Available online at www.sciencedirect.

com

ScienceDirect
IFAC PapersOnLine 55-10 (2022) 133–138
141–146

A Simulation-based Optimization Model for the Vehicle Routing Problem

with Bimodal
A Simulation-based Stochastic
Optimization Travel
Model forTime Distribution
the Vehicle Routing Problem
with Bimodal Stochastic Travel Time Distribution
Amina Antit*,Amel Jaoua**, Safa Bhar Layeb***, Chefi Triki****
*LR-OASIS, Amina
NationalAntit*,Amel
EngineeringJaoua**,
School ofSafa Bhar
Tunis, Layeb***,
University Chefi
of Tunis ElTriki****
Manar, Tunis, Tunisia
(e-mail: [email protected])
*LR-OASIS, National Engineering School of Tunis, University of Tunis El Manar, Tunis, Tunisia
**LR-OASIS, National Engineering [email protected])
School
(e-mail: of Tunis, University of Tunis El Manar, Tunis, Tunisia
(e-mail: [email protected])
**LR-OASIS, National Engineering School of Tunis, University of Tunis El Manar, Tunis, Tunisia
***LR-OASIS, National Engineering (e-mail: [email protected])
School of Tunis, University of Tunis El Manar, Tunis, Tunisia
(e-mail: [email protected])
***LR-OASIS, National Engineering School of Tunis, University of Tunis El Manar, Tunis, Tunisia
(e-mail:for
****Department of Engineering [email protected])
Innovation, University of Salento, Lecce, Italy
****Kent Business School, University
****Department of Engineering for Innovation, of Kent,University
Canterbury, United Kingdom
of Salento, Lecce, Italy
(e-mail:[email protected])
****Kent Business School, University of Kent, Canterbury, United Kingdom
(e-mail:[email protected])
Abstract: This paper proposes a simulation-based optimization model for solving a realistic delivery
problem within an urban area that respects the customers’ availability. Specifically, the originality of this
Abstract: This paper proposes a simulation-based optimization model for solving a realistic delivery
work is to consider the bimodal realistic traffic pattern when solving the classic capacitated vehicle routing
problem within an urban area that respects the customers’ availability. Specifically, the originality of this
problem with time windows. Computational results have shown that under bimodal travel time distribution
work is to consider the bimodal realistic traffic pattern when solving the classic capacitated vehicle routing
the routing solution changes considerately. Thus, this work emphasizes the need to fully integrate the traffic
problem with time windows. Computational results have shown that under bimodal travel time distribution
behavior when solving this class of routing problem with stochastic travel times. Furthermore, it highlights
the
the routing
easinesssolution
of suchchanges considerately.
integration, when theThus, thisproblem
routing work emphasizes
is modeled theand
needsolved
to fully integrate
using the traffic
this promising
behavior when solving this class of routing problem with stochastic travel times. Furthermore, it highlights
simulation-based optimization approach.
the easiness of such integration, when the routing problem is modeled and solved using this promising
simulation-based optimization
© 2022, IFAC (International approach.
Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Copyright
Copyright ©
© 2022
2022 The
The Authors.
Authors. This
This
© 2022, IFAC (International Federation of isis an open access
an open access
Automatic article
article
Control) underbythe
under
Hosting the CC BY-NC-ND
CC
ElsevierBY-NC-ND license
license
Ltd. All rights reserved.
(https://creativecommons.org
(https://creativecommons.org
Keywords: /licenses/by-nc-nd/4.0/)
/licenses/by-nc-nd/4.0/)
Urban Area; Stochastic Travel Time; Bimodal Distribution; Routing Problem; Simulation-based
Optimization.
Keywords: Urban Area; Stochastic Travel Time; Bimodal Distribution; Routing Problem; Simulation-based
Optimization.

Layeb et al. (2018) were the only and first to consider

1. INTRODUCTION
bimodal stochastic travel times when solving a freight
Layeb et al. (2018) were the only and first to consider
This work focuses 1.onINTRODUCTION
transportation problems that emerge transportation problem.
bimodal stochastic travel times when solving a freight
in urban traffic areas, where carriers face a complicated transportation
This work focuses on transportation problems that emerge
stochastic environment. In this context, travel time is In this work,problem.
we investigate the so-called stochastic
in urban traffic areas, where carriers face a complicated capacitated vehicle routing problem with hard time
strongly affected by traffic conditions and isalso seldom In this work, we investigate the so-called stochastic
stochastic environment. In this context, travel time is windows. It assumes that a carrier has to serve a set of
deterministic. Thus,for a delivery company solving the capacitated vehicle routing problem with hard time
strongly affected by traffic conditions and isalso seldom customers from a single depot with a homogeneous fleet,
routing problem with stochastic travel time becomes more windows. It assumes that a carrier has to serve a set of
deterministic. Thus,for a delivery company solving the i.e. all vehicles have the same capacity. Each customer has
challenging. Many recent works on route planning for smart customers from a single depot with a homogeneous fleet,
routing problem with stochastic travel time becomes more a demand and must be delivered within his desired time
cities have emerged during the last decade to address such i.e. all vehicles have the same capacity. Each customer has
challenging. Many recent works on route planning for smart window. In case of delay, the order is no longer accepted.
issues (e.g. Dündar et al. 2021). ). The objective of these a demand and must be delivered within his desired time
cities have emerged during the last decade to address such The carrier tries to find the best routing plan that minimizes
studies is to propose new solutions based on nowadays window. In case of delay, the order is no longer accepted.
issues (e.g. Dündar et al. 2021). ). The objective of these the total travel time. The objective of this study is to analyze
intelligent transportation technologies such as GPS, IoT, The carrier tries to find the best routing plan that minimizes
studies is to propose new solutions based on nowadays the effect of realistic bimodal travel time distribution on the
etc., for a more efficient routing under realistic traffic the total travel time. The objective of this study is to analyze
intelligent transportation technologies such as GPS, IoT, routing solution. Specifically, two stochastic travel time
conditions. For example, in the context of on-demand the effect of realistic bimodal travel time distribution on the
etc., for a more efficient routing under realistic traffic patterns are implemented: Bimodal-Gamma and the
grocery delivery, solving the vehicle routing problem while routing solution. Specifically, two stochastic travel time
conditions. For example, in the context of on-demand Bimodal Lognormal. In order to solve the corresponding
satisfying the desired customer time window is highly patterns are implemented: Bimodal-Gamma and the
grocery delivery, solving the vehicle routing problem while stochastic vehicle routing problem, we will recourse to a
challenging (Hammami et al. 2020). In fact, several studies Bimodal Lognormal. In order to solve the corresponding
satisfying the desired customer time window is highly Simulation-based Optimization (SBO) approach.
have shown that traffic in urban areas exhibit bimodal travel stochastic vehicle routing problem, we will recourse to a
challenging (Hammami et al. 2020). In fact, several studies According to Juan et al. (2019) this highly promising
time (Ji et al. 2015, Taylor et al. 2010).This bimodality in Simulation-based Optimization (SBO) approach.
have shown that traffic in urban areas exhibit bimodal travel method of coupling metaheuristic approach with a simulator
empirical travel time distributions is due to congestion, According to Juan et al. (2019) this highly promising
time (Ji et al. 2015, Taylor et al. 2010).This bimodality in is able to find optimal (or near-optimal) solution for
signal synchronization and several other factors as method of coupling metaheuristic approach with a simulator
empirical travel time distributions is due to congestion, stochastic combinatorial problems. Furthermore, as shown
discussed by Yazici et al. (2017). However, in the existing is able to find optimal (or near-optimal) solution for
signal synchronization and several other factors as in (Abideen et al. 2021) this simulation modeling approach
literature, the bimodal travel times has rarely been stochastic combinatorial problems. Furthermore, as shown
discussed by Yazici et al. (2017). However, in the existing allows providing a digital twin which is a stepping stone for
considered when solving stochastic Vehicle Routing in (Abideen et al. 2021) this simulation modeling approach
literature, the bimodal travel times has rarely been Logistics 4.0. In fact, the simulation software, such as Arena
Problems (VRP).In fact, Müller et al (2021) revealed that allows providing a digital twin which is a stepping stone for
considered when solving stochastic Vehicle Routing
2405-8963 (VRP).In© fact,
ProblemsCopyright 2022 The Authors.
Müller et al This is anrevealed
(2021) open access
that Logistics 4.0. In fact, the simulation software, such as Arena
article under the CC BY-NC-ND license.
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2022.09.380
134
142 Amina Antit et al. / IFAC PapersOnLine 55-10 (2022) 141–146
133–138

used in this work, had already set up a data-driven feed to 3. The MILP MODEL
the simulation model. Then, integrating traffic models in the
To validate the developed SBO model, finding the exact
proposed simulator is straightforward.
solution of the deterministic problem is crucial. Based on
the work of Pérez-Rodríguez and Hernández-Aguirre
The reminder of this paper isstructured as follows. Section
(2016), a MILP model was derived as follows. Let G= (V,A)
2 reviews the most related work. Section 3 presents the
be a full connected directed graph where V is the set of
Mixed Integer Linear Programming (MILP) model required
vertices represented all possible customers including depot,
to validate the proposed SBO model outlined in Section 4.
and A is the set of all possible routes between
Section 5 reports and discusses the conducted experimental
depot/customers.
study. Finally, Section 6 draws some concluding remarks
andpossible future research avenues.
Sets:
2. RELATED WORK N Set of all customers, denoted by j
D Depot node, denoted by w
Integrating time uncertainty in this problem helps L Set of trucks, denoted by l
formalizing a more realistic problem, leading to more V Set of all nodes N⋃D, indexed by i
efficient solutions in real conditions. In the following, we
discuss the most relevant works considering a VRP with Parameters:
stochastic travel time, and all existing works that address
Travel time traversing arc (i,j) ⋴N
dj Demand of customer j⋴N
the VRP with the SBO approach.
TWimin Earliest feasible time to serve customer i ∈V or to
TTij

return to depot i ∈V
Taş et al (2013) provided a tabu search method to solve the
TWimax Latest feasible time to serve customer i ∈V or to
VRP with stochastic travel time following a gamma

return to depot i ∈V
distribution.Three-phase solution algorithm was defined:
starting with constructing an initial solution, moving to
improving the devolved initial solution, and ending with M Big positive number
unifying a post-optimization procedure. Miranda and
Conceição (2016) applied an iterated local search algorithm Decision variables:

vehicle l⋴L is moving from i to node j,(i,j) ⋴A

in order to estimate the arrival times at the customers and xijl Binary decision variable which takes value 1 if
the probability of respecting the time windows. The
considered travel times and the service times are stochastic tjl Continuous non negative variable corresponding to
and both follow the normal distribution.Lombard et al. the arrival time of vehicle l⋴L to node j⋴N
(2018) addressed a new variant of the VRP with random
delivery locations and stochastic travel times and thus the Using these notations, a MILP formulation reads as:
distance between two points is non-deterministic. The travel Minimize ∑𝑙𝑙𝑙𝑙∈𝐿𝐿𝐿𝐿 ∑(𝑖𝑖𝑖𝑖,𝑖𝑖𝑖𝑖)∈𝐴𝐴𝐴𝐴 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖 (1)
time follows the symmetrical triangular distribution. Oyola
(2018) focused on the VRP with stochastic travel time and The objective function (1) minimizes the total routing time
soft time windows. The travel time follows the shifted subject to the following constraints:
gamma distribution.The problem has been solved through
the combination of a genetic algorithm and a neighborhood ∑ 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖 = 1, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉, ∀𝑗𝑗𝑗𝑗 ∈ 𝑁𝑁𝑁𝑁: (𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴 (2)
𝑙𝑙𝑙𝑙∈𝐿𝐿𝐿𝐿
search heuristic. Li and Li (2020) addressed a stochastic
VRP with soft time windows. The service time as well as
Constraints (2) express the condition that all customers are
the travel time are subject of stochasticity and follow the
served exactly once.
normal distribution. The authors have proposed a resolution
algorithm using an improved tabu search algorithm based ∑ 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖 − ∑ 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙 = 0, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉, 𝑙𝑙𝑙𝑙 ∈ 𝐿𝐿𝐿𝐿: (𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴 (3)
on a greedy algorithm. Recently, Rajabi-Bahaabadi et al. 𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉
(2021) proposed a hybridized min-max ant colony with tabu
Constraints (3) are the flow balance equations for each node
search algorithm to solve the VRP with flexible time
and each vehicle.
∑ ∑ 𝑑𝑑𝑑𝑑𝑖𝑖𝑖𝑖 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖 ≤ 𝑄𝑄𝑄𝑄, ∀ 𝑙𝑙𝑙𝑙 ∈ 𝐿𝐿𝐿𝐿 ∶ (𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴
windows. The travel times are considered correlated and
stochastic following the Lognormaldistribution.
𝑖𝑖𝑖𝑖∈𝑉𝑉𝑉𝑉 𝑖𝑖𝑖𝑖∈𝑁𝑁𝑁𝑁
(4)
As far as SBO is concerned, only two works were interested
in solving the VRP, without taking into account the real-
Constraints (4) ensure that the capacity of each vehicle is
world stochastic aspects. Mazzuco et al. (2018) have not exceeded.
∑ ∑ 𝑥𝑥𝑥𝑥𝑤𝑤𝑤𝑤𝑖𝑖𝑖𝑖
𝑙𝑙𝑙𝑙
≤ 1, ∀𝑙𝑙𝑙𝑙 ∈ 𝐿𝐿𝐿𝐿: (𝑤𝑤𝑤𝑤, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴
employed the simulation approach for the VRP to optimize
freight deliveries by minimizing the total transport cost and (5)
choosing the optimal routes. Vonolfen et al. (2010) have 𝑤𝑤𝑤𝑤∈𝐷𝐷𝐷𝐷 𝑖𝑖𝑖𝑖∈𝑁𝑁𝑁𝑁
implemented a scenario-based simulation approach. The Constraints (5) indicate that each truck can only be used at

𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙 ≥ 𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙 + 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 − 𝑀𝑀𝑀𝑀(1 − 𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖 ), ∀𝑖𝑖𝑖𝑖 ∈ 𝑁𝑁𝑁𝑁, 𝑗𝑗𝑗𝑗 ∈ 𝑉𝑉𝑉𝑉,

optimization is based on the combination of both heuristic most once.

𝑙𝑙𝑙𝑙 ∈ 𝐿𝐿𝐿𝐿 ∶ (𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴

and metaheuristic based-optimizers using permutation
(6)
coding.
 Amina Antit et al. / IFAC PapersOnLine 55-10 (2022) 141–146
133–138 135
143

Constraints (6) express the arrival time to each node and a new better solution, i.e. the configuration to evaluate by
ensure elimination of sub-tours. the simulator.
𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙 ≥ 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖𝑚𝑚𝑚𝑚 , ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉 (7)
𝑡𝑡𝑡𝑡𝑗𝑗𝑗𝑗𝑙𝑙𝑙𝑙 ≤ 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 , ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉
The proposed SBO model is implemented on the
(8) commercial software Arena which is one of the most
deployed software in supply chain management (Oliveira et
of customer or depot j.
Constraints (7)-(8) express the earliest (latest) time to visit
al. 2016). This integrates a powerfuloptimization tool to
𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖 ≥ 0, ∀𝑖𝑖𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉 (9) carry out the optimization function, called Optquest. The
𝑥𝑥𝑥𝑥𝑖𝑖𝑖𝑖𝑙𝑙𝑙𝑙𝑗𝑗𝑗𝑗 ∈ {0,1}, ∀𝑙𝑙𝑙𝑙 ∈ 𝐿𝐿𝐿𝐿, ∀𝑤𝑤𝑤𝑤 ∈ 𝐷𝐷𝐷𝐷 ∶ (𝑖𝑖𝑖𝑖, 𝑗𝑗𝑗𝑗) ∈ 𝐴𝐴𝐴𝐴
search algorithm applied by Optquest is based on a
(10) combination of three metaheuristics: tabu search, neural
Constraints (9)-(10) state the nature of variables 𝑥𝑥𝑥𝑥 and 𝑡𝑡𝑡𝑡. networks and dispersion search (Glover et al. 2003).
Precisely, the SBO approach is used in this study to identify
4. THE SIMULATION-BASED OPTIMIZATION routes allowing all customers to be delivered in an uncertain
MODEL environment while minimizing the expected total transport
time. The transport time on each arc (lane connecting two
The SBO, also known as black-box optimization, consists customers) is assumed to be stochastic and follows a
of coupling a simulation model with an optimization tool. bimodal distribution.
The simulator acts as an evaluator of the configurations
(input parameters) sent by the optimizer. The cost of the
solution (response) is then sent back to the optimizer to find

Figure 1. The CVRP-HTW model

In order to embed the simulation model, we used the Arena stochastic context, where travel time of each arc follows
software. Specifically, as shown in Figure 1, we have unimodal and bimodal distributions. For that purpose, we
employed the modules CREATE, ASSIGN, VARIABLE, modified the deterministic data sets provided by Solomon
ATTRIBUTE and DECIDE. These modules enable us to (1987). Specifically, three different cases are taken from
reproduce the same process of evaluation. For more details R101 instances. The number of customers is varying
on Arena models, the interested readeris referred to the between 5 and 10. In order to analyze the effect of the
work of Kelton et al. (2007). Once the simulator is different bimodal distributions as well as to compare the
developped, the next step is to couple it with the Optquest results with the effect of unimodal distribution, these chosen
optimizer. The optimizer will then start the process of Solomon’s data sets were extended by considering
searching for a new/ better route to send to the simulator for deterministic travel times as the expected values of random
evaluation. This iterative process aims at finding the lowest travel times following the three probability distributions:
total travel time. bimodal Gamma distribution, bimodal Lognormal
distribution and the unimodal Lognormal distribution. The
5. COMPUTATIONAL RESULTS AND ANALYSIS
same method of transforming data from deterministic to
In order to validate the proposed SBO model, the stochastic data set was applied by Layeb et al.(2018) and
deterministic version of the problem was firstly assessed Juan et al. (2019).
using the MILP model (1)—(10) solved by the state-of-the-
art solver IBM ILOG CPLEX (version 12.9). Then, once An example of a histogram of travel time following bimodal
validated, the CVRP-HTW model is implemented in a Gamma distribution is given in Figure 3. We first assume
136
144 Amina Antit et al. / IFAC PapersOnLine 55-10 (2022) 141–146
133–138

𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉[𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘 ]
that travel times are fitted with the bimodal Gamma
σk,ij = |√(1 + ln ( )) |
E [𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘 ]2
distribution according to (16)
TTij∼ p1Gamma(α1,ij ,β1,ij )+ p2 Gamma (α2,ij ,
β2,ij),
(11) The values of μk,ij , and σk,ij are estimated using the method
where p1is the probability of encountering the first mode, of moment. Besides, the unimodal Lognormal distribution
is applied to the deterministic travel times.
the corresponding Gamma (αk,ij, βk,ij) mode, k=1,2, (i,j) ∈A,
αk,ij is the shape parameter, and βk,ij is the scale parameter of
We had also assumed3 levels of Coefficient of Variation
where A is the set of arcs.These parameters are respectively (CV), namely CV=10%, 40% and 70%.Recall that the CV
expressed as value is the standard deviation divided by the mean. The
𝐸𝐸𝐸𝐸[𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘,𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ]2
k= 1,2, (i,j) ∈A,
objective herein is to reproduce a realistic level of
αk,ij = ,
Var [𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘,𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ]
(12) variability encountered in empirical study in urban areas (Ji

𝐸𝐸𝐸𝐸[𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘,𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ]2
et al. 2015, Yazici et al. 2017).
βk,ij = , k= 1,2,(i,j) ∈A,
Var [𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘,𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ]
(13) 5.1. Results under different stochastic travel time
withVar [𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘𝑘𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ]= (CV 𝐸𝐸𝐸𝐸𝐸𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘𝑘𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 ])2 , k=1,2, (i,j)∈A.
distributions

Parameters αk,ij and βk,ij are estimated using the method of At first, in order to validate our SBO solving approach we
moment. use it to solve the deterministic instances. The total travel
time obtained with 5, 8 and 10 customers is 156.35, 216.84,
and 269.54 respectively.
Then, we solve the CVRP-HTW model while considering
the lastly discussed different stochasticity of the travel
times. The results are summarized in Table 2.

Table 2 shows the costs calculated herein as the estimated

total travel time for the three instances, namely, for 5, 8 and
Figure 2. Histogram of travel time following bimodal Gamma 10 customers. Since we are using stochastic simulation
distribution model, this solution cost, i.e. the total travel time output
performance is estimated as the sample average over n
As illustrated in Figure 2, the first mode represents travel independent replications. We fixed n such that, at a 95%
time in a congested condition. The second mode represents confidence level, the half-width of the confidence interval
free-flow movement. In the first mode, the average travel is less than 0.05. We observe that, for each instance, an
time is increased by 40%. It is assumed that the first mode increase in the CV implies an increase in the total travelled
is met with a probability p1 = 70% against a probability of time. In fact, the increase in travel time variability augment
p2 = 30% for the second mode. the risk of missing the time window and then the search for
Next, in order to investigate the effect of the bimodal another solution with higher cost but less risk of missing the
Lognormal distribution, the following transformation are desired delivery time window.
applied to the deterministic travel times:

TTij∼p1Lognormal(μ1,ij,σ 1,ij)+ p2Lognormal(μ2,ij,σ 2,ij),

From Table 2, we also depict that by increasing the value of
(14)
the CV as well as the number of clients, the results
where these parameters are expressed as generated with travel times following the bimodal gamma

1 𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉[𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘 ]
distribution differ considerably from the results generated

μk,ij = ln(𝐸𝐸𝐸𝐸[𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘 ]) − ln(1 + )

with the bimodal lognormal distribution.
2 E [𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑘𝑘𝑘𝑘 ]2
(15)

Table 2. Solutions for the stochastic case

Total Travel Time

Distribution #Customers
CV=10% CV=40% CV=70%
5 163.85 195.83 201.58
Lognormal 8 264.37 505.10 570.56
10 392.39 441.08 471.57
5 165.05 178.34 205.59
Bimodal
8 276.87 307.08 462.32
Gamma
10 422.02 453.46 551.68
5 165.00 178.46 196.46
Bimodal
8 276.82 286.06 453.95
Lognormal
10 422.07 440.98 507.02
 Amina Antit et al. / IFAC PapersOnLine 55-10 (2022) 141–146
133–138 137
145

For example, in the case of 10 clients, with a CV equal to Now, let’s turn our attention the corresponding stochastic
70% and the travel time follows the bimodal gamma case. Figure 3 shows the best routing solution found by the
distribution, thetotal travel time is equal to 551.68 minutes, SBO in the bimodal Gamma case, for the three CV levels
whereas the total travel time obtained with the bimodal 10%, 40% and 70%.
lognormal distribution is 507.02 minutes. That means that As shown in Figure 3, the routing plan varies from case to
the nature of bimodality has a considerable impact on the case. In fact, when switching from the deterministic to the
solution. stochastic case with CV equal to10%, customer 3 is no
longer delivered with customer 5, the latter must be
Also, it is worth mentioning that, given the hard time delivered with customers 4 and 2. Moving to the case with
window constraint, when the number of customers CV = 40%, customer 3 must be delivered alone. Then, by
increases, the probability of obtaining an infeasible solution raising the CV to 70% customer 3 has to be delivered just
increases. Indeed, the feasibility of the solution depends on before customer 7.A given route cannot be considered valid
the arrival time of the delivery to customer. If the arrival at all times. The variability of the traffic flow may lead to
time of at least one customer does not belong to the time the deterioration of the proposed solution under different
window, the solution is considered infeasible. In the sequel level of stochasticity.
we illustrate this change on the routing plan.
Furthermore, we can see in Figure 4 that the routing solution
5.2. Impact on the routing solution when travel times follow a bimodal lognormal distributions
Let’s consider the case of 8 customers indexed from 2 to 9. differs from the optimal routing plan found under the
1 is the index of the depot. The 4 vehicle routes obtained in assumption of bimodal gamma distribution under CV=40%.
a deterministic context combine the following customers (3,
5), (4, 2), (6, 7) and (8, 9) in that specific order.

Figure 3. Solution for 8 customers with CV= 10%, 40%, and 70% (Bimodal Gamma Distribution)

proposes Read/Write blocks to interact in real-time with

external applications.
6. CONCLUSION
This paper presents an SBO model for carriers operating in
urban areas. Computational Experiments were conducted to
investigate the effect of realistic bimodal travel times on the
routing solution, due to traffic specification in urban areas such
as bimodal gamma and bimodal lognormal distributions.
Figure 4. Solution for 8 customers with bimodal Lognormal Results revealed that the solution is no more reliable under
distribution and CV=40%
different variability and stochastic behavior.
Thus, the reliability of the solution considerably deteriorated
when applying a different bimodal distribution. Then, when This study highlights the weakness of decoupling the solving
solving such vehicle routing problem, a company should of vehicle routing problems from the real traffic behavior, and
integrate the real empirical data available nowadays through emphasizes the importance of collecting traffic data through
the new technologies. For that purpose, recent works on traffic new technologies to enhance solution reliability. One of the
reliability have proposed new methods such as Google Maps strengths of the our SBO tool is the easiness to integrate on the
Distance Matrix API and Global Positioning System proposed simulation model the collected data from real-world
(Dumbliauskas et al., 2017). Using such API with the herein traffic data. Further contributions may consist in investigating
proposed SBO method could greatly enhance the routing the behavior of our proposed SBO model when solving large-
solution. Integrating these traffic models in the proposed scale instances using Google Maps Distance Matrix API and
Arena simulator is straightforward since this software already also incorporating the effect of delays on the model (Leggieri
et al., 2007).
138
146 Amina Antit et al. / IFAC PapersOnLine 55-10 (2022) 141–146
133–138

REFERENCES and stochastic travel times (VRPRDL-S).

Transportation research procedia, 30, 167-177.
Abideen, A. Z., Sundram, V. P. K., Pyeman, J., Othman, A.
K., and Sorooshian, S. (2021). Digital twin integrated Mazzuco, D. E., Danielli, A. C., Oliveira, D. L., Santos, P. P.
reinforced learning in supply chain and logistics. P., Pereira, M. M., Coelho, L. C., and Frazzon, E. M.
Logistics, 5(4), 84. (2018). A concept for simulation-based optimization in
Vehicle Routing Problems. IFAC-Papers OnLine,
Dumbliauskas, V., Grigonis, V., and Barauskas, A. (2017). 51(11), 1720-1725.
Application of Google-based data for travel time
analysis: Kaunas city case study. Promet-Traffic & Miranda, D. M., and Conceição, S. V. (2016).The vehicle
Transportation, 29(6), 613-621. routing problem with hard time windows and stochastic
travel and service
Dündar, H., Ömürgönülşen, M., and Soysal, M. (2021).A time.ExpertSystemswithApplications,64, 104-116.
review on sustainable urban vehicle routing.Journal of
Cleaner Production, 285, 125444 Müller, J. P., Elbert, R., and Emde, S. (2021). Integrating
vehicle routinginto intermodal service network design
Glover, F., Laguna, M., and Marti, R. (2003). Scatter search with stochastic transit times. EURO Journal on
and path relinking: Advances and applications. In Transportation and Logistics,10, 100046.
Handbook of metaheuristics (pp. 1-35).Springer, Boston,
MA. Oliveira, J. B., Lima, R. S., and Montevechi, J. A. B. (2016).
Perspectives and relationships in Supply Chain
Hammami, N. E. H., Layeb, S. B., and Jaoua, A. (2020, Simulation: A systematic literature review. Simulation
November).On Solving the on-Demand Grocery Modelling Practice and Theory,62, 166-191.
Delivery Problem. In 2020 International Conference on
Decision Aid Sciences and Application (DASA) (pp. 284- Oyola, J. (2019). The capacitated vehicle routing problem with
287). IEEE. soft time windows and stochastic travel times. Revista
Facultad de Ingeniería,28(50), 19-33.
Ji, Y., Jiang, S., Du, Y., and Zhang, H. M. (2015). Estimation
of bimodal urban link travel time distribution and its Pérez-Rodríguez, R., and Hernández-Aguirre, A. (2016).
applications in traffic analysis. Mathematical Problems Simulation optimization for the vehicle routing problem
in Engineering, 2015. with time windows using a Bayesian network as a
probability model. The International J. of Advanced
Juan, A. A., Panadero, J., Reyes-Rubiano, L., Faulin, J., de la Manufacturing Technology, 85(9), 2505-2523.
Torre, R., and Latorre, I. (2019). Simulation-based
optimization in transportation and logistics: Comparing Rajabi-Bahaabadi, M., Shariat-Mohaymany, A., Babaei, M.,
sample average approximation with simheuristics. In and Vigo, D. (2021). Reliable vehicle routing problem in
Winter Simulation Conference, pp. 1906-1917. IEEE. stochastic networks with correlated travel times.
Operational Research, 21(1), 299-330.
Kelton, W. D., Sadowski, R. P., andSadowski, D. A. (2007).
Simulation with Arena, McGraw-Hill. New York. Solomon, M. M. (1987). Algorithms for the vehicle routing
and scheduling problems with time window constraints.
Leggieri, V., Haouari, M., Layeb, S., and Triki, C. (2007). The Operations research, 35(2), 254–265.
Steiner tree problem with delays: A tight compact Taş, D., Dellaert, N., Van Woensel, T., and De Kok, T. (2013).
formulation and reduction procedures. Technical report, Vehicle routing problem with stochastic travel times
University of Salento, Lecce. including soft time windows and service costs.
Computers & OperationsResearch,40(1), 214-224.
Laporte, G., andOsman, I. H. (1995). Routing problems: A
bibliography. A. of operations research,61(1), 227-262. Taylor, M. A., and Somenahalli, S. (2010). Travel time
reliability and the bimodal travel time distribution for an
Layeb, S. B., Jaoua, A., Jbira, A., and Makhlouf, Y. (2018). A arterial road. Road & Transport Research A Journal of
simulation-optimization approach for scheduling in Australian and New Zealand, 19(4), 37-50.
stochastic freight transportation. Computers & Industrial
Engineering, 126, 99-110. Vonolfen, S., Wagner, S., Beham, A., Kofler, M., Affenzeller,
M., Lengauer, E., and Scheucher, M. (2010). A
Li, G., and Li, J. (2020). An improved tabu search algorithm simulation-based approach to the vehicle routing
for the stochastic vehicle routing problem with soft time problem. In 22nd European modeling and simulation
windows.IEEE Access,8, 158115-158124. symposium EMSS, pp. 363-368.

Lombard, A., Tamayo-Giraldo, S., and Fontane, F. (2018). Yazici, A., Ozguven, E. E., and Kocatepe, A. (2017). Urban
Vehicle routing problem with roaming delivery locations travel time variability in New York City: A spatio-
temporal analysis within congestion pricing context.