Archives of Computational Methods in Engineering (2022) 29:195–221

https://doi.org/10.1007/s11831-021-09574-x

ORIGINAL PAPER

Review of Vehicle Routing Problems: Models, Classification

and Solving Algorithms
Haifei Zhang1,2 · Hongwei Ge1,2 · Jinlong Yang1,2 · Yubing Tong3

Received: 25 August 2020 / Accepted: 13 March 2021 / Published online: 19 April 2021
© CIMNE, Barcelona, Spain 2021

Abstract
With the rapid development of logistics industry, vehicle scheduling is the key to the optimization of transportation links.
Vehicle transportation route planning is becoming more and more important to reduce logistics costs. In recent decades, the
research on VRP and related derivative problems has become more and more extensive. Based on the basic VRP, this paper
classifies VRP according to its characteristics and practical application. It focuses on the analysis of VRP with capacity
constraint, VRP with time window, VRP with demand splitting and dynamic VRP, and gives the unified description and
mathematical model of each type of problem, and then analyzes the solution methods of each type of VRP Finally, combined
with other types of VRP, the future research and development trend of VRP are given.

1 Introduction loading and delivery. Under the development of intensive

and integrated logistics distribution, we often integrate all
Logistics industry is developing rapidly in the direction of aspects of distribution, and the core part is the collection,
global integration and operation informatization. Logistics loading and delivery process of distribution vehicles. When
distribution is an important link connecting producers and we start to optimize the logistics distribution system, we
consumers, and plays an increasingly important role in the mainly focus on the key technologies of specific links.
whole supply chain. With the development of social econ- To optimize the distribution system, the main task is to
omy and the continuous improvement of people’s demand, optimize the scheduling of distribution vehicles, including
logistics distribution gradually has the characteristics of freight collection lines, goods distribution lines, and the
multi variety, less batch, real-time, personalized and so on. integration of goods collection, cargo loading and delivery.
In order to meet the requirements of the times, the whole In particular, the optimization of goods distribution line is
society has put forward higher goals for logistics distribu- a key link in logistics distribution optimization. Whether
tion. One of the system’s formulation is to require the logis- the distribution line is reasonable or not has a great impact
tics distribution to achieve “7R”, which is: right product, on the speed, time, cost loss and benefit of distribution,
right quantity, right time, right place, right condition, right especially the multi-user distribution line is more complex.
customer, right cost. Using scientific and reasonable methods to optimize the
As an important part of transportation and production distribution vehicle route is a very important activity in dis-
process, logistics distribution is generally considered to be tribution activities.
the core part including the process of goods preparation, Raff [1] pointed out vehicle routing problem (VRP) stud-
ies how to arrange the route of vehicles to transport goods
from the depot to multiple geographically dispersed cus-
* Hongwei Ge
[email protected] tomer points or transport goods back to the depot under
certain constraints. With the rapid development of logistics
1
Key Laboratory of Advanced Process Control for Light industry, vehicle transportation route planning is becoming
Industry (Jiangnan University), Ministry of Education, more and more important to reduce logistics costs.
Wuxi 214122, China
In recent decades, the research on VRP and related deriv-
2
School of Artificial Intelligence and Computer Science, ative problems has become more and more extensive, the
Jiangnan University, Wuxi 214122, Jiangsu, China
related research on VRP and related derivative problems
3
Medical Image Processing Group, Department of Radiology, mainly includes. Solomon [2] reviewed the vehicle routing
University of Pennsylvania, Philadelphia, PA 19104, USA

13
Vol.:(0123456789)
196 H. Zhang et al.

and scheduling problem with time window constraints.

Laporte [3] reviewed the exact and approximate algorithms
for vehicle routing problem. El-Sherbeny [4] reviewed three
kinds of algorithms for vehicle routing problem with time
windows. Psaraftis et al. [5] analyzed the impact of the new
generation of information technology on DVRP and sum-
marized the existing research literature according to 11 Depot
classification standards. These reviews mainly focus on the
analysis of a certain type of VRP and its solution algorithm,
but less attention is given to the derivation and classification
of VRP, and the unified analysis of mathematical model for
each class of problems. Based on the basic VRP, this paper
classifies VRP according to its characteristics and practi-
cal application, and analyzes VRP with capacity constraint,
VRP with time window, VRP with demand splitting and Fig. 1  VRP diagram
dynamic VRP. The unified description and mathematical
model of each type of VRP are given, and then the solu-
tion methods of each type of VRP are analyzed. Finally, the 1. Road: it is usually represented by the network formed
future research and development trend of VRP are given by starting from the depot, passing through the customer
combined with other types of VRP. points in a certain route, and finally returning to the
depot.
2. Customer point: represents the service object.
3. Depot: the starting or ending point of each vehicle route.
2 Basic Vehicle Routing Problem 4. Vehicle: complete the distribution activities from the
depot to the customer point.
Vehicle routing problem (VRP) was first proposed by [6], 5. Transportation arrangement requirements: the route of
which was used to study the routing optimization problem vehicles often depends on the nature of the goods trans-
of oil tankers between Atlanta refinery and its subordinate ported, the level of service quality, and the characteris-
gas stations. It refers to a certain number of customers with tics of customers and vehicles,
different demand for goods. The depot provides goods to 6. Objectives: ① to minimize the total transportation cost:
customers, and a fleet is responsible for distributing goods its size depends on the number of vehicles required to
and organizing appropriate driving routes. The goal is to serve all customers, the fixed and variable costs of each
meet the needs of customers and achieve the goals of short- vehicle; ② the shortest transportation distance; ③ the
est distance, minimum cost and minimum time consumption minimum number of vehicles required to complete dis-
under certain constraints. tribution; ④ the shortest running time of empty vehicles;
The great value of vehicle routing and economy has been ⑤ the penalty value related to incomplete service of cus-
widely concerned by scholars at home and abroad. tomers.
In general, VRP can be described as follows: under the
constraint conditions, the optimal routing route for delivery 2.1 Problem Description
from one or more initial points to multiple cities or customer
points in different locations is designed, as shown in Fig. 1: There is a depot to deliver goods to m customer points, the
That is to design a route set with minimum total cost: demand of the i -th customer point is gi (i = 1, 2, 3 … , m), the
distribution center sends n vehicles with a load capacity of
1. Each city or customer is visited only once by one vehi- Cn to transport the goods to each customer point, and finally
cle; returns to the depot. It is known that gi ≤ Cn , it is required
2. All vehicles start from the starting point and return to to arrange the vehicle route reasonably to meet the needs of
the starting point after completing the distribution task; each customer point, and the journey is the shortest.
3. Some constraints are satisfied. The most common con- The constraints mainly include:
straints include capacity constraints, time window con-
straints, etc. 1. The total amount of freight carried by each vehicle shall
not exceed the maximum carrying capacity of the vehi-
VRP is mainly one or more combinations of the follow- cle;
ing factors: 2. Each customer site can only be accessed by one vehicle;

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 197

3. The length of each distribution route does not exceed the each vehicle must not exceed the maximum load capacity
maximum distance for the car to be distributed at one of the vehicle;
time; The formula (5) ensures that each customer point can only
4. Priority relationship between multiple customer points. be completed by one vehicle, and all transportation tasks are
For example, you must visit point b before visiting point coordinated by n vehicles.
a.
2.2 Common Solutions of VRP
Let the number of the depot be 0, the number of each
customer point is i(i = 1, 2, … , m), and the number of each In the past, the solution methods of vehicle routing prob-
vehicle is s(s = 1, 2, … , n). The goal of building a model lems can be divided into exact algorithms and heuristics.
is to minimize the total transportation cost. In general, the Among them, precision algorithms include branch and
transportation cost is proportional to the vehicle’s travel bound methods, branch cutting methods, and set coverage
path. The shorter the travel path, the lower the fuel consump- methods; heuristic solutions include saving method, simu-
tion of the vehicle, the less the driver’s working time, and the lated annealing method, deterministic annealing method,
lower the total transportation cost. Therefore, the following tabu search method, genetic algorithm, neural network, ant
mathematical model is established with the shortest driving colonization algorithm, etc.
path as the objective function: On the basis of [7], the algorithm for solving vehicle rout-
m m ing problem is divided into three stages:
∑ ∑
cij xijs The first stage is from 1960 to 1970, which belongs to a
J = min (1)
i=0 j=0 simple heuristic method, including various local improve-
ment heuristic algorithms and greedy methods (Greedy),
Equation (1) is the objective function, cij represents the etc.; the second stage is from 1970 to 1980, which belongs
transportation cost from the customer point i to the customer to a mathematical Planning-based heuristic solution, includ-
point j , and xijs indicates that the vehicle s is driving from ing assignment method, set segmentation method and set
point i to point j . When the event occurs, the value is 1, coverage method; the third stage is from 1990 to the present,
otherwise it is 0; which is a newer method, including the use of rigorous heu-
m ristic method, artificial intelligence method, etc.
∑
xijs = ysj0 , j = 1, 2, … , m;∀s. (2) Classify the common solutions of VRP, as shown in
i=0 Fig. 2.

m
∑ 2.3 Classification of VRP
xijs = ys0i , i = 1, 2, … , m;∀s. (3)
j=0
Based on the basic vehicle routing problem (VRP), the vehi-
cle routing problem has produced many different extensions
and changes in academic research and practical applications,
m
∑
gi ys0i ≤ qs , ∀s. (4) including:
i=0

n
{ 1. Capacitated Vehicle Routing Problem, CVRP
∑ 1 i = 1, 2, … , m 2. Vehicle Routing Problem with Time Window, VRPTW
ys0i = . (5)
s=1 n i=0 3. Split Delivery Vehicle Routing Problem, SDVRP
4. Dynamic Vehicle Routing Problem, DVRP
In Eqs. (2) and (3), ys0i indicates that the freight task at 5. Vehicle Routing Problem with Stochastic Demand,
the customer site i is completed by the vehicle s. The value VRPSD
is 1 when the event occurs, otherwise it is 0. Putting the 6. Vehicle Routing Problem with Simultaneous Delivery-
two together means that for any customer point j served by Pickup, VRPSDP
vehicle s, there must be one and only one (different from j ) 7. Open Vehicle Routing Problem, OVRP
customer point i (including distribution center) served by 8. Green Vehicle Routing Problem, GVRP
vehicle s, vehicle s arrives at customer point j from customer 9. Vehicle Trip Routing Problem, MTVRP
point i , the customer point i served by the vehicle s also
has another customer point served by the vehicle s, and the As a basic problem in the field of logistics and distribu-
vehicle s arrives from the customer point i ; tion, VRP has been widely studied by scholars. Accord-
In the formula of (4), qs represents the maximum load ing to different influence conditions, there are a series of
capacity of the vehicle s , and the total mixing amount of derivative types of vehicle routing problems. According

13
198 H. Zhang et al.

Fig. 2  Algorithm classification Branch and bound approach

of VRP
Cutting planes approach

Exact
algorithm Network flow approach Two-stage method

Scanning algorithm
Dynamic programming approach
C-W saving algorithm

Nearest neighbor method

VRP
algorithm Construct heuristic Recently inserted method

k-opt algorithm
Heuristic Improved heuristics ¬ -interchange algorithm
algorithm
Tabu search algorithm

Sub-heuristic Genetic algorithm

Neural network algorithm

Simulated annealing algorithm

Particle Swarm Optimization

Ant colony algorithm

to different conditions, there are many types of VRP that 3.1 CVRP Description and Mathematical Model
can be derived, limited to research trends and space, and
will not be repeated here. Under the premise of knowing the location of the depot
and the customer, the customer’s demand, the number of
vehicles and the maximum load, the vehicle distribution
route is designed to meet the needs of all customers, and
3 VRP with Capacity Constraints the transportation cost is minimized, that is, the total cost
is minimized (the total distance traveled is the shortest or
CVRP is based on VRP and adds vehicle load constraints. takes the least time).
Due to the addition of load constraints, the optimization The route diagram of CVRP model is shown in Fig. 3.
process needs to comprehensively consider the mutual Expressed by a mathematical model, CVRP can be
coupling relationship of the number of vehicles, cargo described as the problem that n vehicles from the distri-
distribution and path planning to achieve the shortest total bution center 0 serve m customers. Let gi be the demand
travel distance. When reflected in the specific solution, of customer i , the transportation cost from customer i to
there are usually two types of problems: customer j , and bs be the load of the vehicle s.
{
Constraints limit the generation of new solutions for 1, vehicle s serves customer i followed by customer j;
xijs =
optimized objects, reduce the algorithm’s global search 0, else.
capability, and make it easy to fall into local optimal
solutions. And as the types and strengths of constraints The goal of CVRP is to find the optimal transportation
increase, this effect gradually strengthens; path for the fleet, so that the transportation cost J reaches
For the fusion of constraints and VRP problems, it is the minimum value, that is:
difficult to achieve the overall coordination and effi-
ciency of the algorithm in terms of structure, computa-
tional complexity and solution accuracy.

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 199

Customer7
Customer7
Customer5
Customer5 Customer1
Customer1

Rout2
Customer6
Customer6
Depot Customer4
Depot Customer4
Rout1

Customer2
Customer2
Customer3
Customer3

Fig. 3  CVRP problem example

n
∑ m
∑ in evolutionary algorithm, ant colony algorithm, the basic
J = min cij xijs (6) idea when applying it to solving optimization problems
s=1 i,j=0
is: the ant’s walking path represents the feasible solution
of the problem to be optimized, and all paths of the entire
m m
∑
s
∑
s ant group constitute the solution space of the problem to
x0j = xj0 = 1, s ∈ (1, 2, … , n) (7) be optimized. The ants with a shorter path release more
j=1 j=1
pheromone. As time progresses, the concentration of
n m pheromones accumulated on the shorter path gradually
∑ ∑
xijs = 1, i ∈ (1, 2, … , m) increases, and the number of ants who choose the path
(8) increases. Eventually, the entire ant will concentrate on
s=1 j=0
j≠i the best path under the action of positive feedback. At this
time, the corresponding solution is the optimal solution of
n m
∑ ∑ the problem to be optimized.
xijs = 1, j ∈ (1, 2, … , m) Ant colony algorithm can determine a better solution in
(9)
s=1 i=0
i≠j a suitable time for complex combinatorial problems, but it
is limited to small-scale problems. In order to improve the
m
∑ m
∑ performance of ant colony algorithm, Dorigo and Colorni
gi xijs ≤ bs , s ∈ (1, 2, … , n) [8] proposed ant colony system (ACS) algorithm.
(10)
i=1 j=0 Bullnheimer and Strauss [9] applied ant colony algorithm
i≠j
to CVRP for the first time, and an improved ant colony algo-
Equation (7) indicates that the vehicle must start from rithm based on ant system (AS) is proposed, which takes the
the depot and return to the depot after completing the task; path savings between customers as heuristic information,
Eqs. (8) and (9) indicate that each customer is visited once and uses 2-opt to search the optimal solution after each itera-
and can only be accessed once; Eq. (10) indicates that each tion. Bullnheimer et al. [10] is improved on the basis of [9],
vehicle’s service to customers does not exceed its own load. and the candidate list is added in each search process, which
accelerates the convergence speed of the optimal solution.
3.2 Research Progress of CVRP Belf and Mcmullen [11] and Gajpal and Abad [12] pro-
posed a multi-ant colony method to solve CVRP based on
Capacity constraints generally exist in various studies as ant system and ant colony system (ACS), respectively. Chen
a basic constraint of vehicle routing problems. There are and Ting [13] used an improved ant colony system to solve
currently many methods for solving CVRP, which can be CVRP. Yu et al. [14] proposed an improved Ant Colony
roughly divided into two types of algorithms: precise algo- Optimization (IACO) with strong local search capability by
rithms and heuristic algorithms. updating the pheromone accumulation strategy of the ant
In practical application to solve CVRP related problems, colony algorithm.
heuristic algorithm is mainly used to solve. In addition to the ant colony algorithm, other intelligent
Ant colony algorithm (ACO) is a kind of simulated algorithms have also shown good results in solving CVRP.
optimization algorithm which simulates ants’ foraging Juan et al. [15] proposed a hybrid method based on SR-
behavior. It was first proposed by Dorigo et al. and used to GCWS (simulation in routing via the generalized Clarke &
solve TSP. As an heuristic global optimization algorithm Wright savings) to solve CVRP.

13
200 H. Zhang et al.

Korayem et al. [16] proposed a K-means clustering and 4.1 Description and Mathematical Model of VRPTW
gray wolf optimization hybrid algorithm for grouping and
then routing for solving CVRP. Ng et al. [17] proposed Let S = {1, 2, … , n} be the collection of depot,
a novel scheme using a Multiple Colonies Artificial Bee I = {1, 2, … , m} is the customer set of depot, V = {0, I} is the
Colony algorithm for CVRP with re-routing strategy. set of depots (defined as "0") and all customers. Vs is the set of
Wei et al. [18, 19] summarized the transformation of the i n c l u d e d c u st o m e r s a c c e s s e d by ve h i c l e s ,
neighborhood structure and introduced two-dimensional cij (cij > 0, cij = 0, i, j ∈ V) is the distance between customer
packing constraints under the Simulated Annealing CVRP points, C is the maximum load of the vehicle, gi (i ∈ V) is the
solution framework. Tao and Wang [20] and Zhang et al. required load capacity of each customer, xijs indicates whether
[21] integrated three-dimensional packing and loading the vehicle s will serve the customer j after it has served the
capacity constraints in tabu search and random local customer i(xij = 1 for service, xijs = 0 for no service), ysi indi-
j

search, respectively, to achieve 3L-CVRP problem solving. cates whether customer i is served by [ vehicle
] s(yi = 1 is to be
s
Akpinar [22] combined large-scale neighborhood served, yi = 0 is not to be served), ai , bi is the time window
s
search with ant colony algorithm to improve the solution limit of customer i , ai is the allowed service start time of cus-
scheme optimization. Sze et al. [23] constructed a two- tomer i , bi is the termination time of customer i , si is the start-
stage hybrid variable neighborhood algorithm by intro- ing service time of customer i.
ducing large neighborhood search to expand population The mathematical model of VRPTW is as follows:
diversity. The optimization objective function is as follows:
Akhand and Peya [24] proposed an adaptive scanning { }
and velocity speculation Particle Swarm Optimization { } ∑∑ ∑∑ ∑
s s
algorithm for path optimization. min F1 , F2 = min xoj , cij xij
Hannan et al. [25] proposed an improved particle swarm j∈Vs s∈S k∈S i∈Vs j∈Vs �{i}
(11)
optimization algorithm to solve the optimal path of gar-
bage collection. where, F1 is the number of vehicles and F2 is the total driv-
In general, as an implicit constraint of VRP, capacity ing distance of all vehicles.
constraint exists in all types of VRP, so in the follow- ∑
xijs = ysi , ∀i ∈ V, ∀s ∈ S
ing discussion, it will continue to combine with other (12)
j∈Vs �{i}
constraints.
∑
xijs = ysj , ∀j ∈ V, ∀s ∈ S
(13)
i∈Vs �{j}

4 VRP with Time Window

∑
qi ysi ≤ C, ∀s ∈ S
VRPTW is a vehicle routing problem with time windows. It (14)
i∈Vs
adds a time window constraint to each demand point, that is,
for each demand point, the earliest time and the latest time of ∑
s
∑
s
x0j = xi0 = 1, ∀s ∈ S
service start are set, and vehicles are required to start serving (15)
j∈I i∈N
customers in the time window.
There are two kinds of time window restrictions of ∑
demand point. One is the hard time window, which requires ysi = 1, ∀i ∈ I (16)
that the vehicle must start to serve customers within the time s∈S

window, and must wait if it arrives early, and will be rejected ∑

if it is late; the other is the soft time window, which does xijs ≤ ||Vs || − 1, ∀s ∈ S
(17)
not have to start serving customers within the time window, i,j∈Vs
but must be punished when starting service outside the time
window. The biggest difference between soft time window ai ≤ si ≤ bi , ∀i ∈ I (18)
and hard time window is to replace waiting and rejection
with punishment. where,
Solomon [26] first considered the vehicle routing problem xijs ∈ {0, 1}, ∀i, j ∈ V, ∀s ∈ S
with time window (VRPTW) in 1987, which requires that
the delivery vehicles providing service must deliver within
the time range of customer satisfaction, and give a classic ysi ∈ {0, 1}, ∀i ∈ I, ∀s ∈ S
example of this kind of problem.

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 201

Equation (12) indicates that the vehicle directly serves the vehicle routing and scheduling problem with soft time
customer j after serving customer i , (13) indicates that the windows, and a heuristic algorithm of "clustering first and
vehicle only serves one customer i before serving customer constructing path" was proposed.
j , (14) indicates that each vehicle shall not exceed the maxi- Balakrishnan et al. [31] constructed a linear penalty
mum carrying capacity, Eq. (15) indicates that the vehicle function under the condition of setting upper limits for
must return to the depot after service, (16) means that each both the fine payable and waiting time of customers, and
customer can only be served by one vehicle, Eq. (17) means solved Solomon’s R1 and RC1 examples with three simple
to avoid the generation of sub loops in the service process, heuristic algorithms.
(18) represents a hard time window constraint. Taillard et al. [32] proposed a soft time window and
corresponding
[ ] penalty function on the basis of hard time
4.2 Research Progress of VRPTW window ai , bi , that is, if the vehicle arrives before ai , it
will wait until time ai for service and treat it as a hard time
Based on the literature retrieval and analysis, the research on window. If the vehicle arrives after bi and the customer
VRPTW can be divided into two categories: the research of allows, the service will be carried out, but the correspond-
time window constraints and the combination of time win- ing penalty fee (defined as a linear function of the viola-
dows and other constraints. At present, there are many meth- tion amount) should be paid. The Tabu search algorithm
ods to solve VRPTW, which can be roughly divided into is used to solve the soft time window problem.
three categories: simple heuristic method; hybrid algorithm Guertin et al. [33] extended the vehicle routing and
of heuristic constrained programming and local search; scheduling problem with soft time windows from static
precise optimization algorithm and intelligent optimiza- to dynamic. Taking express service as an example, tabu
tion method, such as genetic algorithm, tabu search, simu- search algorithm was used to solve the dynamic problem
lated annealing and improved particle swarm optimization on a parallel platform.
algorithm. Guertin et al. [34, 35] studied the loading and unload-
Many problems in production practice can be attributed ing problem of multi ships with soft time window. Based
to VRPTW. For example, the problem of making hot strip on the maximum
[ ] allowable violation di of internal time
max

rolling plan in iron and steel plant is actually a VRPTW window ai , bi , the outer time window was expanded to
problem. Some service industries also have such problems: [ai − dimax , bi + dimax ],
such as postal delivery, aircraft, train and bus scheduling. Hashimoto et al. [36] established a dynamic program-
Since [27] proved that VRPTW is a NP hard problem, the ming model for the case that the travel time penalty func-
research on VRPTW algorithm has focused on various heu- tion is convex and the time window penalty function is non
ristic algorithms. The emergence of intelligent heuristic convex function, and an iterative local search algorithm is
algorithms such as genetic algorithm, tabu search method designed to solve the optimal service start time of visited
and simulated annealing method provides new tools for solv- customers.
ing VRPTW problem. Fu et al. [37] summarized six types of VRPSTW, and
designed a unified penalty function. The setting of different
1. Separate time window constraints parameter values corresponds to different types of VRPSTW.
At the same time, a unified tabu search algorithm was pro-
Both [28] and [29] have applied genetic algorithm to posed to solve the six types of VRPSTW.
solve VRPTW problem. The goal of the former is to mini- Iqbal et al. [38] used artificial bee colony to stimulate bee
mize the total service cost, while the latter has two objec- foraging behavior, and used hybrid hyper heuristics algo-
tives: first, to use the least number of vehicles; second, to rithm to solve multi-objective vehicle routing problem with
minimize the total cost under the premise of using the least soft time window.
vehicles. Since the structure of adaptive large neighborhood search
Due to the time window limitation of the demand point, (ALNs) was proposed by [39], it can automatically select
vehicles will be punished when they arrive outside the a good operator to destroy and repair the solution, so as to
time window. When solving the problem, it is necessary get a better solution, Ropkes and Pisingerd [40] used this
to construct a penalty function that conforms to the actual algorithm to optimize five types of vehicle routing problems,
situation. including VRPTW, CVRP, MDVRP, SDVRP and OVRP.
Koskosidis et al. [30] considered that no matter whether The reliability of the algorithm was verified by 486 standard
the vehicle arrived before ai or after bi , customers were problems.
allowed to pay corresponding penalty fees (defined as Since [8], ant colony algorithm has been mainly used to
a linear function of the amount of violations). Then, a find the optimal solution of the optimization problem. A
hybrid integer programming model was established for large number of studies show that the ant colony algorithm

13
202 H. Zhang et al.

has a good effect on solving the constrained vehicle routing window assignment vehicle routing problem (TWAVRP)
problem. with time-dependent travel times. Aiming at minimizing
Gambardella et al. [41] successfully applied ant colony the expected transportation cost, they developed a branch-
algorithm to VRPTW for the first time. Wang et al. [42] price-And-cut algorithm.
considered the vehicle routing problem with time window
(VRPTW-STC) and designed a hybrid heuristic algorithm 2. Mixed constraints of time window and other factors
based on multi ant system. Zhang et al. [43] proposed a
multi-objective vehicle routing problem with flexible time ① Vehicle Routing Problem with Simultaneous Delivery
window (MOVRPFlexTW), and proposed a solution strat- -Pickup and Time Windows, VRPSDPTW
egy based on ant colony optimization and three mutation
operators. VRPSDPTW is based on the classic vehicle routing prob-
Memetic algorithm (MA) was first proposed by [44] to lem (VRP) and adds time window constraint. Vehicles must
simulate cultural evolution, the algorithm simulates the pro- serve customers within the time window specified by cus-
cess of cultural evolution. For cultural genes, genes always tomers. At the same time, on the basis of distribution for
develop in the direction of advancement. The transmission customers, the vehicle should not only provide distribution
process of genes is strictly copied in the previous genera- service to customers, but also pick up goods at the custom-
tion of genes. While inheriting the excellent genes of their er’s location and transport the goods to the depot.
parents, the offspring mutation operation is not random, but Vehicle Routing Problem with Simultaneous Pickup-
develops in a better direction based on a lot of knowledge. Delivery (VRPSPD) was first proposed by [54] and used to
Nagata et al. [45] proposed a boundary combination solve the problems of public libraries. VRPSDPTW was first
meme algorithm based on penalty function to solve VRPTW. proposed by [55], and proposed to use an accurate algorithm
Qi et al. [46] proposed a decomposition based MA to to solve the problem.
solve the multi-objective vehicle routing problem with time VRPSDPTW problem studies a team of vehicles from the
windows. distribution center to carry out distribution and pick-up ser-
Nalepa and Blocho [47] used the MA with adaptive vices for a series of customers. It requires a reasonable path
parameters to solve the VRPTW problem. planning to minimize the distribution cost on the premise
Molina et al. [48] pointed out that when a limited fleet of meeting the constraints of time window and vehicle load.
characterized by different capacity, fixed cost and variable There are few literatures about VRPSDPTW at home and
cost can be used to service a group of customers who must abroad, and there is a lack of effective algorithm. There are
be visited within a predetermined time window, the vehi- two conclusions: most of VRPSDPTW solving methods are
cle heterogeneity problem will arise. A memetic algorithm meta heuristic algorithms, the data set of [56] becomes the
based on ACS is proposed to solve the heterogeneous vehicle only international universal test data set for solving VRPS-
routing problem with time window (HVRPTW). DPTW. However, none of the known meta heuristic algo-
For other VRP problems which only consider time rithms in the literature can obtain the optimal solution of all
window constraints, Calvete et al. [49] studied the exact 65 test cases, so scholars are still looking for more stable and
algorithm of "enumerating first and then optimizing" to efficient algorithms.
solve the medium-sized vehicle routing problem with time In recent years, the heuristic algorithms for solving
windows. Considering the service efficiency and customer VRPSDPTW problem are shown in Table 1.
reliability, Tas et al. [50] used tabu search algorithm and Wang et al. [57] proposed vehicle routing problem with
adaptive large neighborhood search algorithm to solve simultaneous delivery and pickup while sharing load and
vehicle routing problem with time dependence and random time window (VRPSDPSLTW), as a mixed integer program-
travel time with soft time windows. A case study showed ming problem. To solve this problem, a hybrid heuristic
that the method can be used to solve large-scale problems. algorithm is developed.
Niu et al. [51] proposed a hybrid evolutionary algorithm
based on Membrane Computing (MC) to solve the sin- ② Multi-trip VRP with Fuzzy Demands Considering
gle delivery vehicle routing problem, and achieved good Time Windows preference, MVRPFDTW
results. Xu et al. [52] proposed hybrid genetic algorithm
and particle swarm optimization (PSO) for vehicle routing MVRPFDTW is based on the VRP with fuzzy demands
problem with time window, which decoded the path by (VRPFD) to open the vehicle travel limit and consider the
particle real number coding method, so as to reduce the customer time window preference. The solution of mvrpfdtw
computational burden; at the same time, it was combined is to determine a vehicle multi journey planning route and
with crossover operator of genetic algorithm to avoid fall- real-time adjustment scheme to meet the customer’s fuzzy
ing into local optimum. Spliet et al. [53] proposed the time demand and time window preference, so that the logistics

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 203

cost is the lowest and the overall customer satisfaction is threshold value, it will return to the distribution center to
the highest. reload the goods, and then serve the next customer; other-
Open vehicle travel limit can be described as allowing wise, it will directly serve the next customer. The on time
vehicles to travel in multiple trips. It is defined that a vehicle delivery probability model of vehicle routing problem with
departs from the distribution center and returns to the distri- stochastic demand and time window was built.
bution center as a journey. Each vehicle has only one service Gutierrez et al. [66] designed a hybrid meta heuristic
route. Each service route can contain multiple single trips algorithm to solve the vehicle routing problem with sto-
from and to the distribution center. Each customer can only chastic demand (VRPSD). The algorithm combines the
be served once by one trip of one vehicle. cultural gene algorithm and greedy random algorithm.
Neira et al. [58] established a two integer programming Compared with the existing meta heuristic methods, the
(IP) model for multi trip vehicle routing problem with time effectiveness of the proposed method is verified.
window, service-related loading time and limited travel Salavatikhoshghalb et al. [67] proposed a hybrid algo-
duration (MTVRPTW-SDLT). rithm, which set the maximum travel threshold and the
minimum replenishment threshold. If the value is greater
③ VRP with Fuzzy Time Windows,VRPFTW than the minimum replenishment threshold, the goods will
be loaded back to the distribution center; otherwise, the
VRPFTW is a VRPTW expansion problem caused by uncer- next customer will be served. A hybrid resource strategy
tain customer service time. for vehicle routing problem with stochastic demand was
Li et al. [59] constructed a single objective fuzzy chance proposed.
constraint model based on fuzzy possibility theory, and
proposed an improved tabu search algorithm to solve the ⑤ Vehicle Routing Problem with Service Choice and
problem. Time Window, VRPSCTW​
Zulvia and Kuo [60] studied the VRPTW with fuzzy
travel time and demand. Based on the basic model of Wang et al. [68] proposed an improved ant colony algorithm
VRPTW, the fuzzy chance constrained model was estab- to solve the periodic vehicle routing problem with time win-
lished and the corresponding hybrid genetic ant colony algo- dow and service selection.
rithm was designed to solve the problem.
Chen et al. [61] established a two-stage single objective ⑥ Multi-depot vehicle routing problem with time win-
programming model, and used the nearest neighbor method dows, MDVRPTW
and improved particle swarm optimization algorithm to
obtain the optimal solution. Wang et al. [69] proposed a multi-objective and multi-site
Cao and Lai [62] and Kuo and Zulvia [63] established vehicle routing problem with time windows, and developed
a single objective fuzzy chance constraint model based on a two-stage multi-objective evolutionary algorithm (TS-
Fuzzy credibility measure theory, and solved the problem MOEA) to solve the problem.
by using improved hybrid difference algorithm and hybrid Zhen et al. [70] considered the vehicle routing problem
genetic ant colony optimization algorithm respectively. with multiple stops and multiple trips with time windows
and delivery times, established a mixed integer program-
④ Vehicle Routing Problem with Stochastic Demand and ming model, and proposed a hybrid particle swarm opti-
Time Window,VRPSDTW mization algorithm and a hybrid genetic algorithm to solve
the model.
Lei et al. [64] proposed a DTD (detour-to-depot) modified Compared with VRP, VRPTW must take into account
algorithm (DTD) to solve the capacity vehicle routing prob- the delivery time and time window. The main reason is that
lem with stochastic demand and time window, it means that customers have the deadline of service time and the earli-
only when the remaining capacity of the vehicle is zero, est service start time. Therefore, under this constraint, the
the vehicle will return to the distribution point to reload the original VRP Problem has to consider scheduling in time
goods, otherwise it will continue to serve the next customer. in addition to routing in space. At the same time, due to the
This method ignores the actual demand of the next customer, limitation of time window in distribution center, the path
resulting in the vehicle must return to the distribution center length is also limited indirectly, therefore, the total operating
to reload the goods, resulting in a huge cost. cost of VRPTW includes not only the transportation cost,
Zhang et al. [65] put forward a preventive replenish- but also the time cost and the penalty cost for not delivering
ment idea, that is, a threshold is set in advance according within the time window. Therefore, in order to get a good
to the formula. When a vehicle has finished serving a cus- answer, it is very important to discuss the temporal and spa-
tomer, if the remaining capacity is less than or equal to the tial problems.

13
204 H. Zhang et al.

5 Split Delivery VRP Model hypothesis:

In the traditional VRP, all task points can only receive one ① The distance between any two points is symmetric,
service, and most of the vehicles are not fully loaded. The that is, dij = dji.
collection and delivery demand of the task point is insepara- ② The distance relationship between points conforms to
ble, but the demand for collection and delivery of some or all the triangle inequality, that is, dik + dkj > dij.
task points in logistics service may be large, but the vehicle ③ All vehicles must start from the depot and return to
capacity is very small, so the vehicle can not complete the the depot after completing a mission.
task in one visit, so how to solve this kind of problem is ④ Each customer’s needs must be met, but can be met
worth studying. by one or more vehicles.
At present, most of the research on VRP has set a condi-
tion in advance, that is, each customer’s demand (which is Objective constraint
less than the vehicle’s maximum carrying capacity) must be The goal is to arrange the vehicle route reasonably, so that
completed in one vehicle and one service. the vehicle operation cost is as low as possible.
Dror and Trudeau [71] proposed the split delivery VRP Minimum total distance traveled:
(SDVRP) for the first time, and discussed the pure delivery R m m
∑ ∑∑
problem with demand splitting. After that, [72] compared J = min dij xijr (20)
the solution obtained by demand splitting and demand insep- r=1 i=0 j=0
arable, and proved that if the demand of the customer point
is allowed to be split and distributed when the distribution Flow conservation, that is, the number of vehicles enter-
center delivers products to the customer points, it is better ing a point is equal to the number of vehicles leaving the
than the case that the customer demand can only be served point:
by one vehicle in terms of the driving distance and the num- n
∑ m
∑
ber of vehicles used. r
xik = r
xkj , (k = 0, 1, … , m;r = 1, 2, … , R) (21)
Since then, many scholars have improved the SDVRP i=0 j=0
algorithm and compared the examples, and obtained better
Ensure that each point is visited at least once and the
solutions in different problem situations.
requirements are met:
R m
∑ ∑
5.1 SDVRP Mathematical Model and Feasible xijr ≥ 1, (j = 0, 1, … , m) (22)
Solution Analysis r=1 i=0

1. Mathematical model R
∑
yri = gi , (i = 1, 2, … , m) (23)
Let I be the set of customer points, I = {1, 2, 3, … , m}, set
r=1

I0 = 0 represents the distribution center, that is, the depot. The number of arc sides between the customers to be
dij is the distance between any two points i, j in the set I and served in each line is equal to the number of customer points
I0 . gi (i = 1, 2, … , m) is the demand of customer i , C is the to be served minus 1:
maximum carrying capacity of the vehicle, R is the number ∑∑
of vehicles required to complete the task, its value is: xijr = |Sr | − 1, (r = 1, 2, … , R;Sr ⊆ I − I0 )
r r
(24)
i∈S j∈S
⌈m ⌉
∑ gi
R= (19) Vehicle capacity limit:
i=1
C
m
∑
“⌈ ⌉ ” m e a n s t o t a k e a n i n t e g e r u p w a r d . yri ≤ C, (r = 1, 2, … , R) (25)
i=1
xijr (r = 1, 2, … , R, i, j = 0, 1, … , m and i ≠ j) is the decision
variable. If and only if the vehicle in the r route passes Customer i can only be served if and only if the vehicle
t h r o u g h a r c s (i, j) , xijr = 1 , ot h e r w i s e , xijr = 0 . passes by:
yri (r = 1, 2, … , R, i = 1, 2, … , m) is the demand of customer m
∑
i in the r line. Sr represents the set of customers served in the xijr gi ≥ yri , (r = 1, 2, … , R, i = 1, 2, … , m) (26)
r route, |Sr | represents the number of elements contained in j=0
the set S.
where,

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 205

xijr ∈ {0, 1}, (i, j = 1, 2, … , m;r = 1, 2, … , R) C C

gi ≥ yri ≥ 0, (i = 1, 2, … , m;r = 1, 2, … , R) rk+1 rk+1

B B
rk
rk
2. Feasible solution analysis O O
A A

① If one SDVRP has a feasible solution and the dis- (a) Split meet (b) Separat ely meet
tance relationship between points conforms to the triangle
inequality, then when the customer demand is equal to the Fig. 5  B is on the left side of the AC line
maximum carrying capacity of the vehicle, that is, when
gi = C , this point in the optimal solution of the problem
should be satisfied separately. It is not suitable to split. In case 1) and case 2), the sum of the two sides
of the triangle is larger than the third side, so that
Suppose that the demand of a customer point B is equal dCB + dBA − dCA > 0. In case 3) dCB + dBA − dCA = 0.
to the vehicle carrying capacity w, and the demand of this Therefore, when the customer demand is equal to the
point is split up and satisfied by rk and rk+1 respectively. maximum carrying capacity of the vehicle, it is suitable to
The demand met in each line are recorded as y kB and meet it alone, not split.
(y k+1)B respectively. The other demand points of line rk
and rk+1 are aggregated into one point, which is recorded ② If a SDVRP has a feasible solution and the distance
as A and C respectively. The maximum demand of these between the points conforms to the triangle inequality,
two points is w−ykB and w−y(k+1)B respectively. The depot then there will be at most one common point in any two
is represented by O point, and the coordinate position of routes of the optimal solution of the problem.
B is relative to AC line in three cases: 1) On the right side
of the AC line; 2) on the left side of the AC line; 3) on It can be inferred from ② that:
the AC line.
If the demand of point B are met separately, the sum ③ If a SDVRP has a feasible solution and the distance
of the requirements of points A and Cq AC should be the relationship between points conforms to the triangle
maximum of: inequality, there will be no k-split cycle in the optimal
solution of the problem.
Max qAC = (W − ykB ) + (W − y(k+1)B ) = W,
K-split cycle means that k lines can be constructed for k cus-
Therefore, two points A and C can be merged into one tomers, line rm includes customers im and m = 1, 2, … , k − 1,
route, and the routes that are split and satisfied separately rk includes customers ii and ik , this structure is called k-split
are shown in Figures 4, 5, 6. cycle. Figure 7 shows a 3-split cycle.
As can be seen from Figures 4, 5, 6, the total
route length satisfied by splitting under three con- ④ If a SDVRP has a feasible solution and the distance
d i t i o n s i s Zsplit = dOC + dCB + dOB + dBA + dAO , between points conforms to the triangle inequality, then
and total route length to be met individually is the number of demand points to be split is less than the
Zunsplit = dOC + dCA + dAO + dOB + dAO , a n d t h e n number of routes in the optimal solution of the problem.
Zsplit − Zunsplit = dCB + dBA − dCA .

C C
B B
C
C B B
rk rk+1 rk+1
rk+1
rk+1 rk
rk rk
O A O A O A O A

(a) Split meet (b) Separately meet (a) Split me et (b) Separat ely meet

Fig. 4  B is on the right side of the AC connection Fig. 6  B on AC line

13
206 H. Zhang et al.

Table 1  Heuristic algorithms Author (year) Solution method Algorithm parameters

for VRPSDPTW
Cao et al. 2009 [52] Genetic algorithm (GA) Cross probability
Wang et al. 2012 [51] Mutation probability
Lai et al. 2010 [53] Differential evolution (EA) Weighting factor, Mutation probability
Wang et al 2015a, b [54] Simulated annealing (SA) Initial temperature, cooling rate

Table 2  Exact algorithm for SDVRP

Author (year) Solving algorithm

Dror et al. (1994) [71] Branch and bound method

i2 r1 Lee et al. (2006) [72] Dynamic programming (shortest path)
i1
Jin et al. (2007) [73] Cluster path (two stage method)
Moreno et al. (2010) [74] Column generation and section method
Archetti et al. (2011) [75] Branch pricing section method
O Archetti et al. (2014) [76] Branch section method (two kinds)
Bianchessi et al. (2019) [77] Branch and cut
r2 r3
optimization effect, it is still necessary to explore new
algorithms.
The exact algorithm can only be used to solve small
scale NP hard problems. For large-scale SDVRP, heuristic
strategy is often used to help determine the upper bound of
i3 the problem (in order to speed up the solving speed) when
using accurate algorithm to solve SDVRP, which is mainly
manifested as the combination of heuristic strategy, col-
Fig. 7  3-split cycle
umn generation and section method. However, although the
hybrid exact algorithm can obtain the upper bound of the
5.2 Research Progress of SDVRP problem with better quality by using heuristic algorithm, it
can not get a better lower bound of the problem, and most of
Dror and Trudeau [71] pointed out that SDVRP is a NP the algorithms still can not be used to solve SDVRP. With
hard problem, and it is very difficult to solve it. For large- the rapid development of heuristic algorithm research, the
scale problems, even if VRP model can be established eas- research literature of SDVRP heuristic algorithm also shows
ily, the corresponding SDVRP model may not be able to an obvious growth trend, especially meta heuristics (also
obtain the optimal solution of the problem. The main rea- known as intelligent optimization algorithm, hyper heuristic
son is that the number of split points is a decision variable algorithm, sub heuristic algorithm, general heuristic algo-
in SDVRP model, which leads to a huge incomplete gap in rithm, etc.) and hybrid heuristic algorithm have become the
the integer programming model. However, due to the split main research direction of SDVRP solution.
vehicle routing problem is more in line with the actual SDVRP is a NP hard problem, it is difficult to get an accu-
situation of vehicle distribution, and also more conducive rate solution in a reasonable time [73]. Summarized several
to reduce the cost of distribution enterprises, therefore, variants of SDVRP and the existing literature.
after the split vehicle routing problem was put forward, a
large number of scholars have studied it according to dif- An accurate algorithm for SDVRP
ferent backgrounds and assumptions.
The algorithms for solving SDVRP mainly focus on pre- Since 1994, many scholars have used accurate algorithms
cise algorithm and heuristic algorithm. The key problem to solve SDVRP. The main accurate algorithms for SDVRP
to be solved in the algorithm is how to split the require- are shown in Table 2.
ments reasonably. Due to the complexity of SDVRP, it is Dror et al. [74] gave the mixed integer programming
difficult for existing algorithms to balance two aspects: model of SDVRP-UF and its branch and bound method.
computational time and optimization degree. In order to According to the characteristics of the optimal solution of
make the algorithm have fast calculation speed and good the problem, several valid inequalities were proposed. In

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 207

[74], the solution of the heuristic algorithm given by [71, include classical heuristic algorithm (neighborhood search,
72] was taken as the upper bound of the solution, and the scanning algorithm, etc.), hybrid heuristic algorithm and
lower bound of the solution was improved by using the effec- meta heuristic algorithm. Among them, meta heuristic
tive inequality. Jin et al. [75] proposed a two-stage method algorithm and hybrid heuristic algorithm have become the
with valid inequality constraints to solve SDVRP, including main algorithms to solve SDVRP. Table 3 summarizes the
customer clustering stage and TSP path arrangement stage. heuristic algorithms for SDVRP.
The former aimed to determine the set of customer points
visited by each vehicle without considering the vehicle run- At present, many hybrid heuristic algorithms have been
ning cost, while the latter aimed to solve the TSP of the path applied to solve SDVRP. The results show that the hybrid
to be solved. The cost obtained was used as the lower bound heuristic algorithm based on precise algorithm can get better
of the problem and inserted into the next iteration cluster- solution than single heuristic algorithm.
ing process with constraints. Moreno et al. [76] proposed In terms of meta heuristic algorithm for SDVRP, Khme-
the column generation and section method for SDVRP, lev and Kochetov [84] showed that tabu search algorithm
which can effectively obtain the lower bound of the solu- was superior to genetic algorithm in solving performance.
tion. They use dynamic programming method to solve the
pricing problem, and use two heuristic algorithms to solve 5.3 Research Progress of SDVRP Derived Types
the pricing problem to speed up the solution. Archetti et al.
[77] first solved SDVRP-LP and SDVRP-UF by branch pric- In the theoretical research and practical application of vehi-
ing section method. At the same time, they gave a heuristic cle routing problem, scholars will consider some charac-
algorithm, which takes the solution (column) obtained by teristics and constraints of the problem, such as customers
the subproblem as the initial solution, and obtains the upper need to obtain service in a specific period of time, or have
bound of the solution of the problem. Based on the relaxa- the demand of picking up and delivering goods at the same
tion model which guarantees the lower bound of the optimal time. According to different problem characteristics and con-
solution, Archetti et al. [78] proposed two branch tangent straints, SDVRP can be derived into many types, including
plane methods for SDVRP. The first relaxation model was SDVRP with time windows(SDVRPTW), pick-up and deliv-
actually an improved model of [79], and the second was a ery VRP(PDVRP), HFVRP with split delivery(HFVRPSD),
new relaxation model based on vehicle loading. By compar- etc.
ing the two branch tangent plane methods, it can be seen SDVRPTW means that each customer can only be
that the performance of the first improved relaxation model accessed in one (or more) specific time periods. If the actual
algorithm is better. visit time of the vehicle to the customer is not within the
Batch delivery usually brings inconvenience to customers required time period, it will be punished.
who want to meet their needs in a single visit. The methods to solve SDVRPTW mainly include: Tabu
Gulczynski et al. [80] and Xiong et al. [81] studied the search algorithm [85], Column generation method [86],
split delivery vehicle routing problem with minimum deliv- Branch-and-price-and-cut [87] [88], Branch and price [89],
ery amounts (SDVRP-MDA), and pointed out that batch Branch and Price and Cut [90], Branch-and-Cut [91] et al.
delivery is allowed only when each visiting vehicle meets at In the vehicle routing problem, sometimes customers
least a small part of customer demand [82]. Proposed a new have the demand of picking up and delivering goods at the
multi start solution for SDVRP-MDA separated delivery same time. At present, a large number of literatures have
vehicle routing problem. The initial solution was generated studied the pick-up and delivery VRP (PDVRP) with pick-up
by both the node insertion process and the route addition and delivery problem with split load (PDPSL).
process, which had a single parameter to control the restart. The methods to solve PDPSL mainly include: Construct
Then, these solutions were improved by a new search opera- heuristic algorithm [92], Heuristic algorithm [93], Variable.
tor inspired by the node pop-up chain and the variable neigh- neighborhood search heuristic algorithm [94], Two stage
borhood descent meta heuristic algorithm. heuristic algorithm [95], Competitive decision algorithm
Chen et al. [83] proposed that each customer’s demand [96] et al.
should be divided into small pieces to solve SDVRP, and the In practical application, the demand of a considerable
calculation was carried out on 82 benchmark examples, and part of the task points may be large. At this time, if only one
good results were achieved. vehicle is required to provide service at each customer point,
it will inevitably lead to the increase of vehicle no-load rate
2. Heuristic algorithm for SDVRP and waste transportation resources. Therefore, logistics
enterprises will split customer demand and complete it by
Heuristic algorithm is often used to solve large-scale multiple vehicles, so as to make full use of vehicle loading
SDVRP. The heuristic algorithms used in the literature capacity and reduce vehicle running cost.

13
208 H. Zhang et al.

Table 3  Heuristic algorithm for Heuristic algorithm Literature Algorithm

SDVRP
Hybrid heuristic algorithm [82] C-W conservation method and record updating method
[83] Tabu search algorithm and optimization idea
[84] Column generation method and bounded search algorithm
[85] Variable neighborhood descent (VND), random tabu
search, fast heuristic algorithm
[86] Hybrid genetic algorithm
Meta heuristic algorithm [87] Tabu search algorithm
[88] Meta heuristic algorithm based on neighborhood search
[89] Adaptive memory algorithm
[90] Tabu search algorithm
[183]
[91] Genetic algorithm (two kinds)
[92] Particle size calculation method
[93] Iterative local search algorithm
[94] Ant colony algorithm and hybrid meta heuristic algorithm
[95] Particle swarm optimization algorithm
General heuristic algorithm [68] Neighborhood search (two stage method)
[78] Cutting plane method
[96] Scanning algorithm
[184] Two stage method
[97] Pre-cutting method
[77] Branch and cut
[98]
[99] Adaptive large neighborhood search method
[100]

6 Dynamic Vehicle Routing Problem (DVRP) In addition to designing a reasonable and effective vehi-
cle scheduling scheme according to the existing distribu-
At present, most literatures on VRP assume that the factors tion orders, there is also an important problem that needs to
such as customer demand and vehicle travel time are static form a variety of distribution schemes that meet the needs of
or known, and will not change in the process of path plan- people in real time and dynamically according to the chang-
ning and implementation. In other words, if the vehicle ing actual needs of customers and the influence of random
routing is arranged when the vehicle, time, personnel, cus- changing factors, so as to meet the real-time and personal-
tomer demand and other information are determined, once ized needs of customers.
the vehicle path is determined, the vehicle route will not
change in the whole operation process. This kind of VRP
is called static vehicle routing problem (SVRP). However, 6.1 Definition and Characteristics of DVRP
there may be many dynamic factors in the actual logistics
distribution, such as customer demand, traffic conditions, Based on previous studies, DVRP is defined as: according to
weather conditions, personnel, vehicles, etc., which cause the known information, the system initially plans the vehicle
great trouble to vehicle routing. It is necessary to dynami- path, receives the dynamic information such as customer
cally arrange vehicle routing according to the constantly demand and traffic condition in real time during the imple-
updated system information. For these dynamic changes, mentation process, and re-plans the vehicle path according
the traditional static VRP theory and method are no longer to certain rules, so as to achieve the goal required by the
applicable. This kind of problem belongs to dynamic vehi- system.
cle routing problem (DVRP). With the development of The schematic diagram of dynamic vehicle routing prob-
intelligent transportation system, Internet of things, arti- lem is shown in Fig. 8.
ficial intelligence and network communication technology, The main characteristics of DVRP are as follows:
it is possible to obtain dynamic demand and real-time traf-
fic information in real time, which provides strong support 1. It can receive all kinds of dynamic information in real
for DVRP research. time.

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 209

Fig. 8  Diagram of vehicle

scheduling problem

Depot

Depot Static Key Point Vehicle

Customer

Unfinished Finished Current Dynamic

Route Route Route Customer

The most important feature of DVRP is that the system In DVRP, after receiving all kinds of dynamic information,
can receive all kinds of dynamic information in real time, the system needs to process in real time, and re plan the vehi-
including the demand of new customers, the adjustment of cle path through fast calculation to ensure the fast response
demand and service time of old customers, and the occa- of the system and the real-time of vehicle path update.
sional traffic congestion on the road.

2. All kinds of dynamic information are random and uncer- 6.2 Classification of DVRP
tain.
In DVRP, the main dynamic elements include: dynamic
The time for the system to receive dynamic information demand, dynamic road travel time, dynamic service vehi-
such as customer demand and real-time traffic conditions cles, etc. According to the difference of dynamic elements
is random and unpredictable, the content of dynamic infor- and their characteristics, the existing research on DVRP can
mation, such as the number of newly added customers, be divided into the following three types:
service time window and traffic congestion, is uncertain.
1. Dynamic Demand based VRP, DDVRP
3. Vehicle paths need to be updated dynamically.
The demand is dynamic, including new customer demand,
After receiving all kinds of dynamic information, the sys- old customer canceling distribution service, changing deliv-
tem needs to re plan the vehicle path and update the vehi- ery location, demand or time window, etc., which results
cle route in real time according to the determined route in vehicle route updating according to dynamic demand
update strategy. information.

4. It has fast response to dynamic information. 2. Real-time traffic information based VRP, RTVRP

13
210 H. Zhang et al.

Traffic conditions change in real time. Traffic congestion is on preference degree was established. Pavone et al. [107]
caused by accidental events such as traffic accidents, vehicle first considered the patient factor of customer waiting in the
failures, weather changes, etc., resulting in vehicle speed dynamic vehicle routing problem.
affected by real-time traffic conditions, and road travel time
changes dynamically. Vehicles need to update their routes 1. Research on DDVRP
according to real-time traffic information during driving.
According to the different number of dynamic demand fac-
3. Dynamic demand & real-time traffic information based tors, the existing DDVRP can be divided into single dynamic
VRP, DDRVRP demand element VRP and multi dynamic demand element
VRP.
This is the most complex problem of DVRP. The dynamic
elements include not only the dynamic demands of custom- ① Single dynamic demand element
ers, but also the dynamic changes of road network traffic
conditions. Single dynamic demand elements mainly focus on the new
customer demand.
6.3 Research Progress of DVRP For the new customer demand and considering the capac-
ity constraint, Table 4 summarizes the research under the
The research on DVRP can be traced back to the 1980s [97], optimization goal of minimizing the total cost.
first proposed DVRP, and then, [98–102] further studied the Considering the time window, capacity constraints, multi
connotation and characteristics of DVRP. At present, there cycle and other constraints, Table 5 summarizes the relevant
are a certain number of research literature and review papers research progress.
in the field of DVRP [100, 103]. Psaraftis [100] introduced
the application fields of DVRP and the progress and influ- ② Multi dynamic demand elements
ence of related technologies. Ghiani et al. [104] analyzed
the types and main characteristics of real-time VRP, and Schyns et al. [108] considered various dynamic demand fac-
introduced the research progress of solving algorithms. Pil- tors, such as new customer demand, customer cancellation,
lac et al. [105] introduced the application fields of DVRP, demand change, time window change, customer or vehicle
and summarized the research progress of solving methods location change, and took the minimization of dynamic event
for dynamic deterministic VRP and dynamic stochastic VRP. response time as the optimization objective, constructed a
Psaraftis [103] analyzed the impact of the new generation multi vehicle DVRP model with capacity and time window
of information technology on DVRP and summarized the constraints, and compared and analyzed the impact of tim-
existing research literature according to 11 classification ing update strategy and dynamic event update strategy on
standards. responsiveness.
Based on the characteristics of real-time and complex-
ity of DVRP, DVRP needs corresponding new models and
theories. Based on the research of SVRP, some scholars put 2. Research on RTVRP
forward some new models to solve DVRP.
Bertsimas et al. [98] studied the DVRP and dynamic According to the different constraints, the existing literature
traveling repairman problem with capacity constraints for mainly studies three types of RTVRP with capacity con-
multiple vehicles, and proposed a queuing theory model to straint, RTVRP with time window constraint, and RTVRP
solve the problem. Teodorovi and Pavkovi [106] first stud- with capacity and time window constraints. Table 6 sum-
ied the VRP of fuzzy customer demand. Customer infor- marizes the three types of studies under RTVRP.
mation and decision-maker’s preference were expressed
by fuzzy numbers, and a fuzzy judgment criterion based 3. Research on DDRVRP

Table 4  Demand and capacity Optimization objectives Literature Methods

constraints of new customers
with minimum total cost Minimum total cost [122] Build an object oriented event
driven DVRP model
The Minimum of total cost, the number of second [186] Two stage optimization strategy
visit customers and the total incompatibility
Minimum total cost [123] Convert DVRP to SVRP

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 211

Table 5  Research on multiple constraints under single dynamic

Additional restrictions Optimization objectives Literature Methods

Time windows The Minimum of driving distance and num- [124] Two stage optimization strategy
ber of delayed service customers
Maximize the number of service customers [125] Improved fuzzy change constrained program-
and minimize the average waiting time of ming
customers
Capacity and time window constraints Minimize the number of vehicles, total driv- [126] The two-stage strategy is to generate the
ing distance and waiting time initial path and update the path periodically
Minimum total cost [127] according to the time period
Time window, Maximum number of accepted requests and [128] Re-route vehicles regularly according to time
capacity constraint, minimum operation cost period
multi period Minimum total cost [129] Adaptive service strategy
Maximum number of accepted service [130] Expected dynamic strategy based on approxi-
requests mate dynamic programming
Capacity, Minimum total travel distance or total man [131] Decision support system is used to solve the
time window, multi Depot hours problem
Minimum total cost [132] Evolutionary algorithm is adopted according
to time period

Table 6  Research on RTVRP

Research directions Optimization objectives Literature Methods

RTVRP with time window constraint Minimize total cost and vehicle usage [134] The method based on Markov decision is used
to determine the best departure time and
route
Minimum travel time [135] A DVRP model based on Markov decision
process is established
RTVRP With capacity and time Minimum of the weighted sum of opera- [136] Considering the real-time change of road travel
window constraints tions, service cancellations and path time, the backup vehicle path planning is
interruptions carried out with the vehicle fault location as
the starting point
Minimum total cost [137] Receive travel time information from custom-
ers and update route
[138] Three shortest path models are used to deter-
mine the shortest path between two nodes,
and the impact on the total cost is compared
[139] Receive the dynamic travel time information in
real time and update the vehicle path
RTVRP with capacity constraint Minimum travel time [140] Construct the DVRP model with road capacity
constraints
[141] The dynamic path guidance model with road
capacity constraint was constructed, and the
vehicle path was updated by receiving traffic
information regularly

DDRVRP is one of the most complex problems in Dynamic factors such as dynamic demand and real-time
DVRP. The dynamic elements include not only the traffic information have an important impact on the con-
dynamic demand of customers, but also the dynamic struction and solution of DVRP model, and have strong
changes of road network traffic conditions. Table 7 sum- correlation with constraints such as capacity and time win-
marizes the research progress of DDRVRP. dow. Table 8 summarizes the relevant studies on dynamic
features associated with other constraints.
4. Dynamic feature and Constraint Association

13
212 H. Zhang et al.

Table 7  DDRVRP related studies

Research directions Optimization objectives Literature Methods

New customer request and dynamic road Minimize user and operator costs [142] Hybrid adaptive predictive control method
network for real-time updating of vehicle routing
New customer orders and dynamic road Minimize weighted sum of travel time, [143] Receive customer request and dynamic
travel time total delivery delay time and return to travel time information from customers,
distribution center and update vehicle path
DVRP with capacity and time window Minimize total cost and total delay [144] Real time optimization
constraints
Real time demand and emergencies Minimizing the weighted sum of total [145] Proposed a strategy to distinguish served
travel time and total waiting time customers from non service customers by
key points
New orders and dynamic travel time Minimize number of vehicles, total travel [146] Hierarchical clustering method clustering
time customers, real-time update vehicle path

6.4 Research Progress on DVRP Solving Methods 1. Vehicle routing problem with stochastic demand
(VRPSD)
The solution of DVRP involves two aspects: one is what
rules the system updates the vehicle route, the other is VRPSD problems can be generally understood as follows:
which optimization algorithm the system uses to optimize For a series of customers with random demand, organ-
the route. ize appropriate driving routes to make the vehicles pass
through them orderly, under certain constraint conditions
1. Route update strategy for DVRP (such as the demand of goods, the quantity of delivery, the
time of delivery and delivery, the limit of vehicle capac-
As for the route updating strategy of DVRP, the existing ity, the limit of driving mileage, the limit of time, etc.) to
literature puts forward four strategies, as shown in Table 9. achieve certain goals (such as the shortest distance, the
least cost, the least time, the least number of vehicles used,
2. DVRP optimization algorithm etc.).
Gaur et al. [109] gave randomized approximation algo-
DVRP algorithm mainly includes two categories: precise rithms for stochastic cumulative VRPs for the split and
algorithm and heuristic algorithm. Table 10 summarizes the unsplit deliveries.
DVRP algorithm. Dinh et al. [110] studied the chance constrained vehi-
In order to improve the quality of optimal solution of cle routing problem (CCVRP), which is a form of vehi-
DVRP model, many hybrid heuristic algorithms have been cle routing problem with stochastic demand (VRPSD), in
proposed. Table 11 summarizes the hybrid heuristics for which the possibility that the capacity of each vehicle is
DVRP. exceeded is limited. And then investigated the use of a
The existing DVRP models mainly consider the dynamic branch-and-cut-and-price (BCP) algorithm.
information such as dynamic demand elements and dynamic
travel time, and also consider the constraints such as capac- 2. Vehicle routing optimization problem with simultaneous
ity or time window, and establish a variety of DVRP models, delivery and pickup (VRPSDP)
which lay a good foundation for the study of DVRP. How-
ever, with the rapid development of intelligent transportation Problem Description: the goods needed by customers are
and intelligent logistics, the research of DVRP model is fac- sent from the site to each customer by multiple vehicles,
ing some new challenges and problems worthy of discussion. and the goods exported by each customer are collected
back to the site from the customers. The location of each
customer is fixed, the quantity of goods to be distributed
and collected at the customer is certain and known in
7 Other types of VRP advance. The customer has certain constraints on the time
period of distribution and collection, and the carrying
In the actual logistics, there are many derivatives of VRP. capacity of each vehicle is fixed. Therefore, the reason-
In addition to the VRP derivatives which are studied more, able arrangement of vehicle routes is required to optimize
they also include: the objective function.

13
 Table 8  Research on the association between dynamic elements and constraints
Directions Constraints Objective Literature Methods

Association of dynamic elements with capacity New customer demand Minimum total cost [122] Object oriented event driven DVRP model
constraints Minimize the total cost, the number of second- [186] Two stage optimization strategy
ary customers and total incompatibility
Real time traffic information Minimum travel time [141] Update vehicle path by receiving traffic informa-
tion regularly
Association of dynamic features with time New customer demand Minimum total driving distance, number of [124] Two stage optimization strategy
window constraints delayed service customers
Maximize the number of service customers and [125] Improved fuzzy change constrained programming
minimize the average waiting time of custom-
ers
Real time traffic information Minimize the weighted sum of travel time, total [143] Receive customer request and dynamic travel
delivery delay time and return to depot time information from customers, and update
vehicle path
Minimum total cost [137] Receive travel time information from customers
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms

and update route

[138] Three shortest path models are used to determine
the shortest path between two nodes, and the
impact on the total cost is compared
Minimum travel time [135] A DVRP model based on Markov decision pro-
cess is established
Dynamic features are associated with capacity Real time traffic information Minimize the weighted sum of operations, ser- [136] Considering the real-time change of road travel
and time window constraints vice cancellations, and path disruptions time, the backup vehicle path planning is car-
ried out with the vehicle fault location as the
starting point
New customer demand, Real Minimize the weighted sum of total travel time [145] Proposed a strategy to distinguish served custom-
time traffic information and total waiting time ers from non service customers by key points
New customer demand Minimum total cost [127] A two-stage strategy is adopted to generate the
initial path and update the path periodically
according to the time period

13
213
214 H. Zhang et al.

Table 9  Route update strategy of DVRP

Strategy Literature Basic thoughts

periodic update [147] Each working day is divided into several equal time slices, and the DVRP is transformed into a series of
(first time) SVRP according to the time slice, and the optimization method of SVRP is used to update the vehicle
path
Dynamic event update [32] Whenever a new dynamic event is received, the system will aggregate it with the original information,
(first time) update the current path, and inform the next customer to be served after the vehicle completes the on-
going demand point service
Customer update [137] Whenever the vehicle arrives at the customer’s place, the dispatching center communicates with the
[143] customer and the driver, receives the dynamic information such as customer request and real-time traffic
situation, re plans the vehicle path, and informs the driver of the next service destination
Key point update [145] Every time the dynamic information is received, the customer points which are in service or about to be
(first time) served are determined as the key points

Table 10  Summary of DVRP Types of algorithms Problem solving Literature Specific algorithm
algorithm
Precise algorithm DVRP [148] Column generation algorithm
[139]
[149] Neural dynamic programming algorithm
[150] Adaptive dynamic programming algorithm
[151] Approximate dynamic programming algorithm
[135] Backward dynamic programming algorithm
Heuristic algorithm DVRPSTW [32] Parallel tabu search algorithm
DDVRP [152] Insertion algorithm
DDRVRP [146]
DDVRPPD [153] Local Search Method
DDVRP [122] Adaptive large neighborhood search algorithm
[154]
[127]
[129] Variable neighborhood search algorithm
[186]
[124]
New order request VRP [155] Ant colony system (first time)
DDVRP [133] Ant colony algorithm
[125]
DVRP [156] Genetic algorithm (first time)
DDVRP [142] Adaptive hybrid two level genetic algorithm
RTVRP [138]
DDRVRP [157]
[126]
[123]
DDVRPPD [158] Self organizing PSO algorithm
MDVRP [132] Two stage multi group PSO algorithm
DDVRP [159] Multi group adaptive PSO algorithm
[160] PSO algorithm

The vehicle routing problem with simultaneous pickup ① Vehicle Routing Problem with Simultaneous Delivery
delivery (VRPSPD) was first proposed by [54] to solve the -Pickup and Time Windows, VRPSDPTW
problem of public library.
VRPSDP can be divided into the following three That is, each customer has a pre-set time window, and the
categories: vehicle can not provide service earlier than the customer’s

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 215

Table 11  Hybrid heuristic Problem solving Literature Specific algorithm

algorithm for DVRP
Multi objective DDVRP [161] Ant colony algorithm based on large neighborhood search
[162] Ant colony algorithm based on local search
[125] Fuzzy ant colony algorithm
RTVRP [163] Clustering ant colony algorithm
DDRVRP [143] Insertion heuristic algorithm
DDVRP [131] Parallel cruise construction algorithm
[164] Multi scene method
[136] Heuristic algorithm based on Lagrange relaxation
[165] Memetic local optimization algorithm
[141] Improved hierarchical Dijkstra algorithm

earliest allowed service time, nor later than the customer’s In the field of GVRP research, Kara et al. [113] first pro-
latest allowed service start time. posed the energy consumption minimization VRP, Bekta
and Laporte [114] put forward the problem of pollution path,
② Vehicle Routing Problem with Simultaneous Delivery Erdoğan et al. [115] put forward the concept of green vehicle
-Pickup and Maximum Distance Length,VRPSDPMML routing problem (GVRP).
Shuai et al. [116] regarded the green vehicle routing prob-
That is, each vehicle starts from the central warehouse, lem with large capacity as a new variant of vehicle routing
serves customers, and returns to the central warehouse. The problem, and proposed two solving methods: two-stage heu-
driving distance of the vehicle is constrained by the maxi- ristic algorithm and meta heuristic algorithm based on ant
mum distance. colony system. Numerical experiments are carried out on a
randomly generated example.
③ Vehicle Routing Problem with Simultaneous Deliv-
ery—Pickup and Time Limit, VRPSDPTL 5. Multi-trip Vehicle Trip Routing Problem, MTVRP

The vehicle must return to the depot within the maximum Cattaruzza et al. [117] introduced the overview of multi
voyage time after serving customers. trip vehicle routing problem (MTVRP) and the related path
problem that allows vehicles to perform multiple trips, gave
3. Open vehicle routing problem (OVRP) a unified mathematical formula, and summarized the precise
and heuristic methods to solve it.
If the vehicle returns to the original depot after completing
the distribution service, it is called closed vehicle routing 6. Vehicle Routing Problem with Backhauls, VRPB
problem. In this case, the distribution vehicle is generally
owned by the distribution center; if the vehicle does not need Pinto et al. [118] discussed the vehicle routing problem
to return after completing the distribution service, it is called with two-dimensional load constraints and mixed routes and
open vehicle routing problem (OVRP), and the distribution backhaul. The problem solved belongs to the subclass of
vehicle in this case is generally rental vehicle. general pick-up and delivery problems, and also considered
Lalla-Ruiz et al. [111] provided a new mathematical for- the two-dimensional loading constraints. A variable neigh-
mula based on two indexes for MDOVRP, as well as the borhood search algorithm is proposed to solve the problem.
enhancement of eliminating constraints on alternative sub
runs. [112] proposed a Hybrid Grasshopper Optimization
Algorithm to minimize the total number of vehicles and the 8 Research Trends
shortest total driving distance.
As the key of vehicle scheduling, vehicle routing problem
4. Green vehicle routing problem (GVRP) can reduce the cost of logistics transportation. Based on the
research of VRP in recent decades, this paper summarizes
Green transportation requires reasonable layout and planning the basic characteristics of VRP, classifies VRP according to
of freight points and distribution centers, and by shorten- the actual transportation situation, and then gives the math-
ing vehicle routes and reducing empty load rate, the goal of ematical models of various VRP, and classifies and analyzes
energy conservation and emission reduction can be achieved. the solving algorithms.

13
216 H. Zhang et al.

After summarizing several artificial intelligence meth- 4. The actual transportation process problem may be the
ods, we can see that: the results of Tabu Search (TS) algo- synthesis of multiple constraints.
rithm are the closest to the optimal solution, but its opera- 5. Multi objective green VRP with mixed constraints
tion time is the longest, which is 2–3 times of Genetic should be considered.
Algorithm (GA) and 20 times of Simulated Annealing
(SA). Because GA can approach the optimal solution and
shorten the operation time, GA can take both the opera- Acknowledgements This paper is supported by the Graduate Innova-
tion Foundation of Jiangsu Province under Grant No. KYLX16_0781,
tion time and efficiency into account, so it is a promising the Natural Science Foundation of Jiangsu Province under Grants No.
method. SA algorithm is very fast and can provide a cer- BK20181340, the 111 Project under Grants No. B12018, and PAPD
tain degree of optimization scheme. It has good effect in of Jiangsu Higher Education Institutions.
solving small-scale problems.
Vehicle routing problem (VRP) has achieved certain
research results, but the depth of its research is not enough.
Generally, only one or two kinds of constraint conditions References
are considered, and the model of multi constraint and
1. Raff S (1983) Routing and scheduling of vehicles and crews: the
multi-objective dynamic vehicle routing problem is rarely state of the art. Comput Oper Res 10(2):63–67
studied. 2. Solomon MDJ (1988) Time window constrained routing and
The existing literature mainly uses accurate algorithm or scheduling problems: a survey. Transp Sci 22(1):1–11
3. Laporte G (1992) The vehicle routing problem: an overview of
heuristic algorithm to solve the problem, and its applica-
exact and approximate algorithms. Eur J Oper Res 59(2):231–247
tion scenarios are mostly small-scale simulation cases or 4. El-Sherbeny NA (2010) Vehicle routing with time windows: an
simulation cases. overview of exact, heuristic and metaheuristic methods. J King
Therefore, the following aspects should be considered in Saud Univ Sci 22(3):123–131
5. Psaraftis HN, Wen M, Kontovas CA (2016) Dynamic vehicle
the follow-up study.
routing problems: three decades and counting. Wiley-Intersci-
ence 67(1):3–31
1. Parallel algorithm research which suitable for large- 6. Dantzig GB, Ramser JH (1959) The truck dispatching problem.
scale. Manag Sci 6(1):80–91
7. Fisher (1995) Vehicle routing. In: Handbooks in operations
research & management science
The number of customers in the existing literature is 8. Dorigo MMV, Colorni A (1996) Ant system: optimization by
mostly less than 100, and the scale of VRP is not large. With a colony of cooperative agents. IEEE Trans Syst Man Cybern
the wide application of parallel computing, strengthening 26(1):29–41
9. Bullnheimer BHRF, Strauss C (1998) Applying the ant system
the research of parallel algorithm and designing a parallel
to the vehicle routing problem. Paper presented at the advances
algorithm with fast operation speed and suitable for large- and trends in local search paradigms for optimization, Boston
scale VRP is a direction worthy of in-depth study. 10. Bullnheimer B, Hartl RF, Strauss C (1999) An improved ant
system algorithm for thevehicle routing problem. Ann Oper Res
89:319–328
2. Design of standard test case base for solving algorithm.
11. Belf JE, Mcmullen PR (2004) Ant colony optimization tech-
niques for the vehicle routing problem. Adv Eng Inform
Most of the existing literatures are based on the needs 18(1):41–48
of research to transform or design the Solomon benchmark 12. Gajpal Y, Abad PL (2009) Multi-ant colony system (MACS)
for a vehicle routing problem with backhauls. Eur J Oper Res
case base. Simulation cases are used as simulation examples
196(1):102–117
of various branch studies. Because the relevant parameters 13. Chen C, Ting C (2006) An improved ant colony system algo-
of simulation examples are inconsistent, it is not convenient rithm for the vehicle routing problem. J Chin Inst Ind Eng
to compare and analyze the simulation results. Therefore, 23(2):115–126
14. Yu B, Yang ZZ, Yao B (2009) An improved ant colony optimiza-
it is of great value to establish a standard test case library
tion for vehicle routing problem. Eur J Oper Res 196(1):171–176
for VRP model solving algorithm to promote the further 15. Juan AA, Faulin J, Ruiz R, Barrios B, Caballé S (2010) The
research of VRP. SR-GCWS hybrid algorithm for solving the capacitated vehicle
routing problem. Appl Soft Comput 10(1):215–224
16. Korayem L, Khorsid M, Kassem SS (2015) Using grey wolf
3. An empirical study on the model and solving algorithm.
algorithm to solve the capacitated vehicle routing problem. IOP
Conf Ser Mater Sci Eng 83(1):012–014. https://​doi.​org/1​ 0.​1088/​
In order to verify the feasibility and effectiveness of 1757-​899x/​83/1/​012014
the proposed model and the proposed algorithm, most of 17. Ng KKH, Lee CKM, Zhang SZ, Wu K, Ho W (2017) A multiple
colonies artificial bee colony algorithm for a capacitated vehicle
the existing literatures use VRP benchmark case library or
routing problem and re-routing strategies under time-dependent
small-scale simulation cases, and there are few studies on traffic congestion. Comput Ind Eng 109:151–168. https://d​ oi.o​ rg/​
the actual road network as application scenarios. 10.​1016/j.​cie.​2017.​05.​004

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 217

18. Wei L, Zhang Z, Zhang D, Lim A (2015) A variable neigh- 38. Iqbal S, Kaykobad M, Rahman MS (2015) Solving the multi-
borhood search for the capacitated vehicle routing problem objective vehicle routing problem with soft time windows with
with two-dimensional loading constraints. Eur J Oper Res the help of bees. Swarm Evol Comput 24:50–64
243(3):798–814 39. Ropke S, Pisinger D (2006) A unified heuristic for a large class
19. Wei L, Zhang Z, Zhang D, Leung SCH (2017) A simulated of vehicle routing problems with backhauls. Eur J Oper Res
annealing algorithm for the capacitated vehicle routing prob- 171(3):750–775. https://​doi.​org/​10.​1016/j.​ejor.​2004.​09.​004
lem with two-dimensional loading constraints. Eur J Oper Res 40. Ropkes S, Pisingerd D (2006) A unified heuristic for vehi-
265(3):843–859 cle routing problems with backhauls. Eur J Oper Res
20. Tao Y, Wang F (2015) An effective tabu search approach with 171(3):750–775
improved loading algorithms for the 3L-CVRP. Comput Oper 41. Gambardella LM, Taillard É, Agazzi G (1999) MACS-VRPTW:
Res 55:127–140 a multiple colony system for vehicle routing problems with time
21. Zhang Z, Wei L, Lim A (2015) An evolutionary local search for windows [M].
the capacitated vehicle routing problem minimizing fuel con- 42. Wang Y, Wang L, Peng Z, et al (2019) A multi ant system based
sumption under three-dimensional loading constraints. Transp hybrid heuristic algorithm for vehicle routing problem with ser-
Res B Methodol 82:20–35. https://​doi.​org/​10.​1016/j.​trb.​2015.​ vice time customization. Swarm Evolut Comput 50
10.​001 43. Zhang H, Zhang Q, Ma L et al (2019) A hybrid ant colony opti-
22. Akpinar S (2016) Hybrid large neighbourhood search algo- mization algorithm for a multi-objective vehicle routing problem
rithm for capacitated vehicle routing problem. Expert Syst with flexible time windows. Inf Sci 490:166–190
Appl 61:28–38 44. Moscato P (1989) On evolution, search, optimization, genetic
23. Sze JF, Salhi S, Wassan N (2017) The cumulative capacitated algorithms and martial arts—towards memetic algorithms.
vehicle routing problem with min-sum and min-max objec- Caltech Concurrent Computation Program
tives: an effective hybridisation of adaptive variable neigh- 45. Nagata Y, Bräysy O, Dullaert W (2010) A penalty-based edge
bourhood search and large neighbourhood search. Transp Res assembly memetic algorithm for the vehicle routing problem
B Methodol 101:162–184. https://​doi.​org/​10.​1016/j.​trb.​2017.​ with time windows. Comput Oper Res 37(4):724–737
04.​003 46. Qi Y, Hou Z, Li H, Huang J, Li X (2015) A decomposition based
24. Akhand MAH, Peya ZJ (2017) Capacitated vehicle routing prob- memetic algorithm for multi-objective vehicle routing problem
lem solving using adaptive sweep and velocity tentative PSO. Int with time windows. Comput Oper Res 62:61–77
J Adv Comput Sci Appl 8(12):288–295 47. Nalepa J, Blocho M (2016) Adaptive memetic algorithm for
25. Hannan MA, Akhtar M, Begum RA et al (2018) Capacitated minimizing distance in the vehicle routing problem with time
vehicle-routing problem model for scheduled solid waste collec- windows. Soft Comput 20(6):1–19
tion and route optimization using PSO algorithm. Waste Manag 48. Molina JC, Salmeron JL, Eguia I (2020) An ACS-based memetic
71:31–41 algorithm for the heterogeneous vehicle routing problem with
26. Solomon MM (1987) Algorithms for the vehicle routing and time windows. Expert Syst Appl. https://​doi.​org/​10.​1016/j.​eswa.​
scheduling problem with time window constraints. Oper Res 2020.​113379
32(2):254–265 49. Calvete HI, Galé C, Oliveros MJ, Sánchez-Valverde B (2007) A
27. Savelsbergh MWP (1985) Local search in routing problems with goal programming approach to vehicle routing problems with
time windows. Ann Oper Res 4(1):285–305 soft time windows. Eur J Oper Res 177(3):1720–1733
28. Thangiah S et al. (1991) A genetic algorithm system for vehicle 50. Tas D, Dellaert N, Van Woensel T, De Kok T (2014) The time-
routing with time windows. Paper presented at the proceedings dependent vehicle routing problem with soft time windows and
of the seventh conference on artificial intelligence applications stochastic travel times. Transp Res C Emerg Technol 48:66–83
Miami, Florida 51. Niu Y, He J, Wang Z, Xiao J (2014) A P-based hybrid evolution-
29. Blanton JL, Wainwright RL (1993) Multiple vehicle routing with ary algorithm for vehicle routing problem with time windows.
time and capacity constraints using genetic algorithms. In: Inter- Math Probl Eng 2014:1–11
national conference on genetic algorithms. 52. Xu SH, Liu JP, Zhang FH, Liang W, Sun LJ (2015) A com-
30. Koskosidis YA, Powell WB, Solomon MM (1992) An optimiza- bination of genetic algorithm and particle swarm optimiza-
tion-based heuristic for vehicle routing and scheduling with soft tion for vehicle routing problem with time windows. Sensors
time window constraints. Transp Sci 26 15(9):21033–21053
31. Balakrishnan N (1993) Simple heuristics for the vehicle routeing 53. Spliet R, Dabia S, Van Woensel T (2018) The time window
problem with soft time windows. J Oper Res Soc 44(3):279–287 assignment vehicle routing problem with time-dependent travel
32. Taillard RD, Badeau P, Gendreau M, Guertin F, Potvin J-Y times. Transp Sci 52(2):261–276
(1997) A Tabu Search Heuristic for the Vehicle Routing Problem 54. Min H (1989) The multiple vehicle routing problem with simul-
with Soft Time Windows. Transp Sci 31(2):170–186 taneous delivery and pick-up points. Transp Res A General
33. Guertin F, Potvin J-Y, Taillard É (1999) Parallel Tabu search 23(5):377–386
for real-time vehicle routing and dispatching. Transp Sci 55. Angelelli E, Mansini R (2002) The vehicle routing problem with
33(4):381–390 time windows and simultaneous pick-up and delivery. Springer
34. Kjetil F (2000) Evaluating the trade-off between the level of 56. Wang HF, Chen YY (2012) A genetic algorithm for the simulta-
customer service and transportation costs in a ship scheduling neous delivery and pickup problems with time window. Comput
problem. Marit Policy Manag 27(2):145–153 Ind Eng 62(1):84–95
35. Fagerholt K (2001) Ship scheduling with soft time windows: an 57. Wang Y, Ma X, Lao Y, Wang Y, Mao H (2018) Vehicle routing
optimisation based approach. Eur J Oper Res 131(3):559–571 problem: simultaneous deliveries and pickups with split loads
36. Hashimoto H, Ibaraki T, Imahori S, Yagiura M (2006) The vehi- and time windows. J Transp Res Board. https://​doi.​org/​10.​3141/​
cle routing problem with flexible time windows and traveling 2378-​13
times. Discret Appl Math 154(16):2271–2290 58. Neira DA, Aguayo MM, Fuente RDl, Klapp MA (2020) New
37. Fu Z, Eglese R, Li LYO (2008) A unified tabu search algorithm compact integer programming formulations for the multi-trip
for vehicle routing problems with soft time windows. J Oper Res vehicle routing problem with time windows. Comput Ind Eng
Soc 59(5):663–673 144

13
218 H. Zhang et al.

59. Li H, Xu Z, Zhou F (2012) A study on vehicle routing problem 79. Belenguer JM, Martinez MC, Mota E (2000) A lower bound
with fuzzy demands based on improved Tabu search. In: Inter- for the split delivery vehicle routing problem. Oper Res
national conference on computational & information sciences 48(5):801–810
60. Zulvia FE, Kuo RJ (2012) Solving CVRP with time window, 80. Gulczynski D, Golden B, Wasil E (2010) The split delivery
fuzzy travel time and demand via a hybrid ant colony opti- vehicle routing problem with minimum delivery amounts.
mization and genetic algorithm. In: WCCI 2012 IEEE world Transp Res E Logist Transp Rev 46(5):612–626
congress on computational intelligence, Brisbane, Australia, 81. Xiong Y, Gulczynski D, Kleitman D, Golden B, Wasil E
vol 6, pp 10–15 (2012) A worst-case analysis for the split delivery vehicle
61. Chen JQ, Li WL, Murata T (2013) Particle swarm optimization routing problem with minimum delivery amounts. Optim Lett
for vehicle routing problem with uncertain demand. In: IEEE 7(7):1597–1609. https://​doi.​org/​10.​1007/​s11590-​012-​0554-9
international conference on software engineering & service 82. Han FW, Chu YC (2016) A multi-start heuristic approach for
science the split-delivery vehicle routing problem with minimum deliv-
62. Cao E, Lai M (2009) A hybrid differential evolution algorithm ery amounts. Transp Res E Logist Transp Rev 88:11–31
to vehicle routing problem with fuzzy demands. J Comput 83. Chen P, Golden B, Wang X, Wasil E (2017) A novel approach
Appl Math 231(1):302–310 to solve the split delivery vehicle routing problem. Int Trans
63. Kuo Z (2017) Hybrid genetic ant colony optimization algo- Oper Res 24(1–2):27–41. https://​doi.​org/​10.​1111/​itor.​12250
rithm for capacitated vehicle routing problem with fuzzy 84. Khmelev A, Kochetov Y (2015) A hybrid VND method for the
demand—a case study on garbage collection system. Paper split delivery vehicle routing problem. Electron Notes Discrete
presented at the 4th international conference on industrial engi- Math 47:5–12
neering and applications 85. Ho SC, Haugland D (2004) A tabu search heuristic for the
64. Lei H, Laporte G, Guo B (2011) The capacitated vehicle rout- vehicle routing problem with time windows and split deliver-
ing problem with stochastic demands and time windows. Com- ies. Comput Oper Res 31(12):1947–1964
put Oper Res 86. Archetti C, Bouchard M, Desaulniers G (2010) Enhanced
65. Zhang J, Lam WHK, Chen BY (2016) On-time delivery proba- branch and price and cut for vehicle routing with split deliver-
bilistic models for the vehicle routing problem with stochastic ies and time windows. Transp Sci 45(3):285–298
demands and time windows. Eur J Oper Res 249(1):144–154 87. Desaulniers G (2010) Branch-and-price-and-cut for the split-
66 Gutierrez A, Dieulle L, Labadie N, Velasco N (2018) A Hybrid delivery vehicle routing problem with time windows. Oper Res
metaheuristic algorithm for the vehicle routing problem with 58(1):179–192
stochastic demands. Comput Oper Res 99:135–147 88. Luo Z, Qin H, Zhu W, Lim A (2016) Branch and price and cut
67. Salavatikhoshghalb M, Gendreau M, Jabali O et al (2019) A for the split-delivery vehicle routing problem with time win-
hybrid recourse policy for the vehicle routing problem with dows and linear weight-related cost. Transp Sci 51(2):668–687
stochastic demands. EURO J Transp Logist 8(3):269–298 89. Salani M, Vacca I (2011) Branch and price for the vehicle rout-
68. Wang Y, Wang L, Chen G, Cai Z, Zhou Y, Xing L (2020) An ing problem with discrete split deliveries and time windows.
improved ant colony optimization algorithm to the periodic Eur J Oper Res 213(3):470–477
vehicle routing problem with time window and service choice. 90. Luo Z, Qin H, Zhu W, Lim A (2017) Branch and price and cut
Swarm Evol Comput. https://​doi.​org/​10.​1016/j.​swevo.​2020.​ for the split-delivery vehicle routing problem with time win-
100675 dows and linear weight-related cost. Transp Sci 51(2):668–687
69. Wang J, Weng T, Zhang Q (2018) A two-stage multiobjective 91. Bianchessi N, Irnich S (2019) Branch-and-cut for the split
evolutionary algorithm for multiobjective multidepot vehicle delivery vehicle routing problem with time windows. Transp
routing problem with time windows. IEEE Trans Cybern 1–12 Sci 53(2):1–26
70. Zhen L, Ma C, a KW, Xiao L, Zhang W (2020) Multi-depot 92. Mitra S (2005) An algorithm for the generalized vehicle rout-
multi-trip vehicle routing problem with time windows and ing problem with backhauling. Asia Pac J Oper Res 22(2):153–
release dates. Transp Res E Logist Transp Rev 135 169. https://​doi.​org/​10.​1142/​s0217​59590​50005​22
71. Dror M, Trudeau P (1989) Savings by split delivery routing. 93. Nowak M, Ergun Z, White CC (2008) Pickup and delivery with
Transp Sci 23(2):141–145 split loads. Transp Sci 42(1):32–43
72. Dror M, Trudeau P (1990) Split delivery routing. Nav Res 94. Chen Q, Li K, Liu Z (2014) Model and algorithm for an
Logist 37(3):383–402 unpaired pickup and delivery vehicle routing problem with
73. Archetti C, Speranza MG (2012) Vehicle routing problems with split loads. Transp Res E Logist Transp Rev 69:218–235
split deliveries. Int Trans Oper Res 19(1–2):3–22. https://​doi.​ 95. Wang Y, Ma X-l, Lao Y-t, Yu H-y, Liu Y (2014) A two-stage
org/​10.​1111/j.​1475-​3995.​2011.​00811.x heuristic method for vehicle routing problem with split deliv-
74. Dror M, Laporte G, Trudeau P (1994) Vehicle routing with eries and pickups %J Frontiers of Information Technology &
split deliveries. Discret Appl Math 50(3):239–254 Electronic Engineering. J Zhejiang Univ SCI C (Comput Elec-
75. Jin M, Liu K, Bowden RO (2007) A two-stage algorithm with tron) 15(3):200–210
valid inequalities for the split delivery vehicle routing problem. 96. Wang K, Ye C, Ning A (2015) Achieving better solutions for
Int J Prod Econ 105(1):228–242 vehicle routing problem involving split deliveries and pickups
76. Moreno L, AragãO MPD, Uchoa E (2010) Improved lower using a competitive decision algorithm. Asia Pac J Oper Res
bounds for the split delivery vehicle routing problem. Oper 32(4):1550022
Res Lett 38(4):302–306 97. Golden BL, Assad AA (1988) Vehicle routing: methods and
77. Archetti C, Bouchard M, Desaulniers G (2011) Enhanced studies. Elsevier Science Publishers, Amsterdam
branch and price and cut for vehicle routing with split deliver- 98. Bertsimas, Dimitris J, Ryzin V, Garrett (1991) A stochastic
ies and time windows. Transp Sci 54 and dynamic vehicle routing problem in the Euclidean plane.
78. Archetti C, Bianchessi N, Speranza MG (2014) Branch-and-cut INFORMS
algorithms for the split delivery vehicle routing problem. Eur 99. Bertsimas DJ, Gv R (1993) Stochastic and dynamic vehicle
J Oper Res 238(3):685–698 routing in the euclidean plane with multiple capacitated vehi-
cles. Oper Res 41(1):60–76

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 219

100. Psaraftis HN (1995) Dynamic vehicle routing: Status and pros- 122. Lee C-G, Epelman MA, White CC, Bozer YA (2006) A short-
pects. Ann Oper Res 61(1):143–164 est path approach to the multiple-vehicle routing problem with
101. Bertsimas DJS-L (1996) A new generation of vehicle rout- split pick-ups. Transp Res B Methodol 40(4):265–284. https://​
ing: robust algorithms, addressing uncertainty. Oper Res doi.​org/​10.​1016/j.​trb.​2004.​11.​004
44(2):286–304 123. Chen S, Golden B, Wasil E (2007) The split delivery vehicle
102. Gendreau MPJ (1998) Fleet management and logistics. Kluwer routing problem: applications, algorithms, test problems, and
Academic Publishers, Boston computational results. Networks 49(4):318–329. https://​doi.​
103. Psaraftis HN, Wen M, Kontovas CA (2016) Dynamic vehi- org/​10.​1002/​net.​20181
cle routing problems: three decades and counting. Networks 124. Archetti C, Speranza MG, Savelsbergh MWP (2008) An opti-
67(1):3–31. https://​doi.​org/​10.​1002/​net.​21628 mization-based heuristic for the split delivery vehicle routing
104. Ghiani G, Guerriero F, Laporte G, Musmanno R (2003) Real- problem. Transp Sci 42(1):22–31
time vehicle routing: solution concepts, algorithms and parallel 125. Jin M, Kai L, Eksioglu B (2008) A column generation approach
computing strategies. Eur J Oper Res 151(1):1–11 for the split delivery vehicle routing problem. Networks
105. Pillac V, Gendreau M, Guéret C, Medaglia AL (2013) A review 36(2):265–270
of dynamic vehicle routing problems. Eur J Oper Res 225(1):1– 126. Bortfeldt A, Yi J (2020) The split delivery vehicle routing
11. https://​doi.​org/​10.​1016/j.​ejor.​2012.​08.​015 problem with three-dimensional loading constraints. Eur J
106. Teodorovi D, Pavkovi G (1996) The fuzzy set theory approach to Oper Res 282(2):545–558. https://​doi.​org/​10.​1016/j.​ejor.​2019.​
the vehicle routing problem when demand at nodes is uncertain. 09.​024
Fuzzy Sets Syst 82(3):307–317 127. Archetti C, Speranza MG, Hertz A (2006) A tabu search algo-
107. Pavone M, Bisnik N, Frazzoli E, Isler V (2009) A stochastic rithm for the split delivery vehicle routing problem. Transp Sci
and dynamic vehicle routing problem with time windows and 40(1):64–73
customer impatience. Mob Netw Appl 14(3):350–364 128. Derigs U, Li B, Vogel U (2010) Local search-based metaheuris-
108. Schyns M (2015) An ant colony system for responsive dynamic tics for the split delivery vehicle routing problem. J Oper Res Soc
vehicle routing. Eur J Oper Res 245(3):704–718. https://​doi.​org/​ 61(9):1356–1364
10.​1016/j.​ejor.​2015.​04.​009 129. Aleman RE, Zhang X, Hill RR (2010) An adaptive memory algo-
109. Gaur DR, Mudgal A, Singh RR (2020) Improved approximation rithm for the split delivery vehicle routing problem. Journal of
algorithms for cumulative VRP with stochastic demands. Discret Heuristics 16(3):441–473
Appl Math 280:133–143 130. Aleman REHRR (2010) A tabu search with vocabulary building
110. Dinh T, Fukasawa R, Luedtke JR (2018) Exact algorithms for approach or the vehicle routing problem with split demands. Int
the chance-constrained vehicle routing problem. Math Program J Meta-heuristics 1(1):55–80
172:105–138 131. Qiu M, Fu Z, Eglese R, Tang Q (2019) A tabu search algorithm
111. Lalla-Ruiz E, Mes M (2020) Mathematical formulations and for the vehicle routing problem with discrete split deliveries and
improvements for the multi-depot open vehicle routing problem. pickups. Comput Oper Res
Optim Lett. https://​doi.​org/​10.​1007/​s11590-​020-​01594-z 132. Wilck Iv JH, Cavalier TM (2012) A genetic algorithm for the
112. Soto-Mendoza V, García-Calvillo I, Ruiz-y-Ruiz E, Pérez-Ter- split delivery vehicle routing problem. American Journal of
razas J (2020) A hybrid grasshopper optimization algorithm Operations Research 02(02):207–216. https://​doi.​org/​10.​4236/​
applied to the open vehicle routing problem. Algorithms. https://​ ajor.​2012.​22024
doi.​org/​10.​3390/​a1304​0096 133. Berbotto L, García S, Nogales FJJAOOR (2014) A Randomized
113. Kara İ, Kara BY, Yetis MK (2007) Energy minimizing vehicle Granular Tabu Search heuristic for the split delivery vehicle rout-
routing problem. In: International conference on combinatorial ing problem. Ann Oper Res 222(1):153–173
optimization & applications, pp 62–71 134. Silva MM, Subramanian A, Ochi LS (2015) An iterated local
114. Bekta T, Laporte G (2011) The pollution-routing problem. search heuristic for the split delivery vehicle routing problem.
Transp Res B Methodol 45(8):1250 Comput Oper Res 53:234–249. https://​doi.​org/​10.​1016/j.​cor.​
115. Erdoğan S, Miller-Hooks E (2012) A green vehicle routing prob- 2014.​08.​005
lem. Transp Res E Logist Transp Rev 48(1):114 135. Rajappa GP, Wilck JH, Bell JE (2016) An ant colony optimiza-
116. Shuai Z, Gajpal Y, Appadoo SS (2018) A meta-heuristic for tion and hybrid metaheuristics algorithm to solve the split deliv-
capacitated green vehicle routing problem. Ann Oper Res ery vehicle routing problem. Int J Appl Ind Eng 3:55–73. https://​
269(9):1–19 doi.​org/​10.​4018/​IJAIE.​20160​10104
117. Cattaruzza D, Absi N, Feillet D (2018) Vehicle routing problems 136. Shi J, Zhang J, Wang K, Fang X (2018) Particle Swarm Optimi-
with multiple trips. Ann Oper Res 14(3):223–259. https://d​ oi.o​ rg/​ zation for Split Delivery Vehicle Routing Problem. Asia Pac J
10.​1007/​s10288-​016-​0306-2 Oper Res 35(2):18400061840001–18400061840042. https://d​ oi.​
118. Pinto T, Alves C, Carvalho JVd (2020) Variable neighborhood org/​10.​1142/​S0217​59591​84000​67
search algorithms for the vehicle routing problem with two- 137. Campos V, Corberán A, Mota E (2008) A scatter search algo-
dimensional loading constraints and mixed linehauls and back- rithm for the split delivery vehicle routing problem. Adv Comput
hauls. Int Trans Oper Res 27(1):549–572 Intell Transp Logist Supply Chain Manag 144:137–152
119. Cao E-b, Lai M-y (2007) Vehicle routing problem with simulta- 138. Yan SY, Chu JC, Hsiao FY et al (2015) A planning model and
neous delivery and pick-up with time windows. In: 1st interna- solution algorithm for multi-trip split-delivery vehicle routing
tional conference on transportation engineering. and scheduling problems with time windows. Comput Ind Eng
120. Lai M, Cao E (2010) An improved differential evolution algo- 87:383–393
rithm for vehicle routing problem with simultaneous pick- 139. Chen P, Golden B, Wang X, Wasil E (2016) A novel approach to
ups and deliveries and time windows. Eng Appl Artif Intell solve the split delivery vehicle routing problem. Int Trans Oper
23(2):188–195 Res
121. Wang C, Mu D, Zhao F, Sutherland JW (2015) A parallel simu- 140. Li J, Qin H, Baldacci R, Zhu W (2020) Branch-and-price-and-cut
lated annealing method for the vehicle routing problem with for the synchronized vehicle routing problem with split delivery,
simultaneous pickup–delivery and time windows. Comput Ind proportional service time and multiple time windows. Transp Res
Eng 83:111–122 E Logist Transp Rev. https://​doi.​org/​10.​1016/j.​tre.​2020.​101955

13
220 H. Zhang et al.

141. Gu W, Cattaruzza D, Ogier M, Semet F (2019) Adaptive large scheduling problems. Paper presented at the the 6th Int Conf on
neighborhood search for the commodity constrained split deliv- city logistics, Puerto Vallarta
ery VRP. Comput Oper Res. https://​doi.​org/​10.​1016/j.​cor.​2019.​ 160. Qureshi AG, Taniguchi E, Yamada T (2012) A microsimulation
07.​019 based analysis of exact solution of dynamic vehicle routing with
142. Ortega EJA, Schilde M, Doerner KF (2020) Matheuristic search soft time windows. Procedia Soc Behav Sci 39:205–216. https://​
techniques for the consistent inventory routing problem with time doi.​org/​10.​1016/j.​sbspro.​2012.​03.​102
windows and split deliveries. Oper Res Perspect. https://​doi.​org/​ 161. Davis LC (2017) Dynamic origin-to-destination routing of wire-
10.​1016/j.​orp.​2020.​100152 lessly connected, autonomous vehicles on a congested network.
143. Pillac V, Guéret C, Medaglia AL (2012) An event-driven opti- Phys A 478:93–102. https://d​ oi.o​ rg/1​ 0.1​ 016/j.p​ hysa.2​ 017.0​ 2.0​ 30
mization framework for dynamic vehicle routing. Decis Support 162. Wang J, Niu H (2018) A distributed dynamic route guidance
Syst 54(1):414–423 approach based on short-term forecasts in cooperative infrastruc-
144. Hu ZH, Sheu JB, Zhao L et al (2015) A dynamic closed-loop ture-vehicle systems. Transp Res D Transp Environ
vehicle routing problem with uncertainty and incompatible 163. Núñez A, Sáez D, Cortés CE (2008) Hybrid predictive control
goods. Transp Res Part C 55(6):273–297 for the vehicle dynamic routing problem based on evolutionary
145. Abdallah AMFM, Essam DL, Sarker RA (2017) On solving peri- multi objective optimization(EMO). Paper presented at the Proc
odic re-optimization dynamic vehicle routing problems. Appl of the 17th world congress int federation of automatic control.
Soft Comput 55:1–12 The International Federation of Automatic Control, Seoul, Korea
146. De Armas J, Melián-Batista B (2015) Variable neighborhood 164. Potvin J-Y, Xu Y, Benyahia I (2006) Vehicle routing and schedul-
search for a dynamic rich vehicle routing problem with time win- ing with dynamic travel times. Comput Oper Res 33(4):1129–
dows. Comput Ind Eng 85:120–131. https://d​ oi.o​ rg/1​ 0.1​ 016/j.c​ ie.​ 1137. https://​doi.​org/​10.​1016/j.​cor.​2004.​09.​015
2015.​03.​006 165. Fleischmann B, Gnutzmann S, Sandvoß E (2004) Dynamic
147. Kuo RJ, Wibowo BS, Zulvia FE (2016) Application of a fuzzy vehicle routing based on online traffic information. Transp Sci
ant colony system to solve the dynamic vehicle routing problem 38(4):420–433
with uncertain service time. Appl Math Model 40(23–24):9990– 166. Chen H-K, Hsueh C-F, Chang M-S (2006) The real-time time-
10001. https://​doi.​org/​10.​1016/j.​apm.​2016.​06.​025 dependent vehicle routing problem. Transp Res E Logist Transp
148. Ghannadpour SF, Noori S, Tavakkoli-Moghaddam R, Ghoseiri K Rev 42(5):383–408. https://​doi.​org/​10.​1016/j.​tre.​2005.​01.​003
(2014) A multi-objective dynamic vehicle routing problem with 167. Wohlgemuth S, Oloruntoba R, Clausen U (2012) Dynamic vehi-
fuzzy time windows: model, solution and application. Appl Soft cle routing with anticipation in disaster relief. Socioecon Plann
Comput 14(1):504–527. https://​doi.​org/​10.​1016/j.​asoc.​2013.​08.​ Sci 46(4):261–271
015 168. Kilby PPP, Shaw P (1998) Dynamic VRPs: a study of scenarios.
149. Chen S, Chen R, Wang G-G, Gao J, Sangaiah AK (2018) An University of Strathclyde
adaptive large neighborhood search heuristic for dynamic vehicle 169. Chen Z, Xu H (2006) Dynamic column generation for dynamic
routing problems. Comput Electr Eng 67:596–607. https://​doi.​ vehicle routing with time windows. Transp Sci 40(1):74–88.
org/​10.​1016/j.​compe​leceng.​2018.​02.​049 https://​doi.​org/​10.​1287/​trsc.​1050.​0133
150. Angelelli E, Bianchessi N, Mansini R, Speranza MG (2009) 170. Secomandi N (2000) Comparing neuro-dynamic programming
Short term strategies for a dynamic multi-period routing prob- algorithms for the vehicle routing problem with stochastic
lem. Transp Res C Emerg Technol 17(2):106–119 demands. Comput Oper Res 27(11–12):1201–1225
151. Albareda-Sambola M, Fernández E, Laporte G (2014) The 171. Godfrey GA, Powell WB (2002) An adaptive dynamic program-
dynamic multiperiod vehicle routing problem with probabilis- ming algorithm for dynamic fleet management, I: single period
tic information. Comput Oper Res 48:31–39. https://​doi.​org/​10.​ travel times. Transp Sci 36(1):21–39
1016/j.​cor.​2014.​02.​010 172. Novoa C, Storer R (2009) An approximate dynamic program-
152. Ulmer MW, Soeffker N, Mattfeld DC (2018) Value function ming approach for the vehicle routing problem with stochastic
approximation for dynamic multi-period vehicle routing. Eur J demands. Eur J Oper Res 196(2):509–515. https://​doi.​org/​10.​
Oper Res 269(3):883–899. https://​doi.​org/​10.​1016/j.​ejor.​2018.​ 1016/j.​ejor.​2008.​03.​023
02.​038 173. Ichoua S, Gendreau M, Potvin JY (2006) Exploiting knowledge
153. Slater A (2002) Specification for a dynamic vehicle routing and about future demands for real-time vehicle dispatching. Transp
scheduling system. Int J Transp Manag 1(1):29–40 Sci 40(2)
154. Okulewicz M, Mańdziuk J (2017) The impact of particular com- 174. Gendreau M, Guertin F, Potvin JY, Seguin R (2006) Neighbor-
ponents of the PSO-based algorithm solving the dynamic vehicle hood search heuristics for a dynamic vehicle dispatching prob-
routing problem. Appl Soft Comput 58:586–604. https://​doi.​org/​ lem with pick-ups and deliveries. Transp Res C Emerg Technol
10.​1016/j.​asoc.​2017.​04.​070 14(3):157–174
155. Kim S, Lewis ME, White CC (2005) Optimal vehicle routing 175 Azi N, Gendreau M, Potvin JY (2012) A dynamic vehicle
with real-time traffic information. IEEE Trans Intell Transp Syst routing problem with multiple delivery routes. Ann Oper Res
6(2):178–188. https://​doi.​org/​10.​1109/​tits.​2005.​848362 199:103–112
156. Guner AR, Murat A, Chinnam RB (2012) Dynamic routing under 176. Montemanni R, Gambardella LM, Rizzoli AE, Donati AV (2005)
recurrent and non-recurrent congestion using real-time ITS infor- Ant colony system for a dynamic vehicle routing problem. J
mation. Comput Oper Res 39(2):358–373 Comb Optim 10(4):327–343
157. Li J-Q, Mirchandani PB, Borenstein D (2009) Real-time vehi- 177. Hanshar FT, Ombuki-Berman BM (2007) Dynamic vehicle rout-
cle rerouting problems with time windows. Eur J Oper Res ing using genetic algorithms. Appl Intell 27(1):89–99. https://d​ oi.​
194(3):711–727. https://​doi.​org/​10.​1016/j.​ejor.​2007.​12.​037 org/​10.​1007/​s10489-​006-​0033-z
158. Taniguchi E, Shimamoto H (2004) Intelligent transportation sys- 178. Barkaoui M, Gendreau M (2013) An adaptive evolutionary
tem based dynamic vehicle routing and scheduling with variable approach for real-time vehicle routing and dispatching. Comput
travel times. Transp Res C Emerg Technol 12(3–4):235–250. Oper Res 40(7):1766–1776
https://​doi.​org/​10.​1016/j.​trc.​2004.​07.​007 179. Cortes CE, Saez D, Nunez A, Munoz-Carpintero D (2009)
159. Nakamura Y, Taniguchi E, Yamada T, Ando N (2010) Selecting Hybrid adaptive predictive control for a dynamic pickup and
a dynamic and stochastic path method for vehicle routing and delivery problem. Transp Sci 43(1):27–42

13
Review of Vehicle Routing Problems: Models, Classification and Solving Algorithms 221

180. Khouadjia MR, Alba E, Jourdan L, Talbi EG (2010) Multi-swarm 184. Gao S, Wang Y, Cheng J, Inazumi Y, Tang Z (2016) Ant colony
optimization for dynamic combinatorial problems: a case study optimization with clustering for solving the dynamic location
on dynamic vehicle routing problem. In: Swarm intelligence— routing problem. Appl Math Comput 285:149–173
7th international conference, ANTS 2010, Brussels, Belgium, 185. Bent RW, Van Hentenryck P (2004) Scenario-based planning
8–10 Sept 2010. Proceedings. for partially dynamic vehicle routing with stochastic customers.
181. Sabar NR, Bhaskar A, Chung E, Turky A, Song A (2019) A Oper Res 52(6):977–987
self-adaptive evolutionary algorithm for dynamic vehicle routing 186. Mańdziuk J, Żychowski A (2016) A memetic approach to vehi-
problems with traffic congestion. Swarm Evol Comput 44:1018– cle routing problem with dynamic requests. Appl Soft Comput
1027. https://​doi.​org/​10.​1016/j.​swevo.​2018.​10.​015 48:522–534
182. Elhassania M, Jaouad B, Ahmed EA (2013) A new hybrid algo-
rithm to solve the vehicle routing problem in the dynamic envi- Publisher’s Note Springer Nature remains neutral with regard to
ronment. Int J Soft Comput 8(5):327–334 jurisdictional claims in published maps and institutional affiliations.
183. Euchi J, Yassine A, Chabchoub H (2015) The dynamic vehicle
routing problem: solution with hybrid metaheuristic approach.
Swarm Evol Comput 21:41–53. https://​doi.​org/​10.​1016/j.​swevo.​
2014.​12.​003

13