collection containers and for the Jadrný: Writing - review & editing.
subsequent collection of waste (use of vehicle routing models).
Declaration of competing interest
5. Conclusion
The authors declare that they have no known competing
The presented model provides a tool for waste container opti- financial interests or personal relationships that could have
misation. The suitable location of waste collection points can appeared to influence the work reported in this paper.
significantly affect the waste collection (cost and servicing time)
and purchase cost of bins, as a case study for particular municipality Acknowledgement
Tabor shows in sec. 3. With the growing number of collection
points, the demands on waste collection increase due to the allo- The authors gratefully acknowledge the financial support pro-
cation of smaller bins. vided by ERDF within the research project No. CZ.02.1.01/0.0/0.0/
The municipality T abor currently has 184 collection points and 16_026/0008413 “Strategic Partnership for Environmental Tech-
about 230 containers for separated plastic. Ta bor is one of the nologies and Energy Production” and the project Sustainable Pro-
average cities in this context with a walking distance of about 91 m. cess Integration Laboratory e SPIL, funded as project No. CZ.02.1.01/
The compilation of a mathematical model is motivated by the 0.0/0.0/15_003/0000456, by Czech Republic Operational Pro-
improvement of the infrastructure for separated plastic with regard gramme Research and Development, Education, Priority 1:
to more criteria. In general, it is possible to find a compromise be- Strengthening capacity for quality research.
tween the presented results and the current state. Local government
can use a mathematical model with looser criteria to reflect its Nomenclature
economic situation and technical possibilities for collection. In gen-
eral, the tool can be applied to any fraction of municipal waste. A EU European Union
significant benefit is especially for factions that have undergone a GAMS General Algebraic Modeling System
legislative change for collection (Smejkalova  et al., 2020). Similarly, GIS Geographic Information System
the need to implement a collection infrastructure for textiles in EU MILP Mixed Integer Linear Programming
countries can be expected (see Directive (EU) 2018/851).
The real application represents a large task, which belongs to
References
the combinatorial problems. The solvability for the real munici-
pality can be limited by computation time in some cases. Given the Amendment no. 210/2018 Coll. Amending the Amendment No. 321/2014 Coll. On
independence of parts of cities that are far apart, it is possible to the Scope and Method of Ensuring a Separate Collection of Municipal Waste
Components, as Amended by Act No. 185/2001 Coll., on Waste and on
divide the whole modelled territory into individual clusters. Sub-
Amendments to Some Other Acts. In (Czech).
sequently, it is possible to solve individual clusters separately, Asefi, H., Shahparvari, S., Chhetri, P., Lim, S., 2019. Variable fleet size and mix VRP
which simplifies the task and calculates in real-time. with fleet heterogeneity in Integrated Solid Waste Management. J. Clean. Prod.
For further investigation of the best realisation plan, the task 230, 1376e1395. https://doi.org/10.1016/j.jclepro.2019.04.250.
Aremu, A.S., Sule, B.F., 2012. A case study evaluation of the impacts of optimised
could be further updated with other requirements. Based on the waste bin locations in a developing city. Civ. Eng. Environ. Syst. 29 (2), 137e146.
presented results, some criteria could be restricted. Future research https://doi.org/10.1080/10286608.2012.672411.
V. Nevrlý et al. / Journal of Cleaner Production 278 (2021) 123445 13

Barbosa-Po voa, A.P., da Silva, C., Carvalho, A., 2018. Opportunities and challenges in Nations Environment Program. Accessed 21.07.2020.
sustainable supply chain: an operations research perspective. Eur. J. Oper. Res. Karak, T., Bhagat, R.M., Bhattacharyya, P., 2012. Municipal solid waste generation,
268 (2), 399e431. https://doi.org/10.1016/j.ejor.2017.10.036. composition, and management: the world scenario. Crit. Rev. Environ. Sci.
Bautista, J., Pereira, J., 2006. Modeling the problem of locating collection areas for Technol. 42 (15), 1509e1630. https://doi.org/10.1080/10643389.2011.569871.
urban waste management. An application to the metropolitan area of Barce- Knickmeyer, D., 2020. Social factors influencing household waste separation: a
lona. Omega 34 (6), 617e629. https://doi.org/10.1016/j.omega.2005.01.013. literature review on good practices to improve the recycling performance of
Bong, C.P.C., Lim, L.Y., Lee, C.T., Fan, Y.V., Klemes, J.J., 2018. The role of smart waste urban areas. J. Clean. Prod. 245, 118605. https://doi.org/10.1016/
management in smart agriculture. Chem. Eng. Trans. 70, 937e942. j.jclepro.2019.118605.
Boskovic, G., Jovicic, N., 2015. Fast methodology to design the optimal collection pez Alvarez, J.V., Aguilar Larrucea, M., Fern
Lo andez-Carrio n Quero, S., Jimenez del
point locations and number of waste bins: a case study. Waste Manag. Res. 33 Valle, A., 2008. Optimizing the collection of used paper from small businesses
(12), 1094e1102. https://doi.org/10.1177/0734242X15607426. Cited 12 times. through GIS techniques: the Legane s case (Madrid, Spain). Waste Manag. 28 (2),
Coutinho-Rodrigues, J., Tralha ~o, L., Alçada-Almeida, L., 2012. A bi-objective 282e293. https://doi.org/10.1016/j.wasman.2007.02.036.
modeling approach applied to an urban semi-desirable facility location prob- Morato, T., Vaezi, M., Kumar, A., 2019. Developing a framework to optimally locate
lem. Eur. J. Oper. Res. 223 (1), 203e213. https://doi.org/10.1016/ biomass collection points to improve the biomass-based energy facilities
j.ejor.2012.05.037. locating procedure e a case study for Bolivia. Renew. Sustain. Energy Rev. 107,
Di Maria, F., Sisani, F., Contini, S., 2019. Contribution of human labor to emissions 183e199. https://doi.org/10.1016/j.rser.2019.03.004.
from waste collection. J. Clean. Prod. 231, 509e519. https://doi.org/10.1016/ 
Nevrlý, V., Sompl k, R., Khýr, L., Smejkalov
a a, V., Jadrný, J., 2019. Municipal solid
j.jclepro.2019.05.260. waste container location based on walking distance and distribution of popu-
Directive (EU) 2018/851 of the European parliament and of the council of 30 may lation. Chem. Eng. Trans. 76, 553e558. https://doi.org/10.3303/CET1976093.
2018 amending directive 2008/98/EC on waste, 2018, Off. J. L 150, 109e140. Ngan, S.L., How, B.S., Promentille, M.A.B., Yatim, P., Lam, H.L., 2018. Integrating
EKO-KOM, 2018. Annual summary. Available online at: www.ekokom.cz/uploads/ stakeholder’s role in mitigating risks for future cleaner production. Chem. Eng.
attachments/Obecne/Vyrocni_shrnuti_2018_EKO_KOM.pdf. accessed 17 April Trans. 70, 1327e1332. https://doi.org/10.3303/CET1870222.
2020 (in Czech). Paul, K., Krishna, A.P., 2017. Using GIS to locate waste bins: a case study on Kolkata
Elia, V., Gnoni, M.G., Tornese, F., 2015. Designing Pay-As-You-Throw schemes in City, India. J. Environ. Sci. Manag. 20 (1), 95e105.
municipal waste management services: a holistic approach. Waste Manag. 44, Prodhon, C., Prins, C., 2014. A survey of recent research on location-routing prob-
188e195. https://doi.org/10.1016/j.wasman.2015.07.040. lems. Eur. J. Oper. Res. 238 (1), 1e17. https://doi.org/10.1016/j.ejor.2014.01.005.
Erfani, S.M.H., Danesh, S., Karrabi, S.M., Shad, R., 2017. A novel approach to find and Ramson, S.R.J., Moni, D.J., 2017. Wireless sensor networks based smart bin. Comput.
optimize bin locations and collection routes using a geographic information Electr. Eng. 64, 337e353. https://doi.org/10.1016/j.compeleceng.2016.11.030.
system. Waste Manag. Res. 35 (7), 776e785. https://doi.org/10.1177/ Rodrigues, S., Martinho, G., Pires, A., 2016a. Waste collection systems. Part A: a
0734242X17706753. taxonomy. J. Clean. Prod. 113, 374e387. https://doi.org/10.1016/
Fan, Y.V., Klemes, J.J., Perry, S., Lee, C.T., 2018. An emissions analysis for environ- j.jclepro.2015.09.143.
mentally sustainable freight transportation modes: distance and capacity. Rodrigues, S., Martinho, G., Pires, A., 2016b. Waste collection systems. Part B:
Chem. Eng. Trans. 70, 505e510. https://doi.org/10.3303/CET1870085. benchmarking indicators. Benchmarking of the great Lisbon area, Portugal.
Ferronato, N., Preziosi, G., Portillo, M.A.G., Lizarazu, E.G.G., Torretta, V., 2020. J. Clean. Prod. 139, 230e241. https://doi.org/10.1016/j.jclepro.2016.07.146.
Assessment of municipal solid waste selective collection scenarios with Rousta, K., Bolton, K., Lundin, M., Dahle n, L., 2015. Quantitative assessment of dis-
geographic information systems in Bolivia. Waste Manag. 102, 919e931. https:// tance to collection point and improved sorting information on source separa-
doi.org/10.1016/j.wasman.2019.12.010. tion of household waste. Waste Manag. 40, 22e30. https://doi.org/10.1016/
Gallardo, A., Carlos, M., Peris, M., Colomer, F.J., 2015. Methodology to design a j.wasman.2015.03.005.
municipal solid waste pre-collection system. A case study. Waste Manag. 36, Rossit, D.G., Toutouh, J., Nesmachnow, S., 2020. Exact and heuristic approaches for
1e11. https://doi.org/10.1016/j.wasman.2014.11.008. multi-objective garbage accumulation points location in real scenarios. Man-
GAMS Development Corporation, Release 24.2.1, 2013. Washington, DC, USA. agement 105, 467e481. https://doi.org/10.1016/j.wasman.2020.02.016.
Ghiani, G., Lagana , D., Manni, E., Triki, C., 2012. Capacitated location of collection Shi, J., Wang, R., Chen, W., Xing, L., Jin, M., 2020. Bi-objective design of household E-
sites in an urban waste management system. Waste Manag. 32 (7), 1291e1296. waste collection with public advertising and competition from informal sectors.
https://doi.org/10.1016/j.wasman.2012.02.009. Waste Manag. 102, 65e75. https://doi.org/10.1016/j.wasman.2019.10.018.
Ghiani, G., Manni, A., Manni, E., Toraldo, M., 2014. The impact of an efficient Smejkalova 
, V., Sompl k, R., Nevrlý, V., Burcin, B., Ku
a cera, T., 2020. Trend forecasting
collection sites location on the zoning phase in municipal solid waste man- for waste generation with structural break. J. Clean. Prod. 266, 121814. https://
agement. Waste Manag. 34 (11), 1949e1956. https://doi.org/10.1016/ doi.org/10.1016/j.jclepro.2020.121814.
j.wasman.2014.05.026. Tralha ~o, L., Coutinho-Rodrigues, J., Alçada-Almeida, L., 2010. A multiobjective
Godfrey, L., Nahman, A., Yonli, A.H., 2018. Africa Waste Management Outlook. modeling approach to locate multi-compartment containers for urban-sorted
United Nations Environment Programme. Accessed 20.07.2020. waste. Waste Manag. 30 (12), 2418e2429. https://doi.org/10.1016/
Greco, G., Cenciarelli, V.G., Allegrini, M., 2018. Tourism’s impacts on the costs of j.wasman.2010.06.017.
municipal solid waste collection: evidence from Italy. J. Clean. Prod. 177, 62e68. Vu, H.L., Ng, K.T.W., Bolingbroke, D., 2018. Parameter interrelationships in a dual
https://doi.org/10.1016/j.jclepro.2017.12.179. phase GIS-based municipal solid waste collection model. Waste Manag. 78,
Hidalgo, D., Martín-Marroquín, J.M., Corona, F., Juaristi, J.L., 2018. Sustainable vac- 258e270. https://doi.org/10.1016/j.wasman.2018.05.050.
uum waste collection systems in areas of difficult access. Tunn. Undergr. Space Wilson, D.C., Rodic, L., Modak, P., 2015. Global Waste Management Outlook. ISWA
Technol. 81, 221e227. https://doi.org/10.1016/j.tust.2018.07.026. (International Solid Waste Association), United Nations Environment Pro-
Hemmati, S., Fataei, E., Imani, A.A., 2019. Effects of source separation education on gramme, Austria. Accessed 21.07.2020.
solid waste reduction in developing countries (a case study: Ardabil, Iran). Yusof, N.M., Zulkifli, M.F., Mohd Yusof, N.Y.A., Azman, A.A., 2018. Smart waste bin
J. Solid Waste Technol. Manag. 45 (3), 267e272. https://doi.org/10.5276/ with real-time monitoring system. Int. J. Eng. Technol. 7, 725e729. https://
JSWTM/2019.267. doi.org/10.14419/ijet.v7i2.29.14006.
Hemmelmayr, V., Doerner, K.F., Hartl, R.F., Vigo, D., 2014. Models and algorithms for Zamorano, M., Molero, E., Grindlay, A., Rodríguez, M.L., Hurtado, A., Calvo, F.J., 2009.
the integrated planning of bin allocation and vehicle routing in solid waste A planning scenario for the application of geographical information systems in
management. Transport. Sci. 48 (1), 103e120. https://doi.org/10.1287/ municipal waste collection: a case of Churriana de la Vega (Granada, Spain).
trsc.2013.0459. Resour. Conserv. Recycl. 54 (2), 123e133. https://doi.org/10.1016/
IBM. CPLEX solver. www.ibm.com/analytics/cplex-optimizer. Accessed 23.04.2020. j.resconrec.2009.07.001.
Jeff, S., Modak, P., Agamuthu, P., 2017. Asia Waste Management Outlook. United

You might also like