Journal of Physics:

Conference Series

PAPER • OPEN ACCESS You may also like

- Adjustable Silicon Corner Rounding
Logistics Route Optimization Based on Improved Radius by Wet Technique
Pei-Ting Tou, Hsin-Yi Liao, Hui-Chin
Particle Swarm Optimization Huang et al.

- Simulation study of an improved firefly

algorithm for solving multi-dimensional
To cite this article: Zunhai Gao and Hongxia Lu 2021 J. Phys.: Conf. Ser. 1995 012044 complex function optimization
Liping Wang

- Detection of road crack defects based on

improved YOLOv5s model
Pingjun Zhang, Wenwen Li and Yue Weng
View the article online for updates and enhancements.

This content was downloaded from IP address 176.103.18.1 on 20/12/2024 at 22:02

CMSA 2021 IOP Publishing
Journal of Physics: Conference Series 1995 (2021) 012044 doi:10.1088/1742-6596/1995/1/012044

Logistics Route Optimization Based on Improved Particle

Swarm Optimization

Zunhai Gao1 and Hongxia Lu2,*

1
College of Artificial Intelligence, Nanchang Institute of Science and Technology,
Nanchang, China
2
School of Management, Wuhan Donghu University, Wuhan, China

*Corresponding author email: luhx@wdu.edu.cn

Abstract. An improved particle swarm optimization (PSO) algorithm is presented by

dynamically adjusting the inertia weight in the iterative process of PSO, and it is used to solve
the problem of logistics route optimization. This algorithm can not only improve the
convergence speed, but also avoid falling into local optimum. In the process of improving the
standard algorithm, two methods are proposed to adjust the inertia weight value according to
the number of iterations. One is piecewise linear decreasing, another is linear decreasing. The
results show that linear decline is better than piecewise linear decline to achieve the purpose of
optimization, which is more conducive to accelerate the convergence rate and enhance the
ability of optimization. Through the simulation experiment of the specific vehicle routing
optimization problem, the results show that after the improvement, the optimization
performance is enhanced, the optimization speed is accelerated, and the complexity is not
increased, which greatly improves the performance of the original algorithm.
Keywords: Logistics route optimization; Improved PSO algorithm; Inertia weight.

1. Introduction
The basic idea of particle swarm optimization (PSO) comes from the research and simulation of birds'
foraging behaviour. The algorithm is simple in concept, easy to implement, and has strong global
search ability. It is an effective to solve the distribution route optimization problem, but it is easy to
fall into the local optimal solution. In view of this, many improved algorithms have proposed [1-5].
The purpose of logistics path optimization is to find a reasonable distribution scheme with low cost
and high speed among many distribution schemes. Once the problem was put forward, it soon
attracted the attention and research of experts and transportation policy makers. It is a well-known NP
hard problem to optimize logistics distribution routes, including vehicle routing problem (VRP) and
travelling salesman problem (TSP). VRP Problem is defined as: some limited vehicle sets, some
customer points, each customer point has different goods needs, under certain constraints, it is
required to arrange the appropriate distribution scheme, complete the needs of each customer, and
achieve a certain goal (such as the lowest cost, shortest route, fastest speed, etc.). It is not easy to find
the exact solution of VRP Problem. Now the effective algorithm to solve the path optimization is in
constant progress, and there are many discoveries in its theoretical part [6-11], but there are still some
areas in urgent need of improvement and improvement in the operation time and the quality of the
optimal solution. So we present an improved particle swarm optimization (PSO) algorithm by
dynamically adjusting the inertia weight in the iterative process of PSO, and it is used to solve the
problem of logistics route optimization.

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
CMSA 2021 IOP Publishing
Journal of Physics: Conference Series 1995 (2021) 012044 doi:10.1088/1742-6596/1995/1/012044

2. Particle Swarm Optimization and Its Improvement

2.1. Standard Particle Swarm Optimization

In standard particle swarm optimization, the position of each individual is considered as the feasible
solution of the problem. In the solution space, individuals dynamically change their own motion
direction and speed by virtue of their own flight experience and other individuals in the space, until
they find the optimal position, that is, the optimal solution of the problem. With the increase of
iterations, the velocity and position of particles are also changing.
There are some parameters in PSO: balance the global and local search w, algorithm parameters c1, c2,
number of individuals N, speed limit, search times and so on. Now we set up the following: in the
search area of D dimension, there are N individuals in the population, and their positions are as
follows , , ⋯ , ,the speed is , , ⋯ , .Then, the location and velocity of each individual are
initialized randomly, which are all D dimensions. For example, the position of the j-th particle is
, ,⋯, , The velocity of the j-th particle is , ,⋯, .
At the same time, the local optimal value of each individual and the global optimal value of all
individuals are changed according to the fitness value. Let the best location (local best) found by the
individual i-th be , ,⋯, , The best location for all individuals is , ,⋯, .
The position and velocity of the j-th component of particle i are updated according to formulas (1) and
(2), respectively

1 (1)

1 1 (2)
Wherei is the number of the current evolution (i.e. the number of iterations), 1,2, ⋯ , , indicates
that the location and rate of all individuals are updated. c1 and c2 are learning factors (the larger c1 is,
the greater the experience of the representative particles has on the individual, the greater c2, the
greater the experience of the representative group has on each particle, all of which are greater than 0).
r1 and r2 are random numbers distributed evenly between 0 and 1. i determines the effect of the
original rate on the current rate, which is of great significance to balance global search and local
searching ability.

2.2. Improvement of Particle Swarm Optimization

This paper will dynamically change the inertia weight according to the number of search times, so as
to achieve the purpose of optimizing the algorithm. At the beginning of the iteration, taking a larger
inertia weight makes the speed between particles play a greater role, increases the individuality and
diversity of particles, expands the search range of particles, increases the iteration efficiency, and
avoids falling into the local optimum due to the aggregation of particles. With the increase of the
number of iterations, the particles are gradually close to the optimal position. At this time, the smaller
inertia weight is taken to increase the local search ability of particles. Therefore, this paper proposes a
method to dynamically adjust the inertia weight according to the number of iterations, which can not
only enhance its optimization performance, but also prevent it from entering the local optimum. In the
whole iteration process, the inertia weight can be changed from time to time. There are two methods to
change the inertia weight according to the number of iterations (T representing the total number of
iterations):
Improved algorithm 1: = - 100/ (3)
In this way, formula (3) is called to update the value of inertia weight every100 iterations, that is, the
value is reduced by 100/total iterations. In the process of individual search, the inertia weight is
reduced by segments.
Improved algorithm 2: = - 1/ (4)
Every iteration, the formula (4) is used to update the value of inertia weight, that is, reduces it by

2
CMSA 2021 IOP Publishing
Journal of Physics: Conference Series 1995 (2021) 012044 doi:10.1088/1742-6596/1995/1/012044

100/total iterations. With the increase of search times, the inertia weight decreases linearly. In the
whole optimization process, the inertia weight of the above two improved algorithms are reduced by
one unit. Both methods gradually reduce the value of inertia weight in the process of increasing the
number of iterations, but the improved method is piecewise linear decreasing, and the improved
method is linear decreasing.

3. Vehicle Routing Optimization Problem

There are K vehicles participating in the delivery of goods. The maximum length of each vehicle is
Dk 1,2, ⋯ , , What is the loading limit per vehicle 1,2, ⋯ , , Supply goods to
customer L points. The number of goods required for all customer points is 1,2, ⋯ , .
represents the distance from the demand point to the demand point . 1,2, ⋯ , indicates the
distance from the logistics centre to each customer point. Let represent the number of demand
points delivered by the first vehicle, so 0 means that the k-th vehicle did not participate in the
delivery, the demand point of the k-th vehicle distribution route is saved in set , Demand point rki
represents the i-th demand customer in the driving path , 0on behalf of the logistics centre.
Because the first customer point on each delivery route is the logistics centre, so for 1,2, ⋯ , ,
0 holds.
According to different problems and situations, the setting of objective function is different. Most of
them take the shortest route, the fastest speed and the least vehicles as the goal of solving the delivery
route. The objective function of this paper is the shortest delivery route.

F is the sum of the distances of all vehicle delivery routes, and the objective function is to find the
minimum value of F. Let is a 0-1 variable, when the number of demand points on the k-th path
is not 0.That is, when 0, the variable is 1. When there is no demand point on the k-th path,
i.e. 0, this variable is set to 0. The accumulation formula realizes the accumulation of k from 1 to
K to get the sum of the length of all delivery routes.
The sum of the distances from the i-1 customer point to the i ( 1,2, ⋯ , ) customer point delivered
by the vehicle k plus the distance from the customer point to the logistics center is shown in
brackets. That is the total length of route K. Therefore, the 0-1 variable is used to ensure that when the
vehicle participates in the delivery, the length of the delivery route is added to F. If you do not
participate in the delivery, you can directly make the route length of the delivery 0 and do not add it to
F. With 0-1 variable, the objective function formula is more concise.

3.1. Constraint Condition

(1) The sum of the demand of all customer points on each delivery route shall not be greater than the
loading limit of vehicles on the route. represents the demand of goods at the i-th customer point
on the delivery route k. The accumulation formula can traverse all customer points i delivered by
vehicle k from 1 to . is the maximum loading capacity of vehicles for route k.

. .

(2) The length of each delivery route shall not be greater than the limit of the route for the delivery
vehicles. k in formula (5) is each value between 1 and K. on the right side of the inequality
represents the maximum route of the kth car. On the left side of the inequality is the sum of the
distances from the i-1-th customer point to the i-th ( 1,2, ⋯ , ) customer point delivered by the k-
th car plus the distance from the customer point to the logistics centre. The 0-1 variable is
also used here to simplify the formula. When the vehicle does not participate in the delivery, the limit
of its driving length will not be considered.
∑ (5)

3
CMSA 2021 IOP Publishing
Journal of Physics: Conference Series 1995 (2021) 012044 doi:10.1088/1742-6596/1995/1/012044

(3) The total number of customer points delivered by each vehicle is less than the total demand points.
represents the number of demand points delivered by the k-th vehicle, so that the value is in the
range of 0 to L.
0
(4) A collection of all customer points on each path. The demand points on the k-th vehicle
distribution route are stored in the set . When 1,2. . . , represents the entire customer points
on the k-th vehicle distribution route. Of course, we need to satisfy the customer point between 1 and L.
| ∈ 1,2, ⋯ , , 1,2. . .
(5) It is necessary that the requirements of each customer point are met. nk is the number of customer
points delivered by the first vehicle, so when 1,2, ⋯ , , the sum of the total number of customer
points delivered on all routes is solved. To make this value equal to L.

(6) Each customer point can only be delivered by one vehicle. For different k, there is no intersection
in the corresponding customer point set .
∩ ,∀

3.2. Modeling
The mathematical model is shown in Table 1.
Table 1. Mathematical model
Objective function:
Solving the shortest distribution route
The sum of demand on each delivery route shall not be
. .
greater than the loading limit of vehicles on the route
The length of each delivery route shall not be greater than
the limit of the route for the delivery vehicles
The total number of customer points delivered by each
0
vehicle is less than the total demand points
| ∈ 1,2. . . , 1,2. . . A collection of all customer points on each path
The needs of each customer point are met

∩ ,∀ Each customer point can only be delivered by one vehicle

4. Application of Improved Particle Swarm Optimization in Vehicle Routing Problem

4.1. Algorithm Thinking

(1) Encoding particles. There are n demand points, and each demand point corresponds to a two-
dimensional vector. In this space, particle i represents demand point 1,2, ⋯ , ,its corresponding
vector represents the vehicle number of distribution customer point i, and represents the order
of this customer point iin the process of vehicle driving (customer point i is the demand point
delivered by the vehicle with vehicle number ).
(2) Decoding particles. Corresponding relationship between distribution vehicles and number: round
the component of particlei, then the vehicles j assigned by logistics centre for demand point ican be
obtained. The order of customer points passed by vehicle j: using the circulation, find all the demand
points i passed by distribution vehicle j, sort and label the value of from small to large, and
determine the order of customer points passed by vehicle j.
(3) Selection fitness function. Fitness value is used to judge the performance of particles, and the
value is determined by its function. For this algorithm, the fitness function can be expressed as
formula (6).

4
CMSA 2021 IOP Publishing
Journal of Physics: Conference Series 1995 (2021) 012044 doi:10.1088/1742-6596/1995/1/012044

∑ ∑ (6)

(4) The termination condition of the algorithm. The end condition of the algorithm is set to meet the
maximum number of searches.

4.2. Algorithm Steps

(1) Setting parameters. There are mainly learning factors , , , ,the number of individuals
and the highest number of iterations .
(2) Initial population. The random coding particle population initializes the order of the distribution
vehicles and the customer points in the distribution route of each particle.
(3) Information decoding. The location information of each customer point is decoded to generate a
path selection method, which can solve the fitness value of each customer point under multiple
constraints.
(4) Update the optimal value. For each customer point that needs to be delivered, after updating its
location information, it determines whether to update its own optimal experience value and group
optimal experience value according to the fitness value.
(5) Update particle attributes. According to the improved formula, the parameters w , x and each
individual v are updated.
(6) Cycle judgment. Determine whether the search times meet the maximum search , If yes, the
optimal position of the particle is output. If no, return to step (3).
(7) Information decoding. Decoding is called for the location of each customer point, so the best
dispatch scheme can be obtained.

5. Simulation Experiment
Analyse and code the path optimization selection of 1 centre, 2 vehicles (each vehicle's loading
capacity limit is 15 tons) and 8 demand points. Distance between demand points and demand
1,2, ⋯ ,8 . Fill in Table 2 (0 for logistics centre point, 1 to 8 for customer point)
Table 2. Distance between customer points and their demands
Demand distance /km
point 0 1 2 3 4 5 6 7 8
0 0 5.5 4.0 7.5 8.0 13.5 11.0 15.0 9.0
1 5.5 0 7.5 8.0 10.5 5.0 8.0 10.0 11.0
2 4.0 7.5 0 8.5 10.0 12.0 7.5 8.0 8.5
3 7.0 8.0 8.5 0 10.0 5.0 9.0 10.0 12.0
4 8.0 10.5 10.0 10.0 0 10.0 7.5 7.5 10.0
5 13.0 5.0 12.0 5.5 10.0 0 8.0 10.0 7.5
6 11.0 8.0 7.5 9.0 7.5 8.0 0 7.0 12.0
7 15.0 10.0 8.0 10.0 7.5 10.0 7.5 0 10.0
8 9.0 11.0 8.5 11.0 10.0 7.5 12.0 10.0 0
qi 2 1 2 2 2 1 4 1
PSO, improved method 1 and improved method 2 were used to calculate the individual. In order to
compare the performance of each method more conveniently, the parameters are set as follows:
particle population is 8, learning factor c1 = 2.05, c2 = 2.05, the initial value of inertia weight w is 1.5,
r1 and r2 are random numbers between 0 and 1. For 1000 searches, the independent calculation is 10.
The statistical results are shown in Table 3
Table 3. Results of each method running ten times separately
Algorithm 1 2 3 4 5 6 7 8 9 10
Standard PSO 74 77.5 72 82.5 70 69 71 75.5 72 70
Improved algorithm 1 70.5 73 71 69 71 70 68 68 72.5 67
Improved algorithm 2 71 68.5 71 69 67 67.5 67 70.5 69 67
The results of Table 3 are summarized and table 4 is obtained.

5
CMSA 2021 IOP Publishing
Journal of Physics: Conference Series 1995 (2021) 012044 doi:10.1088/1742-6596/1995/1/012044

Table 4. Relevant data of each algorithm running ten times

Algorithm Average value Maximum Minimum value
Standard PSO 73.35 82.5 69
Improved algorithm1 70.0 73 67
Improved algorithm2 68.75 71 67
According to the results in Table 3 and table 4, when all parameters except inertia weight are the same,
the results of the two improved algorithms in the example are better than those of the standard
algorithm, and the convergence performance is better, and the linear decreasing inertia weight is better
than the piecewise linear decreasing inertia weight. On the basis of the improved algorithm, it meets
the requirements of balancing individual optimization performance.

6. Conclusion
A method to dynamically adjust the inertia weight according to the number of iterations is presented.
Considering the individuality and colony of particles, the inertia weight is dynamically changed
according to the iteration of the algorithm in the process of particle search, so as to improve the
convergence speed of the algorithm, increase the diversity of the particle population, and avoid the
premature phenomenon due to the strong individuality of the individual and falling into local optimum.
This algorithm is used to solve the logistics distribution path optimization problem. Through the
simulation experiment of the specific vehicle routing optimization problem, the results show that after
the improvement of the algorithm, the optimization performance is enhanced, the optimization speed
is accelerated, and its complexity is not increased, which greatly improves the performance of the
original algorithm.

Acknowledgments
Authors wishing to acknowledge financial support of the start-up fund for doctoral research in
Nanchang Institute of technology.

References
[1] Guo Kai and Lv Yang 2020 Optimizing Routing Path Selection Method Particle Swarm
Optimization International Journal of Pattern Recognition and Artificial Intelligence 34(12)
[2] Si Han 2020 A new intelligent method for travel path recommendation based on improved
particle swarm optimization International Journal of Computing Science and Mathematics 12(1)
[3] Xun Li and Dandan Wu et al 2020 An Improved Method of Particle Swarm Optimization for
Path Planning of Mobile Robot Journal of Control Science and Engineering 2020
[4] Qiong Gao 2018 Intelligent Vehicle Global Path Planning Based on Improved Particle Swarm
OptimizationOpen Access Library Journal5(4)
[5] Zong-ran Dong and Yan Lin 2017 A particle swarm optimization based approach for ship pipe
route design International Shipbuilding Progress 63(1-2) pp 59-84
[6] Puspitasari F H and Kurniawan V R B 2021 Designing Optimal Distribution Routes using a
Vehicle Routing Problem (VRP) Model in a Logistics Service Provider IOP Conference Series:
Materials Science and Engineering 1071(1)
[7] Yonghao Zeng 2019 Improved Particle Swarm Optimization Based on Tabu Search for VRP
Journal of Applied Science and Engineering Innovation 6(2)
[8] Florian Arnold and Kenneth Sörensen 2019What makes a VRP solution good? The generation
of problem-specific knowledge for heuristics Computers & Operations Research106pp280-288
[9] Tao Shuaihan 2021 An Improved genetic algorithm for solving TSP Journal of Physics:
Conference Series 1952(4)
[10] Sharma Sachin and Jain Vinod 2021 A Novel Approach for Solving TSP Problem Using
Genetic Algorithm Problem IOP Conference Series: Materials Science and Engineering 1116(1)
[11] Hansknecht Christoph and Joormann Imke et al 2021 Dynamic Shortest Paths Methods for the
Time-Dependent TSPAlgorithms 14(1)