Article

Dynamic Multi-Compartment Vehicle Routing Problem for

Smart Waste Collection †
Yousra Bouleft 1, * and Ahmed Elhilali Alaoui 2

1 Modeling and Mathematical Structures Laboratory, Faculty of Science and Technology, Fez 2202, Morocco
2 Digital Engineering and Artificial Intelligence School, EuroMed University of Fes, Fez 2202, Morocco
* Correspondence: [email protected]
† This paper is an extended version of paper published in the international conference: 2022 IEEE 6th
International Conference on Logistics Operations Management (GOL), Strasbourg, France,
29 June–1 July 2022.

Abstract: The rapid increase in urbanization results in an increase in the volume of municipal solid
waste produced every day, causing overflow of the garbage cans and thus distorting the city’s
appearance; for this and environmental reasons, smart cities involve the use of modern technologies
for intelligent and efficient waste management. Smart bins in urban environments contain sensors
that measure the status of containers in real-time and trigger wireless alarms if the container reaches
a predetermined threshold, and then communicate the information to the operations center, which
then sends vehicles to collect the waste from the selected stations in order to collect a significant
waste amount and reduce transportation costs. In this article, we will address the issue of the
Dynamic Multi-Compartmental Vehicle Routing Problem (DM-CVRP) for selective and intelligent
waste collection. This problem is summarized as a linear mathematical programming model to define
optimal dynamic routes to minimize the total cost, which are the transportation costs and the penalty
costs caused by exceeding the bin capacity. The hybridized genetic algorithm (GA) is proposed to
solve this problem, and the effectiveness of the proposed approach is verified by extensive numerical
experiments on instances given by Valorsul, with some modifications to adapt these data to our
problem. Then we were able to ensure the effectiveness of our approach based on the results in the
static and dynamic cases, which are very encouraging.

Citation: Bouleft, Y.; Elhilali Alaoui, Keywords: mathematical programming; dynamic multi-compartment vehicle routing problem; smart
A. Dynamic Multi-Compartment waste collection; Information and communication technologies
Vehicle Routing Problem for Smart
Waste Collection. Appl. Syst. Innov.
2023, 6, 30. https://doi.org/
10.3390/asi6010030 1. Introduction
Academic Editor: Abdelkader Sbihi With growing urbanization, world cities face new challenges. They include traffic jams,
overcrowding, environmental problems, transportation problems, waste management, etc.
Received: 24 November 2022
Revised: 1 February 2023
These issues can be addressed these problems using intelligent networking capabilities.
Accepted: 10 February 2023
The solutions can provide the information needed to contribute to the prosperity of cities,
Published: 15 February 2023 so the “smart city” has exploded. In 2008, at the Council on Foreign Relations in New
York, IBM proposed the concept of a “smart city”; these cities depend on Information and
Communication Technologies (ICT) to improve the urbans’ services quality or reduce costs.
There are other terms for similar concepts, such as the connected city, electronic city, digital
Copyright: © 2023 by the authors. city, and electronic communities.
Licensee MDPI, Basel, Switzerland. A smart city is a well-performing and future-oriented city with different components,
This article is an open access article such as Smart Mobility, Smart Environment, Smart People, Smart Economy, Smart Waste
distributed under the terms and Management, etc. [1]. Smart cities work based on information and communication tech-
conditions of the Creative Commons nologies to find a link between citizens and technology to improve the quality of life and
Attribution (CC BY) license (https:// sustainability [2]. The service uses sensors to collect information and provide resources
creativecommons.org/licenses/by/
and efficiencies to facilitate the manageability of smart city services [3,4]. Most researchers
4.0/).

Appl. Syst. Innov. 2023, 6, 30. https://doi.org/10.3390/asi6010030 https://www.mdpi.com/journal/asi

Appl. Syst. Innov. 2023, 6, 30 2 of 26

define a smart city as a city using the concept of the Internet of Things (IoT) [5], noting that
it is also helpful in municipal solid waste management (MSWM) [6].
Information and Communication Technology (ICT) is a term used in the field of
telecommunications i.e., computer technologies, audio-visual media, multimedia, internet,
and telecommunication, allows users to communicate and explore information in real-time
to improve operational decisions-making. One of the critical areas to building smart cities
is the waste management area. In these cities, waste management explores in real-time the
waste size provided by sensors installed in the bins [7]. If container-filling information is
well documented, productivity will immediately improve, which will have a significant
impact not only on company profits but also on the community at large. This is due to the
fact that waste management is one of the challenges we face in our world today.
Nowadays, the municipality is still dealing with solid waste in the traditional way,
where the containers are unloaded one by one along a fixed route. These containers contain
different types of non-separated waste, and this system has many disadvantages, including
waste bins are not empty when complete [8], which leads to environmental pollution and
distortion of the city’s image. There are bins containing blank spaces, which can be supplied
with more waste, so they do not require emptying. The traditional model relies on the
mixed management of collected waste, which increases the difficulty of disposing of all
waste types.
Therefore, effective waste management relies on the classification of waste types
and the adoption of compartmentalized containers and means of transport, where each
compartment is assigned to a specific waste type [9]. This process will reduce the final
waste disposal cost and improve the overall disposal efficiency, as well as improve the
conservation of resources and protect the environment, and subsequently contribute to
economic, social, and environmental development. The smart city concept can also help
to solve the above problems, as waste collection uses the technology of information and
communication to determine which containers should be emptied by installing a sensor to
monitor the status of each compartment of the container and subsequently make effective
decisions. All these developments increase the complexity of modeling and the difficulty
of designing models.
In order to apply our problem in the actual case, real-time data, such as the waste
quantities of each type at each station, leads to an increase in the waste in the corresponding
bin. However, a compartment in a station can be visited more than once on the same
working day to avoid overflowing containers. In this article, we consider the dynamic
vehicle routing problem for separate waste collection in each smart bin, using information
and communication technologies in the concept of smart cities. To achieve this goal, we
propose a meta-heuristic method for efficient planning of separate waste collection dynamic
routes with the presence of a sensor at each compartment, which must be triggered once
the container has reached a predetermined maximum threshold. We compare the meta-
heuristic to an actual status given by a waste collection company.
There are several studies have treated the topic of waste management problems but
only a few studies thought about selective waste collection as well as waste collection
problems in smart cities using the smart bin. Waste sorting is important not only to
reduce the impact it has on the environment but also for health issues that can arise
from improperly disposed of waste and toxins. Waste sorting is also an economically
beneficial prospect because it makes recycling much easier. We search to combine both
problems in order to avoid environmental and social issues as well as minimize costs; this
method consists of selective waste collection using smart bins containing sensors indicating
the real status of filling. as other objectives, we ensure to avoid altered cities caused by
overflows bins.
We have organized the article as follows: In Section 2, we present a state-of-the-art
on smart waste management, selective waste collection, and the dynamic vehicle routing
problem. Section 3 details the proposed system model and defines the Dynamic Vehicle
Routing Problem of separate waste management using intelligent sensors. The formulation
Appl. Syst. Innov. 2023, 6, 30 3 of 26

of the mathematical model is presented in Section 4. Section 5 describes a real case study
and shows the results of the application of our approach. The conclusion of the article and
future directions are presented in Section 6.

2. Literature Review
The problem of smart waste collection appeared in research a few years ago as part
of the development of cities, what we call smart cities. Many researchers have proposed
models of waste collection using information and communication technologies.
In the literature, most of the authors researching smart waste management consider
that each container contains a sensor that measures waste level and triggers alarms, which
are communicated wirelessly to municipal agents to maximize the amount of waste col-
lected while minimizing costs. In this context, we present below various optimization
studies of separate and intelligent collections of which we are aware.
Pereira Ramos et al. (2018) [10] designed a waste collection system based on the
use of sensors that transmit information in real-time. These sensors are associated with
optimization procedures that provide information about optimal collection methods. The
work compares three different approaches, which define dynamic collection routes taking
into account the information provided by sensors installed inside containers: a limited
approach, an intelligent collection approach, and a more intelligent collection approach.
In the same year, the same authors studied the dynamic problem of waste collection in
the same context of smart bins. A Capacitated Vehicle Routing Problem (CVRP) has been
developed to maximize the amount of collected waste while minimizing the distance
traveled. The results show that using sensors inside containers improves the profit of waste
management companies. Álvaro Lozano et al. (2018) [11] proposed a monitoring and smart
waste management platform in urban areas. A model is developed to obtain the weight
measurements, filling volume, and waste container temperature. The forum contains a
model to improve waste collection methods; this model creates dynamic paths based on
the obtained data to save energy, time, therefore, costs. It also has a mobile application that
guides each driver to the best route.
Omara et al. (2018) [12] propose a linear programming model to find the optimal tra-
jectories to minimize costs or waits so that the trajectory aid works in real-time. The authors
proposed three heuristic solutions addressing the problem of efficient and intelligent waste
management. These heuristics are the closest vehicle first; a collection takes into account an
upper threshold, and supply takes into account the upper and lower thresholds.
In 2020, Lu et al. [13] introduced an intelligent ICT-based system to collect sorted
waste with the objective is to improve sustainability and vehicle workload balancing
and minimizing transportation and carbon tax costs. The article proposes a new hybrid
multi-objective algorithm based on genetic algorithms and the optimization of whales to
optimize the proposed system. In 2019, Hrabec et al. [14] presented the problem of smart
waste collection; the authors are developing a simple mathematical model, and they are
proposing/discussing a computational approach to solve the waste collection problem;
the article proposes two objectives, the primary aim is to optimize the dynamic planning
of the daily schedules of the collection vehicles to minimize transportation costs, and the
secondary objective is to make the optimal decision about the containers that are not entirely
filled, especially those that are in the path that connects the full containers; Theodoros
Anagnostopoulos et al. [15] introduced a dynamic architecture for waste collection; the
authors paid particular attention to advance services in priority areas, such as markets,
universities, hospitals, etc., where there is likely to be hazardous waste that negatively
affects human life and requires quick collection; they proposed new algorithms to provide
practical solutions to the problem of a dynamic and instantaneous waste collection, whose
objective is to reduce the time needed to serve priority areas.
Since our research model is an extension of the classic Dynamic Vehicle Routing
Problem (DVRP) model, we also present some literature reviews on DVRP. We cite Mes-
saoud et al. [16] studied the Vehicle Routing Problem (VRP) problem with a time window
Appl. Syst. Innov. 2023, 6, 30 4 of 26

to study the static case and the dynamic case in which the appearance of new customers
over time. The authors have proposed a resolution approach based on using the hybridized
Ant Colony Optimization Algorithm and the Large Neighborhood Search Algorithm in
the static case, and they have adopted this approach to solve the problem in a dynamic
context. In 2021 Ramachandranpillai et al. [17] proposed a model based on the neural
system with modified rules and learning in combination with Firefly optimization to solve
the combined version of Green VRP and Dynamic VRP with temporal windows, called
DGVRPTW. In the same year, Talouki et al. [18] studied the dynamic green vehicle routing
problem of perishable products under green traffic conditions to optimize the total cost of
the active transportation network, minimize environmental influences, and increase cus-
tomer satisfaction. The authors augmented an algorithm for a new augmented ε-constraint
heuristic approach; Moreover, robust optimization was carried out for the problem with
uncertainties on request and economic parameters. In 2017 Sreelekshmi et al. [19] used a
Modified Capacitated K-Means Algorithm and Variable Neighborhood Search to optimize
dynamic routes to manage solid waste collection efficiently. We find many works in the
literature based on their study of the DVRPV on the working day division into time slices.
Each time slice represents a static VRP; cite MESSAOUD et al. [16–20] (2013 and 2014) and
Bouziyane et al. [21] (2018). The discretization strategy was proposed for the first time by
Kilby et al. (1998) [22] and adopted by Montemanni al. (2002) [23].
In the literature, there are not many works concerning the MCVRP (Multi-Compartment
Vehicle Routing Problem) of separate waste collection, which consists in not mixing the
different types during transport. We cite the work of Bouleft and Elhilali Alaoui [24], who
proposed a new scheme for separate waste collection. The scheme contains three levels,
begin by transferring the different waste types to the separate transfer stations, where each
compartment corresponds to a specific waste type and can accommodate several supplies
of the same waste type, then transport these different types to the factories for treatment
and finally to the landfill. In this context, they presented mathematical formulations of
the waste management system using linear programs. The article adopted heuristics and
metaheuristics for resolutions. Reed et al. [8] (2014) studied the problem of transporting
two-compartment vehicles routing problem of recycling waste collection, and to solve
this problem, the authors proposed the Ant Colony System (ACS). And we cite the work
of Muyldermans and Pang (2010) [25], which tested on CVRP instances and MCVRP in-
stances to assess the benefits of using multi-compartment vehicles for waste collection.
Henke et al. [26] (2019) presented a problem of routing vehicles with several compartments
of different sizes for collecting several types of glass waste. They proposed a branch-and-
cut algorithm to compare the total cost of compartment sizes. Rabbani et al. [27] (2016)
combined a close-open mixed vehicle routing problem (COMVRP) with a multi-depot
vehicle routing problem (MDVRP) to formulate the multi-compartment vehicle waste col-
lection problem. And to solve it, they used the hybrid genetic algorithm. Oliveira et al. [28]
and Silva et al. [29] developed heuristics to solve the multi-compartment vehicle routing
problem to collect two types of recyclable materials (paper and plastic/metal). A real case
study is adopted to justify the effectiveness of the proposed approach. Bouleft and Elhilali
Alaoui (2022) [30] have proposed a mathematical model to study the problem of separate
waste collection in smart cities via compartmentalized vehicles. Using electric vehicles,
Yang, J et al. [31] (2022) introduced a Chance-Constrained Multi-Compartment Electric
Vehicle Routing Problem (CCMCEVRP) for separate and intelligent waste collection and
solved by Diversity Enhanced Particle Swarm Optimization with Neighborhood Search
and Simulated Annealing (DNSPSOSA). The authors proved that using electric vehicles is
more economical than using fuel vehicles and the saving rate is increased with the increase
of the vehicle compartment.
Dynamic vehicle routing models have rarely been applied to waste collection problems.
We cite three articles in which the authors take into account historical data (i.e., without
considering real-time information): Anghinolfi et al. [32], Abdallah et al. [33] (2019), and
Appl. Syst. Innov. 2023, 6, 30 5 of 26

Exposito-Marquez et al. [34] (2019). Anghinolfi et al. solved a dynamic optimization model
for creating the waste collection route.
As a reference, we note that there is only one definition regarding smart waste man-
agement; it is based on sensors installed inside bins to communicate information wirelessly
to collection decision-makers
The contributions of this paper are summarized as follows:
1. We propose a new scheme based on separate waste collection in the smart city through
compartmentalized vehicles. The objective of this problem is to minimize total trans-
port costs, including crossed distance and penalties caused by bin overflowing.
2. The problem is modeled as a dynamic vehicle routing problem, in which we divide the
day into periods of equal duration; in each time slice, we consider the requests known
at the beginning of the day and the new requests that have arrived and which can be
visited. Contrary to the classic dynamic vehicle routing problem, which considers
each visited customer must be removed from the list of customers to be visited for the
rest of the day, we assumed that if a waste type in a station has triggered the alarm
more than once in the same working day, this station can be visited more than once
on the same day in order to avoid the overflow of the bins.
3. The effectiveness of the proposed approach is validated with a used a real dataset
collected by Valorsul with some modifications to adapt these data to our problem.

3. Problem Descriptions
Our model proposes a separate and smart waste management problem that considers
real-time information on the filling level of fragmented bins; each part is dedicated to a
specific waste type to define dynamic routes for each compartmentalized vehicle. The
signal base station receives information on the amount of waste in each intelligent bin
at each station. It then communicates this information to the depot, which then sends
compartmentalized vehicles to collect the selected separate waste stations. When a new
station appears, the main task of the dispatching center is to include the new station in the
current routing plan. Scheduled routes must be rescheduled depending on the position
and waste load of the alarmed stations. The waste types, can be transported on the same
vehicle without mixing the types; this is done by dividing the capacity of each vehicle into
a number of compartments where each compartment can accommodate a waste type from
one or more stations.
As shown in Figure 1, the dynamic event considered is the appearance of a station
Appl. Syst. Innov. 2023, 6, x FOR PEER REVIEW
that contains at least one waste type having a filling level xˆwi such as xˆwi ∈6 [ofα 29
, 100] (α is
the threshold to trigger the alarm for each waste type) (Figure 2).

Figure 1. Dynamic vehicle routing for separate and smart waste collection.
Figure 1. Dynamic vehicle routing for separate and smart waste collection.
Appl. Syst. Innov. 2023, 6, 30 6 of 26
Figure 1. Dynamic vehicle routing for separate and smart waste collection.

Figure 2.
Figure Fill level
2. Fill level of
of each
each type
type in
in each
each station.
station.

Our problem can be defined as follows: given a set of n stations (each station contains
Our problem can be defined as follows: given a set of n stations (each station contains
m̂ bins, and each bin is designated for a specific type of waste), a group of v homogeneous
𝑚 bins, and each bin is designated for a specific type of waste), a group of 𝑣 homoge-
compartmentalized vehicles, a transfer station n + 1 where vehicles unload different types
neous compartmentalized vehicles, a transfer station 𝑛 + 1 where vehicles unload differ-
of waste collected in the bins, and a depot (where all vehicles begin and complete their
ent types of waste collected in the bins, and a depot (where all vehicles begin and complete
path), a complete undirected graph is defined over n + 1 nodes, with distances dij for
their path), a complete undirected graph is defined over 𝑛 + 1 nodes, with distances
each edge (i, j) in the graph. Each bin of type w in station i has a maximum capacity
𝑑 for each edge (𝑖, 𝑗) in the graph. Each bin of type w in station 𝑖 has a maximum ca-
Cpw (in kg). Let Cij be the travel cost per unit of distance travelled. Each compartment in a
pacity 𝐶𝑝 (in kg). Let 𝐶 be the travel cost per unit of distance travelled. Each compart-
vehicle has a capacity Cqm (in kg).
mentThein asensors
vehicle inside
has a capacity 𝐶𝑞 (in information
the bins transmit kg). on the bins’ fill levels in m3 , which is
The sensors
Appl. Syst. Innov. 2023, 6, x FOR PEERtransformed
REVIEW inside the bins transmit information
into kilograms. To model and solve this problem, on the bins’ fill levels
we will adopt inthe𝑚approach
, which
7 of 29
is
proposed by [35], which consists of dividing the working day into time slices ti (Figureap-
transformed into kilograms. To model and solve this problem, we will adopt the 3);
proach proposed by [35], which consists of dividing the working day
each time slice represents a partial static Multi-Compartment Vehicle Routing Problem into time slices 𝑡
(Figure
(M-CVRP), 3); each time slice represents a partial static Multi-Compartment Vehicle Routing
of each casewhich optimizes
is sufficient thevisiting
after route ofall
eachthevehicle,
stationswhere
that vehicles
triggered must
the serve
alarmallfor
known
each
Problem
stations (M-CVRP),
at minimumwhich optimizes
cost while the route
satisfying of each vehicle,
the capacity where compartments,
of the vehicle vehicles must serveand
waste type.
all
theknown stations at minimum cost while satisfying the capacity of the vehicle compart-
bins capacity.
ments, and the bins capacity.
The discretization strategy we followed is different from that proposed in the litera-
ture because we have stations that can be visited more than once in the same working day.
The vehicle visits the stations that triggered the alarm, and collects the alarm types first.
Then it contains the other waste types in the corresponding compartments if the capacity

Figure 3. The division of the working day.

Figure 3. The division of the working day.
The discretization strategy we followed is different from that proposed in the literature
At the
because we start
haveof the day,
stations thatthe
cancompartmentalized
be visited more than vehicles
once inleave the central
the same workingdepot
day.with
The
avehicle
route plan
visits the stations that triggered the alarm, and collects the alarm types first. alarms
containing the stations that were not collected the day before until the Then it
are triggered
contains to redefine
the other waste the
types routes.
in the corresponding compartments if the capacity of each
The problem of separate
case is sufficient after visiting all andtheintelligent waste
stations that collection
triggered the is assessed
alarm here.
for each Consider
waste type.
the input
At thedata,
startwhich
of thecan
day,bethedivided into two main vehicles
compartmentalized types: static
leaveand
thereal-time inputs
central depot dataa
with
(Figure 4):
route plan containing the stations that were not collected the day before until the alarms
-are triggered
The statictoinputs
redefinearethetheroutes.
information on the waste types and the bin’s positions in a
station, the bin’s capacity,
The problem of separate and andintelligent
the distance between
waste the stations.
collection Secondly,
is assessed concern-
here. Consider
ing vehicles, the static inputs are the number of compartments,
the input data, which can be divided into two main types: static and real-time inputs which are identified
data with the4):
(Figure number of waste types and the capacity of each compartment.
-- Real-time
The static inputs
inputsare arereceived by existing
the information tracking
on the wastetechnology
types and the devices
bin’sinside the bins
positions in a
to give the quantity value of each container in each station, the effective
station, the bin’s capacity, and the distance between the stations. Secondly, concerning replenish-
ment of bins after each visit, and the vehicle charge after collecting each type is not
alarming.
The real-time traceability system required for real-time entry is based on three main
levels, namely the bins, the vehicles, and the operation center, each one connected to the
other as indicated in Figure 5.
Appl. Syst. Innov. 2023, 6, x FOR PEER REVIEW 8 of 29

Appl. Syst. Innov. 2023, 6, 30 7 of 26

vehicles, the static inputs are the number of compartments, which are identified with
the number of waste types and the capacity of each compartment.
Appl. Syst. Innov. 2023, 6, x FOR PEER- REVIEW
Real-time inputs are received by existing tracking technology devices inside the bins
8 of to
29
give the quantity value of each container in each station, the effective replenishment of
bins after each visit, and the vehicle charge after collecting each type is not alarming.

Figure 4. Real time data routing model for separate waste collection.

The types of vehicles used are compartmentalized vehicles. For this, it is assumed
that each compartment is dedicated to a specific waste type. The bins of each station in-
form the operations center of the waste type level in terms of volume, which is converted
into weight, and it triggers the alarm when it reaches a predetermined maximum thresh-
old for each type. The operations center knows the location of the bins at each station, the
weight of the loaded waste, and the current weight of vehicle capacity. The operations
Figure 4. Real time data routing model for separate waste collection.
center sends
Figure 4. the information
Real time to the for
data routing model vehicle at the
separate depot
waste and inviting to visit the alarmed
collection.
stations
The real-time traceability system required for real-timeaccount
according to the route model proposed taking into entry isthe ferries
based on mentioned
three main
and The
at types
each of
station vehicles
serviced, used
the are
vehiclecompartmentalized
communicates to vehicles.
the For
operating
levels, namely the bins, the vehicles, and the operation center, each one connectedthis,
centerit is
the assumed
quantity
to the
that
of each
loaded
other compartment
waste forineach
as indicated istype
Figurededicated to a specific waste type. The bins of each station in-
5. not alarmed.
form the operations center of the waste type level in terms of volume, which is converted
into weight, and it triggers the alarm when it reaches a predetermined maximum thresh-
old for each type. The operations center knows the location of the bins at each station, the
weight of the loaded waste, and the current weight of vehicle capacity. The operations
center sends the information to the vehicle at the depot and inviting to visit the alarmed
stations according to the route model proposed taking into account the ferries mentioned
and at each station serviced, the vehicle communicates to the operating center the quantity
of loaded waste for each type not alarmed.

Figure
Figure 5.
5. Real-time
Real-time traceability
traceability system.
system.

The types of vehicles used are compartmentalized vehicles. For this, it is assumed that
each compartment is dedicated to a specific waste type. The bins of each station inform
the operations center of the waste type level in terms of volume, which is converted into
weight, and it triggers the alarm when it reaches a predetermined maximum threshold
for each type. The operations center knows the location of the bins at each station, the
weight of the loaded waste, and the current weight of vehicle capacity. The operations
Figure 5. Real-time traceability system.
Appl. Syst. Innov. 2023, 6, 30 8 of 26

center sends the information to the vehicle at the depot and inviting to visit the alarmed
stations according to the route model proposed taking into account the ferries mentioned
and at each station serviced, the vehicle communicates to the operating center the quantity
of loaded waste for each type not alarmed.

4. Mathematical Modelling
In this problem, some vehicles visit several bins during working hours. Unlike static
M-CVRP, where all stations are defined at the beginning of the operation, DM-CVRP allows
newly arriving stations during working hours (i.e., the stations alarmed during the time
slice). When a new station comes, the dispatcher or the company must decide whether
to accept or reject the service request. At each time slice, getting a new order means the
station will be called in the following time slices. On the other hand, the refusal means it
will be called during the work slices on the next day.
In this paper, a DM-CVRP is modeled as sequences of the time slice. Let H is the total
working hours. A time slice divides the H time into several time intervals b, such that each
time slice is considered a static M-CVRP
Where:
• t1 : Considers the well-known stations from the previous working day; these stations
are overcarried to the next working day.
• tl /l ∈ { 2, . . . , b}: Consider the stations received during the previous time slices and
some of the stations received on the last working day.
Each time slice contains a continuity period, and a decision period (see Figure 6). The
Appl. Syst. Innov. 2023, 6, x FOR PEER REVIEWwill enter the decision period at the end of the time slice to redefine the itineraries
system 10 of of
29
the following time slices, and consider the new stations that have arrived in the current time
slice. The decision period can be divided into preprocessing phase and the solution phase.

DM-CVRP.
Figure 6. Proposed strategies for DM-CVRP.

The preprocessing
4.1. Index Sets phase is used to determine whether new requests can be added or
rejected, and the solution phase is used to reschedule the routes of all vehicles in subsequent
• 𝑆 = {0, 1…𝑛, 𝑛 + 1 }: Set of 𝑛 stations (each station contains 𝑚 bins) and the depot 0,
time slices. Each resolution phase will be considered all the stations already programmed
and the transfer station 𝑛 + 1
in the time slices and the new stations that have arrived and can be visited.
• 𝑊: Set of waste types
At the end of each time slice, the vehicle’s position current is marked, and the last
• 𝑉 ∶ Set of compartmentalized vehicles available (the vehicles have the same number
stations visited are considered fictitious depots in the next time slice.
of compartments)
A static M-CVRP is applied to obtain the optimal route based on the current informa-
• 𝑀: Set of compartments for each vehicle
tion. Figure 3 illustrates this approach.
• For each station 𝑖 either:
At the beginning of the working hours, all vehicles start from the depot. Thus, the
• 𝐽: Set of the waste types that have not alarmed
initial point of all vehicles for the time interval t1 is {0}. For the other time intervals, tl ,
•l = 2𝐼: ,Set
. . . of
, b,the
eachwaste types
vehicle that from
starts have alarmed
its current position. During the time interval tl ,
•l = 1,Note that
. . . , b, thefor each station
constructed 𝑖 we
route have:
must W = I all
consider ∪ Jthe stations which have already been
•programmed
𝑁 : Set containing
and the new stations that arrived duringday
orders known at the working the start, 𝑙 ∈ time
previous {1, ...,slices
b} which can
•be visited.
𝐴 :The In addition, {0} and n + 1 are also taken into account since it is theslice
set of new stations which are triggered the alarm before the time 𝑡 and
last station
visited which
before caneach
be visited
vehicle’s destination.
• 𝑁′
Some : Set the stations
stations may not considered
be served,in thetotime
due 𝑡
slicehours
limited working. In this case, these sta-
tionsIfwill
a newbe served
stationduring the next
has agreed to working
be visitedhours. Thus,
in a time the problem
slice, notwill
this station onlybe determines
added to
the 𝐴 set. Then
the optimal routethe
for all
set stations but considered
of stations also selects which
in the stations
time slice 𝑡 (𝑙 =be2,performed,
should … , 𝑏) is: 𝑁′ and
=
what waste types can be collected during the current hours of the
𝑁 ∪ 𝐴 \ {0 }, because a station may be visited more than once in the same time slice. working day.
For The𝑙 = proposed
1, the set of formulation is based on
stations considered is:the
𝑁′ following
=𝑁 assumptions:

4.2. Parameters
• 0 : The last station served by the vehicle 𝑣 at the time slice 𝑡
• Cp : The bin capacity of type w (in kg)
Appl. Syst. Innov. 2023, 6, 30 9 of 26

• The fleet is homogeneous.

• Vehicles have multiple compartments.
• Each vehicle starts its tour, and the end returns to the depot.
• Each station is visited only once by a vehicle in a time slice. But it can be called more
than once in a working day.
• Each compartment is dedicated to one type of waste.
• The number of compartments is equal to the number of waste types.

4.1. Index Sets

• S = {0, 1 . . . n, n + 1 }: Set of n stations (each station contains m bins) and the depot 0,
and the transfer station n + 1
• W: Set of waste types
• V : Set of compartmentalized vehicles available (the vehicles have the same number
of compartments)
• M: Set of compartments for each vehicle
• For each station i either:
• J: Set of the waste types that have not alarmed
• I: Set of the waste types that have alarmed
• Note that for each station i we have: W = I ∪ J
• Ntl : Set containing orders known at the working day start, l ∈ {1, . . . , b}
• Atl :The set of new stations which are triggered the alarm before the time slice tl and
which can be visited
• N 0 tl : Set the stations considered in the time slice tl
If a new station has agreed to be visited in a time slice, this station will be added
to the Atl set. Thenthe set of stations considered in the time slice tl (l = 2, . . . , b) is:
N 0 tl = Ntl ∪ Atl \ 0vl , because a station may be visited more than once in the same
time slice. For l = 1, the set of stations considered is: N 0 tl = Ntl .

4.2. Parameters
• 0vl : The last station served by the vehicle v at the time slice tl −1
• Cpw : The bin capacity of type w (in kg)
• C: traveling cost per distance unit
• dij : Distance between node i and node j
• qwi : The quantity of waste in the bin (in kg) of type w in station i (calculated using the
information sent by the sensor in m3 , which is converted to kg)
• Qw : The sum of the quantities alarmed at all stations for each waste type w
• Cqm : The capacity of compartment m for each vehicle (in kg)
• pwi : Penalty for bin overflow for type w in station i
• $w : The maximum threshold for the waste amount in the bin w
• m̂ : Number of compartments for each vehicle (it is assumed that the number of waste
types and the number of bins are equal in each station)
• hw : Number of containers that have alarmed for type w

4.3. Decision Variables

We define the variables of our model as follows:

1, i f the vehicle v moves f rom i to j,
xijv =
0, otherwise,



1, i f station i is visited,
gi =
0, otherwise,

Appl. Syst. Innov. 2023, 6, 30 10 of 26


1, i f the vehicle v loads the type

w in the station j,

yvwj =


0, otherwise,



 1, i f the compartment m is selected f or
the waste type w in the vehicle v,

v
zwm =


0, otherwise,


uwi : Level of a bin’s overflow

F vw ij : The amount of waste type w ∈ W flowing on arc (i, j), by vehicle v in kg
The static M-CVRP in each continuity period for each time slice can be formulated as
the following integer program:

4.4. Model
The objective function (1) contains two parts; the first considers the cost for the
total distance traveled, while the second is the cost of penalty caused by exceeding the
bin capacity.
Min f 1 + f 2 (1)
The constraints of the model are as follows:

f1 = C ∑ ∑v ∑v dij xijv f 2 = ∑ ∑ pwi uwi

v∈V i∈ N 0 tl ∪{0,0l ,n+1} j∈ N 0 tl ∪{0,0l , n+1} w ∈W i ∈ N 0 t
l

s.t
∑0 x0vv j ≤ 1 ∀v ∈ V
l
(2)
j∈ N tl

∑0 x vj,n+1 ≤ 1 ∀v ∈ V (3)
j∈ N tl

∑0 x0vv j =
l
∑0 x vj, n+1 ∀ v ∈ V (4)
j∈ N tl j∈ N tl

∑ x vji = ∑ xijv ∀ v ∈ V, ∀i ∈ N 0 tl ∪ {n + 1} (5)

j∈ N 0 tl ∪{0,0vl } j∈ N 0 tl ∪{n+1,0}
j 6 =i j 6 =i

∑ ∑0 x vj, n+1 = ∑ xnv +1, 0 (6)

v ∈V j ∈ N tl v ∈V

∑ y vwj = 1 ∀ j ∈ N 0 tl p , ∀w ∈ W (7)
v ∈V

xijv ≤ ∑ y vwj ∀ v ∈ V, ∀ j ∈ N 0 tl , ∀ i ∈ Ntl p ∪ {0, 0vl }, i 6= j (8)

w ∈W

∑ ∑ z vwm ≤ m̂ ∀v ∈ V (9)
w ∈W m ∈ M

∑0 qwj y vwj ≤ ∑ Cqm z vwm ∀v ∈ V, ∀w ∈ W (10)

j∈ N tl m∈ M

y vwj = ∑0 xijv ∀ v ∈ V, ∀ j ∈ N 0 tl , ∀i ∈ N 0 tl ∪ {0, 0vl }, ∀w ∈ W (11)

j∈ N tl
j 6 =i

gi = 1, ∀ i ∈ N 0 tl , ∃w ∈ W : qwi ≥ $w Cpw (12)

∑0 y vwj ≥ hw ∀ v ∈ V, ∀w ∈ W (13)
j∈ N tl
Appl. Syst. Innov. 2023, 6, 30 11 of 26

yvwi qwi > 0 ∀v ∈ V, ∀ i ∈ N 0 tl , ∀w ∈ I (14)

yvwj qwj ≤ ∑ Cqm z vwm − Qw − Fwij
v
∀v ∈ V, ∀ j ∈ N 0
tl , ∀i ∈ N 0
tl ∪ {0, 0vl }, ∀w ∈ J (15)
m∈ M

∑ v
Fw,n +1,0 = 0 ∀ w ∈ W (16)
v ∈V

uwi ≥ qwi − Cpw ∀i ∈ N 0 tl , ∀w ∈ W (17)

xijv ∈ {0, 1}∀ v ∈ V, ∀ i, j ∈ S (18)

y vwj ∈ {0, 1} ∀ v ∈ V, ∀ j ∈ S\{0, n + 1} ∀w ∈ W (19)

z vwm ∈ {0, 1} ∀ v ∈ V, ∀ m ∈ M, ∀ w ∈ W (20)

gi ∈ {0, 1} ∀ i ∈ S\{0, n + 1} (21)
uwi ≥ 0 ∀ i ∈ S\{0, n + 1}, ∀ w ∈ W (22)
v
F wij ≥ 0 ∀ v ∈ V, ∀ i, j ∈ S , ∀ w ∈ W (23)
Constraints (2)–(4) ensures that each vehicle starts at the last station that has been
committed to it and returns to the transfer station to unload the collected waste. Con-
straints (5) express that when a vehicle v arrives at station i, it must exit. Constraint (6)
ensures that the number of vehicles returning to the transfer station equals the number of
vehicles returning to the depot. Constraints (7) ensure that each waste type w at station j
is assigned to precisely one vehicle; a separate collection of a single waste type does not
exist. Constraint (8) warrants that if the vehicle visits the station, it must precisely collect
at least one waste type. Constraint (9) ensures that the number of waste types does not
exceed the maximum number of compartments. Constraint (10) warrants that for each case
of vehicle v, its capacity is not exceeded by the sum of the quantities collected. Constraint
(11) states that if a bin in station j is selected, it must be visited by a vehicle v. Constraint
(12) guarantees that a station i is visited when the fill-level of at least one bin of type w
achieves a predetermined threshold. Constraints (13) to (15) guarantee that each alarmed
waste type is collected, and can also collect non-alarmed types if the capacity is sufficient
after visiting all the stations that triggered the alarm. Constraint (16) means that vehicles
are empty at the transfer station before heading to the depot. The overflow constraint (17)
ensures that any overflow is considered. Constraints in equations (18) to (23) are domains
for the variables.

5. Methods and Results

5.1. Model Validation
In this section, we study the validation of our mathematical model. Validating the
model with data and constraints makes it possible to verify the operation of the constraints.
Once a solution is found, the formulation checks whose constraints have been modeled so
they can be alerted in case of error.

5.1.1. Transport Network Data

To solve the problem by the CPLEX optimizer, we test our model with small examples.
For this, we are dealing with an example composed of 5 stations, each station contains
two waste types, 11 arcs, and 10 lines between the depot and the transfer station. Since
our objective is to validate our mathematical model, we fixed the arc length in all arcs to
1000 m, the network shown in Figure 7.
 5.1.1. Transport Network Data
To solve the problem by the CPLEX optimizer, we test our model with small exam-
ples. For this, we are dealing with an example composed of 5 stations, each station con-
tains two waste types, 11 arcs, and 10 lines between the depot and the transfer station.
Appl. Syst. Innov. 2023, 6, 30 12 of 26
Since our objective is to validate our mathematical model, we fixed the arc length in all
arcs to 1000 m, the network shown in Figure 7.

Thenetwork
Figure7.7.The
Figure networkof
ofthe
theexample.
example.

In this example, we studied nine paths between the depot and transfer station. The
In this example, we studied nine paths between the depot and transfer station. The
paths are as follows:
paths are as follows:
• Path1: D-1-3-5-Tr
• Path1: D-1-3-5-Tr
• Path2: D-1-3-4-5-Tr
• Path2: D-1-3-4-5-Tr
• Path3: D-1-2-3-4-5-Tr
• Path3: D-1-2-3-4-5-Tr
• Path4: D-1-2-4-5-Tr
• Path4: D-1-2-4-5-Tr
• Path5: D-1-4-5-Tr
• Path5: D-1-4-5-Tr
• Path6: D-1-2-4-3-5-Tr
• Path6: D-1-2-4-3-5-Tr
• Path7: D-2-3-5-Tr
•• Path7: D-2-3-5-Tr
Path8: D-2-3-4-5-Tr
•• Path8:
Path9:D-2-3-4-5-Tr
D-2-4-5-Tr
• Path9: D-2-4-5-Tr
In this subsection, our objective is to validate our mathematical model therefore, we
In thisthat
assumed subsection,
the number our of
objective is to validate
lines limited in nineour
evenmathematical model
there are other therefore,
lines we
to add. The
assumed that the number of lines limited in nine even there are other lines
purpose is to simplify the validation of the model without changing the real problem. to add. The
purpose is to simplify
Table 1 collect the datathe validation
values. of the model
The objective without
is to find changing
an optimal the collection
waste real problem.
that
Table 1 collect the data values. The objective is to find an optimal waste collection
allows each vehicle to go from the depot to the transfer station and return to the depot. that
To
allows
test oureach vehicle
model to go
on this from the
example wedepot to the transfer
give different valuesstation
to theand return
amount of to the depot.
waste in the To
bin
test our of
(in kg) model
each on
typethis
in example we give different values to the amount of waste in the bin
each station
(in kg) of each type in each station
Table 1. Data values.
Table 1. Data values.
Notation Definition Value
Notation Definition Value
C traveling cost per distance unit 50 €
C Cq1
traveling cost per distance unit
The capacity of compartment 1 for each vehicle (in kg)
50 € 600 kg
𝐶𝑞 The capacity
Cq2 The capacity1of for
of compartment 2 for(in
each vehicle
compartment kg)vehicle (in kg)
each 600 kg 500 kg
𝐶𝑞 The capacity
Cp1 of compartmentThe 2 bin
for capacity of type(in
each vehicle 1 (kg)
in kg) 500 kg 180 kg
Cp Cp2 The bin capacity of type 1 (in kg)
The bin capacity of type 2 ( in kg) 180 kg 150 kg
$ The maximum threshold for the waste amount in each bin 70%

5.1.2. Results
The problem is solved for each of the objectives, because CPLEX can only solve
nonobjective problems. The first objective is the cost for the total distance traveled, while
the second is the cost of penalty caused by exceeding the bin capacity.
We can see in Table 2 that the number of stations visited is equal to the number of
stations existing in each line because it is assumed that the stations have at least one bin
with a filling level greater than or equal to 70% of the bin capacity. The vehicles can collect
the bins with a filling level lower than 70% if the corresponding compartment capacity is
sufficient after having visited all the stations triggering the alarm. Otherwise, the vehicle
cannot collect the type in the corresponding compartment. As in lines 3 and 6, we can see
that the vehicle at a station has collected type 1 but not collected type 2. Moreover, all the
bins are visited before overflowing.
Appl. Syst. Innov. 2023, 6, 30 13 of 26

Table 2. Results by CPLEX.

Path 1 2 3 4 5 6 7 8 9 Total
Collected waste 1 (kg) 420 482 511 396 315 563 414 477 389 3967
Collected waste 2 (kg) 398 409 485 388 276 513 297 355 284 3405
Visited bin 1 3 4 5 4 3 5 3 4 3 34
Visited bin 2 3 4 4 4 3 4 3 4 3 32
Level of a bin’s overflow 0 0 0 0 0 0 0 0 0 0
Objective functions The first objective function: 7361 The second objective function: 0

5.2. Genetic Algorithm for Solving the DM-CVRP

DM-CVRP is an NP-hard problem; exact-solving methods have become very time-
consuming. So we have selected the genetic algorithm as an optimization tool in this article.
The choice is mainly due to its simplicity, use ease, and efficiency in high-complexity
problems. In this section, we adapt the genetic algorithm for each static M-CVRP at each
time slice to solve the DM-CVRP.
Our system designed to solve DM-CVRP is composed of two central systems:
- The first preprocessing phase is responsible for receiving and adding new stations and
divides the working day into a predefined number of time slices. Each slice represents
a static VRP. This system manages all the entrances and exits and organizes the GA.
- The second solving phase, which contains the route optimization process and solu-
tions, executes the GA to solve the static VRP in each time slice generated by the
preprocessing step. To this end, a hybrid GA was proposed to find the best solution
for each given time slice until the stopping condition.
John Holland developed a Genetic Algorithm in 1975. The algorithm’s idea is to
translate the problem into a genetic representation or a chromosome, which is used to
define a solution. Each solution is associated with a numerical value called fitness according
to the function of this solution. At each iteration, a certain number of solutions are selected
according to their fitness to create new ones by applying genetic operators (crossover and
mutation). At the end of the iteration, a replacement step is applied to select the solutions
for the next generation. We use GA in this paper, and we hybrid it by heuristic methods.
The general structure of the Genetic Algorithm is as follows:
1. Encoding the solutions
2. Generate the initial population
3. Repeat
• Selection
• Crossover
• Mutation
• Replacement
Until the condition stops.
Our hybrid genetic algorithm is shown in Figure 8.

5.2.1. Preprocessing Phase

The preprocessing phase manages the new stations, creates static problems, and
engages the stations on compartmentalized vehicles. It is a stage between the arrival of
new stations and the optimization phase. The first static problem created for the first time
consists of all remaining stations from the previous business day. The following static
problems consider stations received during the last time slices. In these problems, the
compartmentalized vehicle departs from whatever committed station it has just visited
with appropriate capacity changes. After each time slice, the best solution is chosen. When
a vehicle has used all its capacity, it is sent back to the depot.
 • Selection
• Crossover
• Mutation
• Replacement
Appl. Syst. Innov. 2023, 6, 30 Until the condition stops. 14 of 26
Our hybrid genetic algorithm is shown in Figure 8.

Figure 8. The
Figure steps
8. The of our
steps genetic
of our algorithm.
genetic algorithm.

5.2.2.
5.2.1. Solving Phase
Preprocessing Phase
The preprocessing is
A GA strategy executed
phase for the
manages eachnew
static VRP created
stations, as described
creates static above.
problems, Each
and en-
chromosome in a population represents a possible solution to a static VRP.
gages the stations on compartmentalized vehicles. It is a stage between the arrival of new
stations
Encodingand the
the optimization
Solutions phase. The first static problem created for the first time con-
sists of all remaining stations from the previous business day. The following static prob-
For each time slice tl , each solution is composed of routes for vehicles and collection
lems consider stations received during the last time slices. In these problems, the compart-
bins. The following table represents the solution for a vehicle v.
mentalized vehicle departs from whatever committed station it has just visited with ap-
In the example illustrated in Table 3. Route 1 is served by vehicle 1, which visits the
propriate capacity changes. After each time slice, the best solution is chosen. When a ve-
ordered list from station 2, starting with 012 (0vl is the last station operated by vehicle v
hicle has used all its capacity, it is sent back to the depot.
at time slice tl −1 ), until the last station 1 before visiting the transfer station to unload the
collected quantities, and then returns to the depot. Vehicle 1 collects types 1 and 3 in station
5.2.2. Solving Phase
2; it collects types 2, 3, and 4 at station 5, and it collects types 1 and 4 at station 1.
A GA strategy is executed for each static VRP created as described above. Each chro-
mosome
Table 3.inExample
a population represents
of a solution a possible
of 3 vehicles with solution toand
11 stations a static VRP.
4 types.

Stations
Encoding0the
1 Solutions
Types l 2 5 1 0
For each time slice 𝑡 , each solution is composed of routes for vehicles and collection
T1 /
bins. The following q21 the solution for a vehicle
table represents 0 𝑣. q11 /
T2 / 0 q52 0 /
T3 In the/ example illustrated in
q23Table 3. Route 1 is served
q53 by vehicle0 1, which visits
/ the
ordered list from station 2, starting with 0 (0 is the last station operated by vehicle 𝑣
T4 / 0 q54 q14 /
Stations time slice 𝑡 ), until the last station 1 before visiting the transfer station to unload the
at
Types 01l 2 5 1 0

T1 / 0 q31 0 q41 /
T2 / q82 q32 0 q42 /
T3 / q83 0 0 0 /
T4 / q84 q34 q11,4 0 /
Appl. Syst. Innov. 2023, 6, 30 15 of 26

Table 3. Cont.

Stations
Types 01l 2 5 1 0

T1 / q61 0 0 q91 /
T2 / 0 q72 q10,2 q92 /
T3 / q63 0 0 0 /
T4 / q64 q34 0 q94 /

Initial Population
A routing solution based on a list of integer numbers, where 1 to n represent stations,
n + 1 represents the transfer station, and 0 represents the depot. The heuristic constructs the
routing of available vehicles beginning with the first one by inserting the request’s origin
and destination in its trajectory to find a suitable solution for DM-CVRP. If the compartment
capacity of the vehicle is reached, we move on to the second vehicle, up to all the stations
that have been served. This heuristic is divided into two steps, the first step is for the first
time slice, and the second step is for the other time slices:
 The initial population of the first time slice:
The working day begins with the first time slice Nt1 that indicates that depot 0 is
inserted, and every time, if a station respects the capacity constraint must be inserted in the
route (remaining vehicle capacities for the types triggering the alarm) until all the stations
in the Nt1 list have been affected. If the vehicle compartment capacity is reached for all
waste types, the vehicle visits the transfer station and then returns to the depot. If the
vehicle capacity constraint is not respected for any remaining station in the Nt1 list, the
current vehicle will move to the transfer station, then to the depot. The algorithm marks
the last station visited, and this station is considered a depot for the next time slice Nt2 .
This heuristic is based on the following algorithm:
 The initial population of the other time slices:
The second step of the heuristic create the initial solutions for the time intervals
l = 2, . . . , b. The same process as the first step is used to generate the initial solutions in
the time slices l = 2, . . . , b, except that the list of stations considers already programmed
stations in the current time slice and the new stations triggering the alarm in the previous
time slice. A station can be visited more than once in the same working day because each
time a type in a station triggers the alarm, this station will enter the list of stations to be
visited in the next time slice. Therefore, the  list of stations considered in the time tl such
that l ∈ {2, . . . , b} is N 0 tl = Ntl ∪ Atl \ 0vl .
At each time slice, the depot considered is the last station visited in the previous
time slice.

Selection
Computed fitness allows the selection of members for the next generation; this can be
done using different selection types, like Roulette, Tournament, and Random.
The tournament selection is adopted to choose some chromosomes to undergo genetic
operations. The most common tournament method is the binary tournament, where we
select two individuals randomly in the population. We choose the best one (which has
the smallest fitness value). This process is repeated until the obtained required number
of individuals is reached. On the other hand, after applying genetic operators, selection
for replacement determines which individuals will disappear from the population in each
generation and which strong individuals survive in the next generation population.
Appl. Syst. Innov. 2023, 6, 30 16 of 26

Algorithm 1: Creation of the initial population.

1. Creates a list that contains the stations considered in the time slice Nt1
2. V=1
3. While(Nt1 6= ∅)
Select randomly a station i from the list Nt1
For each waste type w ∈ I in the station i
If there is w ∈ I such that the quantity qwi is less than or equal to the capacity remaining
in compartment m (With m being the corresponding compartment to the type
w in the vehicle v)
Collects the type w in the corresponding compartment and decreases the capacity of the
compartment m of the vehicle v ( Cqm ← Cqm –qwi )
End for
End If
For each waste type w ∈J in the station i
If there exists w ∈ J such that the quantity qwi is less than or equal to the capacity
remaining in the compartment m of the vehicles after visiting all the stations triggering
the alarm for the type w
Collects the type win the corresponding compartment and decrease capacity of the
compartment m of the vehicle v ( Cqm ← Cqm –qwi )
End for
End If
If all types trigger the alarm in the station i have been transported
Remove this station from the list Nt1
Else
Update i in list Nt1
Else
Get back to 3
End If
End While
If the vehicle capacity is reached for all types
The vehicle v visits the transfer station, and returns to the depot
v=v+1
Get back to 3
Else
Get back to 3
End If
4. Returns the last visited station 0v1 (0v1 is the last station visited in time slice Nt1 )
End

For each individual, we associate a fitness function that corresponds to a compromise

between the two objectives, namely the total distance traveled cost and the cost of the
penalty caused by exceeding the bin capacity. We represent this function as follows:

α1 C ∑ ∑ ∑ dij xijv + α2 ∑ ∑ pwi uwi

v ∈ V i∈ N 0 t ∪{0,0vl ,n+1} j∈ N 0 tl ∪{0,0vl , n+1} w ∈W i ∈ N 0 t
l l

where α1 and α2 represent the importance of the two objectives considered.

Crossover Procedure
Crossover is a genetic operator that combines two chromosomes called “parents”
to produce a new solution called “child”. The purpose of using crossover is that the
new solution can be better than both parents can if it takes the best characteristics from
each parent.
Two parents’ p1 and p2, are selected from the population. The proposed crossover
takes place in two phases:
Appl. Syst. Innov. 2023, 6, 30 17 of 26

- The first phase randomly chooses a crossing point k for each arc (the red line in
Figures 9 and 10). The crossover procedure used in this paper is eluted as follows: the
first part of the first parent is copied as the first part of child 1, and then the elements
of the second part of this parent are considered the second part of child 2. The first
part of the second parent is copied as the first part of child 2, and then the elements of
Appl. Syst. Innov. 2023, 6, x FOR PEER REVIEW
the second part of this parent are considered the second part of child 1. The19two of 29
parts
for each child are combined to form a new path (See Figure 9).
Appl. Syst. Innov. 2023, 6, x FOR PEER REVIEW 19 of 29
- After crossing, the result does not always correspond to a feasible solution. We
illustrate this situation with the example in Figure 10.

Figure
Figure 9. Example
Figure 9. Example
Exampleofof the
ofthe
the first
first
first phase
inin
phase
phase inthe
the crossover
the procedure.
crossover
crossover procedure.
procedure.

Figure 10. Example to build the second phase in the crossover procedure.
Figure 10.
Figure 10. Example
Exampleto tobuild
buildthe
thesecond
secondphase
phaseinin
thethe
crossover
crossoverprocedure.
procedure.
Note that the obtained result does not present a feasible solution (see Figure 10), for
childNote
2, thethat
sumthe obtained
of the result
quantities does not in
transported present a feasible
compartments solution
2 and (see significant
4 is more Figure 10), for
child 2, the
than the sum ofof
capacities the
thequantities transported
corresponding in compartments
compartment. Therefore, we2 and
have4 istomore
makesignificant
some
than the capacities
adjustments, of the corresponding
so the second compartment.
phase is testing the feasibility ofTherefore, we have to make
these new chromosomes usingsome
adjustments, so the second
the heuristic illustrated phase
in the is testing
following the feasibility
algorithm for childof1these new chromosomes
(the same algorithm is ap-using
plied
the to modify
heuristic child 2): in the following algorithm for child 1 (the same algorithm is ap-
illustrated
Appl. Syst. Innov. 2023, 6, 30 18 of 26

We assume that we have four types of waste and that the trucks have four compartments.
We assume that each truck has a capacity of 23 tons distributed as follows: The first
compartment capacity is 7 tons, the second compartment capacity is 5 tons, the third
compartment capacity is 6 tons, and the fourth compartment capacity is 5 tons.
Note that the obtained result does not present a feasible solution (see Figure 10), for
child 2, the sum of the quantities transported in compartments 2 and 4 is more significant
than the capacities of the corresponding compartment. Therefore, we have to make some
adjustments, so the second phase is testing the feasibility of these new chromosomes using
the heuristic illustrated in the following algorithm for child 1 (the same algorithm is applied
to modify child 2):

Algorithm 2: Modification of the crossover operator.

For each vehicle v
For each station j in the path of child 1
For each waste type w in the station j
w=p
While (w < m̂ )
If (qwi + . . . + qw,i0 +j0 + . . . + qw,i0 +m0 ≤ Cqw )
Collected all waste type w (if bwj = qwj ) in the corresponding compartment of vehicle
v for child 1
Else
Repeat
For (j=i0 + j0 . . . i0 + m0 )
Calculate (qwi + . . . + qwj )
If (qwi + . . . + qwj > Cqw )
Replace qwj by 0 for child 1
Else
Collected the waste type w in the station j in the corresponding compartment of
vehicle v for child 1, and calculate qwi . . . + qwj +qw,j+1
j ← j + 1 get back to 11
End If
End For
End If
While j = i0 + m0

End While
p=p+1
End While
End For
End For
End For

Mutation Procedure
The mutation is applied with a certain probability. The mutation operator allows
modifying an individual’s genes to find other parts that are very different. At this point,
some parents are chosen from the population to apply the mutation operation. We use
the inversion operator, which is simple in that it generates a point along the length of
two chromosome genes and then inverts those two points (two stations). An example of
inversion is shown in Figure 11.
After the mutation, the sum of the quantities collected for the type w at the stations
visited may be greater than the corresponding compartment capacity.
Appl.
Appl. Syst.
Syst.Innov.
Innov.2023,
2023,6,
6,x30FOR PEER REVIEW 2119ofof29
26

Figure11.
Figure Exampleof
11.Example ofMutation
Mutationprocedure.
procedure.

To keepstudying
Before the feasibility of the solution,
the example based on wethe
propose the of
real data following
valorsulmodification algorithm:
using our hybrid ge-
netic algorithm (HGA), we treat the example presented in Section 5.1 to show the effi-
Algorithm
ciency Modification
of the3: approach usedof the mutation
HGA. operator.
Table 4 shows the results found by applying HGA
method and
(p is the the results
change point) that we found using the CPLEX software (exact method). To illus-
trate
Forour
w =results,
1 to m̂ we calculate the quantity collected for each type in each arc (TCW1 &
TCW2), If (the number
qwi + of +
. . . + qw,p visited
. . . + qbins
w,i +m for each
> Cq w ) type at each station (TVB1 & TVB2), and the
bins overflowqw,p ←level0 in each arc, and the transport cost (TC). We obtained the results shown
Else4.
in Table
Collected the waste type w in the station p in the corresponding compartment of vehicle v
In Table 4, we can see that the results obtained are the same for both methods (CPLEX
w ← w + 1 get back to 3
and HGA). This happens because if the vehicle collects a waste type in a station, it collects
End If
all the
Endquantity
For existing in the bin.
End
Table 4. Comparison of the results found by HGA and CPLEX.

CPLEX Before studying the example based on the realHGAdata of valorsul using our hybrid genetic
algorithm (HGA), we treat the example presented in Section 5.1 to show the efficiency
TCW1 TCW2 TVB1 TVB2 TLBO TC TCW1 TCW2 TVB1 TVB2 TLBO TC
of the approach used HGA. Table 4 shows the results found by applying HGA method
3967 3405 34 32 0
and the results that 7361 3967 the CPLEX
we found using 3405 software
34 32 method).
(exact 0 To illustrate
7361our
results, we calculate the quantity collected for each type in each arc (TCW1 & TCW2), the
6. Experimental
number Study
of visited bins for each type at each station (TVB1 & TVB2), and the bins overflow
level
6.1. in each arc,
Real-World and the transport cost (TC). We obtained the results shown in Table 4.
Study
The case study presented in this article is based on data from those Ramos et al. [10]
Table
and 4. Comparison
Silva, of the
R. F. R [29], results
with some found by HGA andtoCPLEX.
modifications adapt this data to our problem. These
are the waste collection operations by Valorsul–Valorizaç˜ao e Tratamento de Resíduos
CPLEX HGA
S’olidos, S.A. This company operates in the region of Lisbon, the capital of Portugal. Val-
TCW1 TCW2 TVB1 TVB2
orsul TLBO
is responsible TC collecting
for TCW1 differentTCW2types of TVB1 TVB2
waste, including TLBO
recyclableTCwaste
3967 3405 34 32
(glass, 0
paper/cardboard, 7361and plastic/metal).
3967 3405The collection
34 of32recyclable0 materials 7361
is car-
ried out in 14 municipalities by a homogeneous fleet of vehicles based on a single depot.
During In compilation,
Table 4, we canthesee
team
thatrecords the obtained
the results fill level of
arethe
thebins.
sameIt for
classifies them as(CPLEX
both methods empty
(0%), less than
and HGA). half
This (25%),because
happens half (50%), more
if the thancollects
vehicle half (75%), or full
a waste type(100%). The collection
in a station, it collects
system proposedexisting
all the quantity by Valorsul
in thefor
bin.the collection of recyclable waste then estimates the fill-
ing level of the bins and decides whether to include these bins in the route. This estimate
is6. calculated
Experimental Study after having visited all the assigned collection points, the truck
as follows:
6.1. Real-World
returns Study
with information on the filling levels of the bins. This information and the density
values The provided by presented
case study Valorsul from in this Paper
articleand/Cardboard
is based on data (PC),
from Plastic
thoseand/Metal
Ramos et al. (PM),
[10]
and Glass (G) and the density values provided by Valorsul from PC, PM,
and Silva, R. F. R [29], with some modifications to adapt this data to our problem. These are and G (ρ = 40
kg/m 3; ρ = 20 kg/m 3; ρ = 300 kg/m 3; see densities in Table 5) were used to calculate
the waste collection operations by Valorsul–Valorizaç˜ao e Tratamento de Resíduos S’olidos,
the
S.A.amount collectedoperates
This company per bin. inThe formula
the region to of calculate
Lisbon, thethe capital
quantity ofordered
Portugal.perValorsul
station
and per type of waste follows (1.1a), where A is the quantity collected at station 𝑖 of
Appl. Syst. Innov. 2023, 6, 30 20 of 26

is responsible for collecting different types of waste, including recyclable waste (glass,
paper/cardboard, and plastic/metal). The collection of recyclable materials is carried out
in 14 municipalities by a homogeneous fleet of vehicles based on a single depot. During
compilation, the team records the fill level of the bins. It classifies them as empty (0%), less
than half (25%), half (50%), more than half (75%), or full (100%). The collection system
proposed by Valorsul for the collection of recyclable waste then estimates the filling level of
the bins and decides whether to include these bins in the route. This estimate is calculated
as follows: after having visited all the assigned collection points, the truck returns with
information on the filling levels of the bins. This information and the density values
provided by Valorsul from Paper and/Cardboard (PC), Plastic and/Metal (PM), and Glass
(G) and the density values provided by Valorsul from PC, PM, and G (ρPC = 40 kg/m3 ;
ρPM = 20 kg/m3 ; ρG = 300 kg/m3 ; see densities in Table 5) were used to calculate the
amount collected per bin. The formula to calculate the quantity ordered per station and per
type of waste follows (1.1a), where Aiw is the quantity collected at station i of type w, ρw is
the type density w, Cpiw is the total capacity of all bins of waste type w in station i, and
FLiw the average fill level of all containers of type w in station i.

Aiw (kg) = ρw ∗ Cpiw ∗ FLiw (1.1a)

Table 5. Density of the materials collected by Valorsul inside the respective bins, and inside
collection vehicles.

Density Inside the Bins Density in the Collection

Type of Waste
(kg/m3 ) Vehicles (kg/m3 )
Paper and/Cardboard 40 250
Plastic and/Metal 20 150
Glass 300 600

For more clarification, part of the visit sequence carried out is shown in Table 6, along
with information on the filling level of each bin collected.

Table 6. Part of the visit sequence of a shift.

Hour Container Amount

Station Bin Filling Level (%)
(hh:mm:ss) Capacity (m3 ) Collected (kg)
... ... ... ... ... ...
19:25:20 1590 P0800 2.5 0% 0
19:25:33 1590 P2257 2.5 2.5% 25
19:27:00 578 T_P0021 2.5 0% 0
19:29:48 2293 P2306 2.5 50% 50
19:39:38 574 P0438 2.5 50% 50
... ... ... ... ... ...

For example, the vehicle collected 50 kg of PC in bin 2293 (marked in red in Table 6),
calculated in Formula (1.1b).

A2293.PC = 40 ∗ 2.5 ∗ 50% = 50 kg (1.1b)

The estimated fill level of each bin heard by Valorsul is calculated by dividing the
total fill levels recorded for each bin by the number of days in the period multiplied by
the time interval between routes (Ramos et al. [10] and Silva [29]). This estimation allows
a comparison between the current situation of Valorsul and the results obtained when
applying our approach.
Ramos et al. selected three routes to collect the type (paper/cardboard) to analyze the
current situation (blind collection) and the future situation (installation of sensors to obtain
Appl. Syst. Innov. 2023, 6, 30 21 of 26

real-time information on the filling levels of the bins). Route 6 with 68 containers, route 11
with 74 containers, and route 13 with 84 containers, totaling of 226 containers. The scalar
data of the study are those presented in Table 7. Cij is the cost of travel per distance unit,
including fuel consumption, vehicle maintenance, and the salary of the collection team. qi
is the amount of waste at bin i, (considering the density of paper/cardboard), simulates the
values the sensors will read and transmit daily. To apply the resolution method, we use the
bin’s fill levels registered by the collection team after visiting these bins plus an expected
accumulation rate per bin i. This accumulation rate was calculated by dividing the bin fill
levels recorded by the collection team by the number of days that passed after the previous
day the bins were collected (Ramos et al. [10]).

Table 7. Parameters values and sources.

Parameters Values Source

Cij 1 €/km Valorsul
4000 kg Valorsul
Q Cq PC = 2500 kg, Silva et al. [29]
Cq PM = 1500 kg Silva et al. [29]
Cp PC =74 kg Ramos et al. [10]
Cpw
Cp PM = 74 kg —
qi Team records for day 1 Valorsul
ρPC = 29.5 kg/ m3 Valorsul
ρw
ρPM = 29.5 kg/ m3 —
dij Adjusted Euclidian distance Ramos et al. [10]
S 226 Valorsul

For Valorsul, the types of waste are not always in the same station; there are other
routes to collect plastic/metal and glass. To apply our model and our proposed ap-
proach, we assume that each station contains two bins (one for paper/cardboard, one
for Metal/Plastic), and we use a two- compartment vehicle. Table 6 shows the data from
Valorsul, and we add some more data for it to be applied to our problem (from the data
illustrated by Silva [29]).
Ramos et al. considered a planning period of 30 days to collect the paper/cardboard;
the profit obtained by the Valorsul for the three routes is presented in Table 8. During the
planning period, routes 6 and 11 were twice repeated, route 13 four times, and 620 bins
were visited, with 12% empty and 1% overflowing. In these eight routes, the waste collected
amounted to 17,000 kg, and 1200 km were traveled, resulting in an average of around 15 kg
per km traveled), and the vehicle utilization rate varied between 46% and 67%.

Table 8. Actual Valorsul operations (Paper/Cardboard).

Collected Vehicle Usage

Day Route Distance (km) Number of Bins Vehicles Used
Waste (kg) Rate (%)
1 13 1350 117 84 1 67.5%
7 11 1200 129 74 1 60%
8 6/13 900/1400 208/132 152 2 57.5%
15 13 1430 130 84 1 71.5%
19 11 1200 139 74 1 60%
22 6/13 1000/1200 221/127 152 2 55%
Total 9680 1203 620 8 -

Since we do not have data in the literature on the collection of Plastic/Metal, we will
propose data presented in Table 9. On these eight routes, it is assumed that the waste
Appl. Syst. Innov. 2023, 6, 30 22 of 26

collected (Plastic/Metal) amounted to 9000 kg and the rate of use of the vehicles varied
between 55% and 71.5%.

Table 9. Proposed operations (Plastic/Metal).

Over- Empty % of
Collected Distance Visited Vehicles Vehicle Usage
Day Route Flowing Visited Empty
Waste (kg) (km) Bins Used Rate (%)
Bins Bins Bins
1 13 2472 117 84 0 6 7.1 1 61.8
7 11 2343 129 74 0 17 23.0 1 58.6
8 6/13 1420/2564 208/132 152 1 33 21.7 2 49.8
15 13 2693 130 84 1 7 8.3 1 67.3
19 11 2251 139 74 0 2 2.7 1 56.3
22 6/13 1513/2195 221/127 152 2 8 5.3 2 46.3
Total 17450 1203 620 4 73 11.8 8 -

6.2. Study Results

The modified genetic algorithm applied to our model was implemented in C++ lan-
guage on a Windows 10 Professional Intel(R) Xeon(R)W2155 CPU (20 CPUs) at 3.3 GHz,
with 32.768 GB of RAM to test the quality and performance of the modified genetic algo-
rithm and the proposed model.
The choice of parameters influences the results. For that, several tests were carried out
on the parameters of the method (crossing and mutation) to identify the best parameters to
use. We launched our program with different sizes of populations {20, 50,100,500}, and we
found the best solutions for a population of size 50 individuals. Therefore, the parameters
of the proposed approach are set as follows:
Crossover rate: 50% of the population size
Mutation rate: 20% of the population size
Population size: 50 individual
Number of iterations: 100 iterations

6.2.1. Experimental Results: Static M-CVRP

Our approach in the static case was tested, considering the same data presented in
Table 6, except for the amount of waste and gives in real-time. The types selected to collect
with multi-compartment vehicles are Paper and/Cardboard, Plastic, and/Metal. Therefore,
each vehicle needs to have two compartments, one for each type, and we consider the
capacity of the two compartments given by Silva et al. 2016. In addition, we add a penalty
of 1 € per kg for each bin that overflows. The value of the minimum fill level at which the
containers must be visited when applying the proposed Metaheuristic is assumed to be
equal to 70% of the container volume, i.e., each bin that has 70% or more of its capacity for
both waste types.
After applying our approach, we can see in Table 10 that the number of visited stations
is lower than the number of visited bins given by Valorsul (233 against 620). Since the
stations visited are the stations that have at least one container with a fill level greater than
or equal to 70% of the bin capacity (having a bin capacity of 52.5 kg at most). We also note
that there are plastic/metal type bins were collected even if their fill level did not reach
the required weight. This is due to the ability of the plastic/metal type compartment to
collect this waste type at each station visited after calculating the total disturbing quantities
of this type.
Appl. Syst. Innov. 2023, 6, 30 23 of 26

Table 10. Solutions obtained through the hybrid Genetic Algorithm.

Compartment
Collected Visited Level of a Empty Vehicle
Visited Distance Vehicles Usage
Day/Route Waste (kg) Bins Bin’s Visited Usage
Stations (km) Used Rate %
PC PM PC PM Overflow Stations Rate %
PC PM
1940 77.6
1/13 35 35 33 76.41 0 0 1 76.54
1132 75.47
1820 72.8
7/11 33 33 33 78.32 0 0 1 67.1
921 61.4
2890 57.8
8/6 & 8/13 51 51 46 119.10 0 0 2 57.53
1718 57.26
2113 84.52
15/13 39 39 39 87.95 0 0 1 82.43
1205 80.33
1517 60.68
19/11 26 26 26 54.05 0 0 1 58.38
841 56.07
22/6 & 2819 56.38
49 49 47 116.21 0 0 2 53.96
22/13 1546 51.53
13,099
Total 233 233 206 532.04 0 0 8 - -
7363

The 73 empty bins were visited during the period in the real case, while in our
approach, there are no open visited stations. Moreover, no container exceeds its maximum
capacity, i.e., all bins are visited before overflowing, while there are four overflowing bins
for actual Valorsul operations.
Regarding the distances, a reduction of 55.77% of the traveled distance (532 km vs.
1203 km), so we see that the total distance to be crossed in the actual case of Valorsul is twice
as large. This is justified by the fact that the vehicles travel long distances to collect fewer
bins, which requires a model capable of choosing the containers worth visiting according
to their fill levels and location.
When the hybrid metaheuristic is applied, the vehicle usage rates are always higher
than 50%, varying between 53% (day 22) and 82% (day 15). Valorsul vehicle usage rates are
lower, starting at 46% (day 22) and reaching 67% (day 15) at maximum, so this allows us to
say that our approach gives satisfactory results.

6.2.2. Experimental Results: Dynamic M-CVRP

In this part, we present the results of our proposed approach for the resolution of
the dynamic M-CVRP, where we must satisfy the integration of new stations when the
alarms are triggered for at least one waste type, then the routes have been planned. Their
exploitation started by adding the characteristics below to obtain the dynamic problem:
The discretization of a working day.
The time of appearance of a station.
Our tests were carried out over six days, and we kept the parameter values used in
the static case, and we simulated the length of a day by 22,000 s (tl = 22,000 s), and thus
the algorithm is stopped if the elapsed time is greater than or equal to tl . This duration is
divided into time slices having variable sizes according to the course of an iteration of our
approach Titer .
This group contains six days with 5 degrees of dynamism (dod) (Messaoud et al. [16])
which are 90%, 10%, 50%, 70%, and 30% (dod = 0% present the static case).
Table 11 reports the obtained results on the different degrees of dynamism for our
modified genetic algorithm. It indicates the total traveled distance and the number of
stations visited. We note, on the one hand, that respecting all the constraints, the dynamic
aspect influences the number of stations visited and the distances covered, it is easy to see,
on the one hand, that if we increase the degree of dynamism, the total distance increases,
then it decreases when the degree reaches 50%, then it goes up or remains unstable, and
on the other hand, we notice that in some cases, even if the distance increases, the number
of stations visited remains the same, or else diminished, This is due to the fact that the
number of visited stations does not depend only on the distance traveled but specifically
Appl. Syst. Innov. 2023, 6, 30 24 of 26

on the number of stations triggering the alarm, and which are integrated into the different
time slices. And in DVRP, the vehicle can service stations on its way, which can reduce the
distances covered.

Table 11. Results for DM-CVRP.

Our Approach
Day/Route Dod (%) Distance Visited Visited Bins
(km) Stations PC PM
1/13 0 76.34 35 35 30
10 80.32 35 35 30
30 97.91 35 35 30
50 78.56 35 35 30
70 91.47 34 34 30
90 95.52 35 35 30
7/11 0 78.23 33 33 33
10 92.37 32 32 32
30 100.67 33 33 33
50 83.78 32 32 32
70 81.53 31 31 31
90 105.14 33 33 33
8/6 0 119.52 51 51 36
8/13 10 130.74 51 51 36
30 145.23 51 51 36
50 127.17 50 50 50
70 138.31 51 51 51
90 122.28 51 51 51
15/13 0 87.95 39 39 39
10 91.15 38 38 38
30 115.02 39 39 39
50 98.39 39 39 39
70 94.55 39 39 39
90 111.28 39 39 39
19/11 0 54.05 26 26 26
10 60.34 26 26 26
30 77.12 26 26 26
50 71.88 26 26 26
70 65.17 26 26 26
90 57.07 24 24 24
22/6 0 116.21 49 49 42
22/13 10 120.05 49 49 42
30 141.76 49 49 42
50 125.92 45 45 41
70 118.54 47 47 42
90 135.16 48 48 42

In static VRP, the more Ntl contains more stations, the simulation results tend to
be much better because the vehicle often returns to the depot with approximately full
loading, and thus, the vehicle number is reduced. However, in DVRP, it is uncertain
that the optimization result is better when Ntl contains more stations because there are
undefined factors.
Generally, at the end of each day, the number of unvisited stations with at least
one waste type with a filling rate greater than or equal to 70% is negligible compared to
the number of stations having triggered the alarm and which is visited regardless of the
dynamism degree. This makes it possible to say that our approach gives satisfactory results
in terms of quality of service in the dynamic case.
The approaches proposed by existing works in the literature for the smart waste
collection problem did not respect all the constraints, especially in the dynamic case; this is
why our results in the dynamic case, were not compared with those found in the literature.
Appl. Syst. Innov. 2023, 6, 30 25 of 26

In addition, to show the effectiveness of our approach, we focused on comparing our result
in the static case with the current situation of the Valorsul Company.

7. Conclusions
This work introduces the problem of separate waste collection in smart cities, where
it is assumed that each station contains a limited number of bins; each bin is specialized
for a single type of waste. The uncertainty is considered regarding the amount of waste
in each container at each station by installing sensors capable of real-time reading and
transmitting the stations’ fill levels. This makes it possible to consider the problem as a
multi-compartment vehicle routing problem to collect only the stations having at least one
more attractive bin, which improves the efficiency of operations. For that, we proposed a
mathematical model whose objective is to minimize total transport costs, including crossed
distance and penalties caused by bin overflowing. To validate our model we tested the
model using the CPLEX Optimizer on a small example, since the problem was a very
complicated one.
We have proposed a static M-CVRP resolution approach based on genetic algorithm
hybridization. Afterward, we adapted this approach to solve the M-CVRP in the dynamic
case; in the latter, during the execution of the planned route, we integrated new stations
triggering the alarm for at least one waste type.
To validate our approaches, we used a real dataset from Valorsul with some mod-
ifications to adapt these data to our problem, where we dedicated to the collection of
two types of recyclable waste Paper/Cardboard (PC) and Plastic/Metal (PM) using multi-
compartment vehicles. We were able to ensure the effectiveness of our approach based on
the results of the static case, which are very encouraging. Moreover, in the dynamic case,
we used the definition of the degree of dynamism with such significant results established.

Author Contributions: Conceptualization, Y.B. and A.E.A.; methodology, Y.B. and A.E.A.; software,
Y.B.; validation, Y.B. and A.E.A.; formal analysis, Y.B. and A.E.A.; investigation, Y.B. and A.E.A.;
resources, Y.B.; data curation, Y.B.; writing—original draft preparation, Y.B.; writing—review and
editing, Y.B.; visualization, Y.B. and A.E.A.; supervision, A.E.A.; project administration, A.E.A.;
funding acquisition, Y.B. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Balandin, S.; Andreev, S.; Koucheryavy, Y. Internet of Things, Smart Spaces, and Next Generation Networks and Systems. In
Proceedings of the NEW2AN: International Conference on Next Generation Wired/Wireless Networking, Petersburg, Russia,
27–29 August 2014.
2. Habibzadeh, H.; Qin, Z.; Soyata, T.; Kantarci, B. Large Scale Distributed Dedicated- and Non-Dedicated Smart City Sensing
Systems. IEEE Sens. J. 2017, 17, 7649–7658. [CrossRef]
3. Page, A.; Hijazi, S.; Askan, D.; Kantarci, B.; Soyata, T. Research Directions in Cloud-Based Decision Support Systems for Health
Monitoring Using Internet-of-Things Driven Data Acquisition. Int. J. Serv. Comput. 2016, 4, 18–34.
4. Hijazi, S.; Page, A.; Kantarci, B.; Soyata, T. Machine Learning in Cardiac Health Monitoring and Decision Support. IEEE Comput.
2016, 49, 38–48. [CrossRef]
5. Silva, B.N.; Khan, M.; Han, K. Towards sustainable smart cities: A review of trends, architectures, components, and open
challenges in smart cities. Sustain. Cities Soc. 2018, 38, 697–713. [CrossRef]
6. Nirde, K.; Mulay, P.S.; Chaskar, U.M. IoT based solid waste management system for smart city. In Proceedings of the International
Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 15–16 June 2017; pp. 666–669.
7. Hannan, M.A.; Al Mamun, M.A.; Hussain, A. A review on technologies and their usage in solid waste monitoring and
management systems: Issues and challenges. Waste Manag. 2015, 43, 509–523. [CrossRef] [PubMed]
8. Reed, M.; Yiannakou, A.; Evering, R. An ant colony algorithm for the multi-compartment vehicle routing problem. Appl. Soft
Comput. 2014, 15, 169–176. [CrossRef]
Appl. Syst. Innov. 2023, 6, 30 26 of 26

9. Mofid-Nakhaee, E.; Barzinpour, F. A multi-compartment capacitated arc routing problem with intermediate facilities for solid
waste collection using hybrid adaptive large neighborhood search and whale algorithm. Waste Manag. Res. 2019, 37, 38–47.
[CrossRef]
10. Ramos, T.R.P.; Morais, C.S.; Barbosa-Póvoa, A.P. The smart waste collection routing problem: Alternative operational management
approaches. Expert Syst. Appl. 2018, 103, 146–158. [CrossRef]
11. Lozano, Á.; Caridad, J.; De Paz, J.F. Smart Waste Collection System with Low Consumption LoRaWAN Nodes and Route
Optimization. Sensors 2018, 18, 1465. [CrossRef]
12. Omara, A.; Gulen, D.; Kantarci, B.; Oktug, S.F. Trajectory-Assisted Municipal Agent Mobility: A Sensor-Driven Smart Waste
Management System. J. Sens. Actuator Netw. 2018, 7, 29. [CrossRef]
13. Lu, X.; Pu, X.; Han, X. Sustainable smart waste classification and collection system: A biobjective modeling and optimization
approach. J. Clean. Prod. 2020, 276, 124183. [CrossRef]
14. Hrabec, D.; Senland, P.; Nevrlý, V.; Popela, P.; Hoff, A.; Šomplák, R.; Pavlas, M. Quantity-Predictive Vehicle Routing Problem for
Smart Waste Collection. Chem. Eng. Trans. 2019, 76, 1249–1254.
15. Anagnostopoulos, T.; Kolomvatsos, K.; Anagnostopoulos, C.; Zaslavsky, A.; StathesHadjiefthymiades, S. Assessing dynamic
models for high priority waste collection in smart cities. J. Syst. Softw. 2015, 110, 178–192. [CrossRef]
16. Messaoud, E.; Elhilali Alaoui, A.; Boukachour, J. Hybridation de l’algorithme de colonie de Fourmis avec l’algorithme de
recherche à grand Voisinage pour la résolution du VRPTW statique et dynamique. INFOR 2013, 51, 41–51. [CrossRef]
17. Ramachandranpillai, R.; Arock, M. A solution to dynamic green vehicle routing problems with time windows using spiking
neural P systems with modified rules and learning. J. Supercomput. 2021, 77, 9689–9720. [CrossRef]
18. Talouki, R.Z.; Javadian, N.; Movahedi, M.M. Optimization and incorporating of green traffic for dynamic vehicle routing problem
with perishable products. Environ. Sci. Pollut. Res. 2021, 28, 36415–36433. [CrossRef] [PubMed]
19. Sreelekshmi, V.; Jyothisha, J.N. Dynamic Vehicle Routing for Solid Waste Management. In Proceedings of the IEEE Region 10
Symposium (TENSYMP), Cochin, India, 14–16 July 2017.
20. Messaoud, E.; Elhilali Alaoui, A.; Boukachour, J. Solving the Dynamic Vehicle Routing Problem using Genetic Algorithms. In
Proceedings of the 2nd IEEE International Conference on Logistics Operations Management 2014, Rabat, Morocco, 5–7 June 2014.
21. Bouziyane, B.; Dkhissi, B.; Cherkaoui, M. Solving a Dynamic Vehicle Routing Problem with Soft Time Windows Based on Static
Problem Resolution by a Hybrid Approach. Int. J. Supply Oper. Manag. 2018, 5, 134–151.
22. Kilby, P.; Prosser, P.; Shaw, P. Dynamic VRPs: A Study of Scenarios; Technical Report APES-06-1998; University of Strathclyde:
Glasgow, UK, 1998.
23. Montemanni, R.; Gambardella, L.M.; Rizzoli, A.E.; Donati, A.V. A New Algorithm for a Dynamic Vehicle Routing Problem Based on
Ant Colony System; Technical Report IDSIA-23-02; IDSIA: Lugano, Switzerland, 2002.
24. Bouleft, Y.; Elhilali Alaoui, A. Modeling and simulation of multi-compartment vehicle routing problems to transport different
types of solid waste. Int. J. Oper. Res. 2021. [CrossRef]
25. Muyldermans, L.; Pang, G. On the Benefits of Co-collection: Experiments with a Multi-compartment Vehicle Routing Algorithm.
Eur. J. Oper. Res. 2010, 206, 93–103. [CrossRef]
26. Henke, T.; Speranza, M.G.; Wascher, G. A branch-and-cut algorithm for the multi-compartment vehicle routing problem with
flexible compartment sizes. Ann. Oper. Res. 2019, 275, 321–338. [CrossRef]
27. Rabbani, M.; Farrokhi-asl, H.; Rafiei, H. A hybrid genetic algorithm for waste collection problem by heterogeneous fleet of
vehicles with multiple separated compartments. J. Intell. Fuzzy Syst. 2016, 30, 1817–1830. [CrossRef]
28. Oliveira, A.D.; Ramos, T.R.P.; Martins, A.L. Planning Collection Routes with Multi-compartment Vehicles. Oper. Res. Big Data
2015, 15, 131–138.
29. Silva, R.F.R. The Multi-Compartment Vehicle Routing Problem in the Collection of Recyclable Municipal Solid Waste. Master’s
Thesis, Iscte-University Institute of Lisbon, Lisbon, Portugal, 2016.
30. Yousra, B.; Ahmed, E.A. Multi-compartment vehicle routing problem for waste collection in smart cities. In Proceedings of the
IEEE 6th International Conference on Logistics Operations Management (GOL), Strasbourg, France, 29 June–1 July 2022.
31. Yang, J.; Tao, F.; Zhong, Y. Dynamic routing for waste collection and transportation with multi-compartment electric vehicle using
smart waste bins. Waste Manag. Res. 2022, 40, 1–13. [CrossRef] [PubMed]
32. Anghinolfi, D.; Paolucci, M.; Robba, M.; Taramasso, A.C. A dynamic optimization model for solid waste recycling. Waste Manag.
2013, 33, 287–296. [CrossRef] [PubMed]
33. Abdallah, M.; Adghim, M.; Maraqa, M.; Aldahab, E. Simulation and optimization of dynamic waste collection routes. Waste
Manag. Res. 2019, 37, 793–802. [CrossRef] [PubMed]
34. Expósito-Márquez, A.; Expósito-Izquierdo, C.; Brito-Santana, J.; Moreno-Pérez, A. Greedy randomized adaptive search procedure
to design waste collection routes in La Palma. Comput. Ind. Eng. 2019, 137, 106047. [CrossRef]
35. Montemanni, R.; Gambardella, L.M.; Rizzoli, A.E.; Donati, A.V. Ant colony system for a dynamic vehicle routing problem.
J. Comb. Optim. 2005, 4, 327–343. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.