Next Article in Journal
Developed Design of Battle Royale Optimizer for the Optimum Identification of Solid Oxide Fuel Cell
Previous Article in Journal
Drivers of Environmental Conservation Agriculture in Sado Island, Niigata Prefecture, Japan
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Vehicle Routing Problem for the Simultaneous Pickup and Delivery of Lithium Batteries of Small Power Vehicles under Charging and Swapping Mode

School of Information, Beijing Wuzi University, Bejing 101149, China
*
Author to whom correspondence should be addressed.
Sustainability 2022, 14(16), 9883; https://doi.org/10.3390/su14169883
Submission received: 12 June 2022 / Revised: 7 August 2022 / Accepted: 8 August 2022 / Published: 10 August 2022

Abstract

:
Due to the national policy of encouraging the development of power exchange modes, the reasonable planning of vehicle distribution paths to meet the demand of lithium battery power exchange points has become a topic of considerable research interest. In this study, we propose the “centralized charging + unified distribution” power exchange mode for optimizing the charging and transporting of lithium batteries. Considering lithium batteries are dangerous goods, the vehicle path problem of simultaneous pickup and delivery of lithium batteries with vehicle load and soft time window constraints is studied. The model objective is to minimize the transportation risk and total cost of delivery. By performing crossover and mutation operations on the initial solutions generated by the ant colony algorithm, a hybrid ant colony genetic algorithm (ACO-GA) is designed to solve the model. The results of ACO-GA are compared with the GA, ACO, and SAA methods using the Solomon dataset; the results show that the optimized ant colony algorithm can achieve a smaller total cost in solving the model. Finally, taking a lithium battery leasing business in Company A, we determine the optimal path under different preferences by setting different weights for distribution cost and transportation risk in the model transformation, which provides a reference for the company to select the distribution route. Thus, the model provides a reference for companies that intend to develop power exchange businesses.

1. Introduction

With the strengthening of the concept of healthy travel and the further improvement and supplementation of urban infrastructure, the urban micro-travel tools market has continued to expand, and micro-travel tools such as electric bicycles have become common, safe, and reliable means of transportation. In 2020, the ownership of electric bicycles in China was approximately 320 million units [1]. Electric bicycle batteries are mainly run on lead-acid batteries and lithium batteries. Driven by factors such as product technology advancement and environmental protection requirements for green travel, the sales of lithium batteries for two-wheeled electric vehicles have increased significantly from 5.4% in 2016 to 23.4% in 2021 [2]. Lithium batteries offer many advantages, such as long life, being light weight, and having a high energy density. To ensure the safe and standardized usage of electric vehicles, the New National Standard stipulates that the weight of electric bicycles should not exceed 55 kg [3]. The capacity of the battery is limited by the weight of the vehicle, making the farthest travel distance with a single battery limited to less than 60 km, which requires frequent charging operations of the battery during use, subsequently increasing the demand for power exchange [4].
By the end of 2020, 1500 highway service areas (including parking areas) in 31 provinces in China had been replaced by charging and replacement facilities [5]. New infrastructure such as charging piles and replacement stations are also under active construction. With the continuous development of the urban express service industry, the demand for lithium battery leasing and replacement from takeaway, express, and other groups has gradually increased. However, this increase in usage of green energy is not without risks. In 2021, nearly 18,000 fires and 57 deaths were caused by electric bicycles and their battery failures [6]. The high incidence of accidents is mainly concentrated in self-built houses, residences, and stores along the streets, with overcharging and wire short-circuiting being the root causes of the fires. As a result of policy and market demand, the electric vehicle power exchange model is once again being promoted and developed. There are two main types of power exchange modes: the centralized charging mode and the battery swap mode [7]. The centralized charging mode refers to the use of charging stations to store and charge a large number of batteries in a centralized manner; then, unified distribution is provided to each demand point as well as damaged battery replacement services. The battery swap mode refers to the use of the exchange station as a carrier to provide users with battery charging and battery replacement services. Both of these two modes require the construction of a large number of charging stations (piles) as carriers and need to solve a series of problems such as voltage and land. As a result, the “centralized charging + unified distribution” mode was created, that is, using centralized charging stations to store and charge a large number of batteries centrally, and then arranging vehicles to collect the fully charged batteries for distribution. In this mode, we can choose to build large-scale centralized charging stations in the suburbs, and then carry out unified distribution, which can alleviate the power and land pressure caused by the construction of a large number of urban exchange stations [7].
Taking a lithium battery leasing business in Company A, we propose the “centralized charging + unified distribution” power exchange mode (see Figure 1) for optimizing the charging and transporting of lithium batteries. Since lithium batteries belong to the ninth category of dangerous goods in the “United Nations Model Regulations on the Transport of Dangerous Goods”, we add the consideration of transportation risks to the objective function; the vehicle path problem of simultaneous pickup and delivery of lithium batteries with vehicle load and soft time window constraints is studied; the model objective is to minimize the transportation risk and total cost of delivery. This model is intended to help companies optimize delivery routes while taking into account safety and cost, reducing delivery costs and transportation risks. At the same time, the battery swap mode of “centralized charging + unified distribution” will be developed to improve the mileage anxiety of small power vehicles.
The main contributions of this study are as follows:
  • We first recommend a “centralized charging + unified distribution” power exchange mode, which can help companies take into account safety and cost, optimize their distribution routes, and reduce distribution costs and transportation risks.
  • We proposed a dual-objective model to minimize delivery cost and transportation risk, and study a time-windowed vehicle routing problem for simultaneous pickup and delivery of dangerous goods.
  • We improved the traditional ant colony algorithm and added the crossover and mutation operations of the genetic algorithm. Based on the R-type data in the Solomon dataset, the effectiveness of the proposed algorithm is verified by comparing the effects of the improved algorithm with the ant colony algorithm, genetic algorithm, and simulated annealing algorithm.
  • Change the dual-objective model to a single-objective model by giving different weights to distribution cost and transportation risk. We also explore the optimal distribution paths under different weight combinations to provide decision-making reference for companies with different risk preferences.

2. Literature Review

In 1959, Dantzig and Ramser [8] first proposed the Vehicle Routing Problem (VRP) for logistics distribution. The problem states that a certain number of customers require picking up and delivering goods; the distribution center provides goods to customers and arranges a fleet to provide pickup and delivery services for these customers. The distribution center needs to arrange appropriate driving routes in advance that can meet the needs of customers under certain constraints and achieve goals such as the lowest vehicle mileage, lowest total cost, and smallest number of vehicles used. The VRP is an NP-hard problem, and the solution algorithm contains exact and approximate algorithms. The exact algorithms include [9]: branch and bound (BB), cutting plane method (CPM), dynamic programming (DP), and more. Approximation algorithms can be divided into metaheuristic-based algorithms such as [10]: tabu search algorithms (TSA), simulated annealing algorithm (SAA), genetic algorithm (GA), population intelligence-based algorithms, ant colony algorithm (ACO), glowworm swarm optimization (GSO), particle swarm optimization (PSO), and more. When the size of the problem is small, the exact algorithm can find the optimal solution in an acceptable time. When the problem size is large, heuristic algorithms are more suitable.
(1)
VRP with Time Windows
The vehicle path problem with time windows (VRPTW) is an extension of the classical path problem, which was reviewed and expanded in detail by Solomon et al. [11]. VRPTW generally refers to vehicles with weight constraints that provide delivery or pickup services within a time specified by the customer [9]. Common time window path planning problems consist of: the vehicle routing problem with hard time windows (VRPHTW), the vehicle routing problem with soft time windows (VRPSTW) and the vehicle routing problem with mixed time windows (VRPMTW). As early as 1992, Koskosidis et al. [12], in their study of VRPTW, regarded the time window constraint as a soft constraint that can violate the cost. The research problem in this study also belongs to the soft time window problem. If the goods are delivered within the time interval requested by the customer, then no penalty cost is incurred, and if they are delivered earlier or later than the time interval requested by the customer, then a penalty is imposed depending on the length of the delay (or early arrival). Soft time windows can effectively reduce the delivery cost without significantly reducing customer satisfaction, which is more realistic [13]. Existing studies on VRPTW are shown in Table 1.
The vehicle path problem with time window constraints has been studied in different application scenarios, expanding it to HFGVRPTW, MDOVRPTW, etc. Most of the literature takes cost minimization as the objective function. In terms of solution algorithms, most of them are heuristic-based and improve the existing heuristic algorithms for specific problems to obtain faster convergence and smaller cost.
(2)
VRP with simultaneous pickup and delivery
The vehicle path problem with simultaneous pickup and delivery (VRPSPD) belongs to the vehicle path problem with backhaul pick (VRPB) [22]. In 1989, Min [23] first proposed the VRPSPD problem and defined it as follows: all demand points are set to have pickup and delivery demands, the vehicles are loaded in the warehouse and arrive at each demand point in order from the warehouse, but finally return to the warehouse after satisfying the pickup and delivery demands of each demand point. When studying VRPSPD, Reil et al. [24] considered the three-dimensional loading constraints, aimed at minimizing the total driving distance, based on the principle of first packing and then routing, and considered the influence of unloading and reloading as appropriate, and finally used the TSA to find the most optimal path.
By reviewing the literature, we found (see Table 2) that algorithms are most often used to solve the VRPSPD problem [25]. Some scholars also combine the metaheuristic algorithm with the variable neighborhood search (VNS) [26,27] or the variable neighborhood descent algorithm (VND) [28] to obtain a hybrid metaheuristic algorithm for optimizing the capability of local search.
(3)
Vehicle routing problem with simultaneous pickup and delivery with time window constraints (VRPSPDTW)
In actual logistic services, time window and pickup and delivery services are realistic requirements that need to be met; hence the vehicle routing problem with simultaneous pickup and delivery with time window constraints (VRPSPDTW) has received increasing attention [35]. Existing studies on VRPSPDTW are shown in Table 3.
In analyzing the existing literature, it can be seen that in VRPSPDTW, the most basic constraint is the vehicle load constraint or the distance constraint. The objective function is mainly the multi-objective function, and total cost minimization is the main goal of optimization. The cost includes distribution, time window, and distance cost. In terms of solving algorithms, heuristic algorithms are often used to solve the model. This study researches the vehicle path planning problem from the centralized charging station to the lithium battery exchange stations under the “centralized charging + unified distribution” mode. It is necessary to consider the time window requirements of each site, and the simultaneous pickup and delivery requirements, so it is a VRPSPDSTW problem.
(4)
Research on the path problem of dangerous goods distribution and the solution method
The dangerous goods distribution problem refers to the organization of a suitable transportation route (e.g., minimum risk to personnel, shortest distance, lowest cost, minimum time, etc.) for a series of loading and unloading points for dangerous goods, so that the vehicles transporting these goods can pass through in an orderly manner and achieve certain optimal goals under certain constraints (such as transportation volume, speed, cycle time, and acceptable risk criteria) [43].
Due to the flammable and explosive nature of hazardous materials, accidents during the transportation of these materials can be extremely dangerous. Therefore, in the actual transportation process, it is necessary to increase the consideration of risks when transporting dangerous goods in order to ensure transportation safety. Erkut et al. [44] summarized eight common methods to measure transport risk: traditional risk, population coverage, accident rate, perceived risk, expectation-variance risk, minimum-maximum risk, negative utility risk, and conditional risk. Also studied was the vehicle route optimization problem of hazardous materials transportation. The research on the hazardous materials path problem advanced from a single-objective problem to a multi-objective hazardous materials vehicle path problem, especially in recent years. The research objectives now encompass transportation cost, transportation risk, transportation distance, transportation time, accident rate, and transportation loss [45]. Kara et al. [46] first proposed a bi-optimal objective model considering both transportation risk and transportation cost and converted the bi-objective model into a single-objective model using Kuhn–Tucker conditions and complementary slackness conditions. Zografos et al. [47] proposed a two-objective hazardous materials transportation model and gave a simplified transportation risk formula, which measured transportation risk in terms of conventional risk. The proposed bi-objective model was later converted into a single-objective model using a linear weighting approach. Androutsopoulos et al. [48] considered time constraints and used an improved insertion method to solve the model with the objective of minimizing transportation costs and risks when transporting chemical supplies. Chai et al. [49] first simplified the process of quantifying the risk factors of vehicles passing through densely populated areas, and used the small number of vehicles, the total transport distance, and the shortest driving distance through densely populated areas as the objective function in constructing the model and solved it using an improved genetic algorithm. Zhang et al. [50] established a route optimization model for dangerous goods transportation vehicles that simultaneously minimizes the maximum accident consequences and transportation costs and designed the exact algorithm for solving the model based on the ε-constraint method. Li et al. [51] constructed a mathematical model with the objectives of minimizing transportation cost and transportation risk, and the transportation risk was measured by the number of people affected by the accident, after which the improved NSGA-II algorithm was applied to solve the model.
In general, research on the vehicle path problem focuses on two aspects: model construction and solution algorithms. In terms of model construction, scholars have studied the vehicle path problem from two aspects: adding constraints or increasing the number of objective functions. The constraints mainly focus on time constraints and vehicle volume constraints, and the objective functions can be divided into single-objective and multi-objective, of which the main objective is the minimization of total cost. In terms of solution algorithms, most scholars use heuristic algorithms, among which ant colony algorithms have been widely used. However, studies on the simultaneous pickup and delivery vehicle routing problem ignore the inherent dangers of transporting goods. Combining the characteristics of lithium battery dangerous goods and the service time constraints of the power exchange point, this study adds the consideration of “transportation risk” when building the model and studies the routing problem of vehicles for simultaneous pickup and delivery of dangerous goods with on-board capacity and time window constraints in detail.

3. Basic Model

3.1. Problem Description and Model Assumptions

In view of the current battery exchange mode, this study proposes a “centralized charging + unified distribution” battery exchange mode, with a specific operation flow as seen in Figure 2. In other words, there is a centralized charging station and multiple lithium battery exchange stations in a particular area, and the centralized charging station provides a battery replacement service by centralized storage, centralized charging, and unified distribution for a large number of batteries.
The problem can be specifically described as follows: there exists a centralized charging station and multiple lithium battery exchange stations in a certain area, and the centralized charging station is responsible for providing pickup and delivery services of lithium batteries from the lithium battery exchange stations in the area. The locations of the centralized charging station and the battery exchange stations are determined each time a distribution vehicle with a certain load capacity departs from the centralized charging station, delivers the fully charged batteries to each lithium battery exchange station during the service time, and then transports the lithium batteries that need to be charged back to the centralized charging station for charging. If the delivery vehicle does not deliver the lithium battery within the service time required by the lithium battery exchange station, a corresponding penalty cost will be paid. In this case, the following factors are known: the location and number of centralized charging stations and each lithium battery exchange station, the demand for lithium batteries the day before delivery, the service time limit, the number of vehicles, the vehicle travel speed, and information related to transportation risk. The objective of the problem is to reasonably arrange vehicles as well as driving routes under certain constraints to satisfy the demand for all lithium battery exchange stations while minimizing distribution costs and transportation risks. The following assumptions are made to facilitate the study.
  • The “centralized charging + unified distribution” mode of power exchange has been applied on a large scale in a certain area, and the number of batteries can meet the demand of all lithium battery exchange stations.
  • There is a centralized charging station (24 h charging service) in the region, the number of vehicles is certain, and the starting and ending points of vehicle distribution are centralized charging stations.
  • The location of each lithium battery exchange station and the demand for picking up and delivering lithium batteries are known before the vehicle departs.
  • The distance between the centralized charging station and each lithium battery exchange station, the service time, the accident probability of each road section, the accident impact radius, the population density, and the number of people affected are known.
  • Each vehicle has the same specification and maximum load capacity. Assume that the number of vehicles is three, and the driving speed is constant at 60 km/h.
  • Each lithium battery exchange station is delivered and picked up by only one vehicle and the pickup and delivery demand is met at one time.
  • The lithium batteries can be mixed and delivered with the same specifications, and the actual loading capacity of each vehicle cannot exceed the maximum loading capacity of the vehicle.
  • Each lithium battery exchange station has a designated service window, and the distribution vehicle needs to provide service within this timeframe.
  • The service time of the distribution vehicle at the lithium battery exchange station is not related to the distribution volume.
  • The distribution cost only considers variable cost (proportional to the number of miles driven by the vehicle) and time window penalty cost. Fixed costs (constants, including driver’s salary, vehicle insurance, etc.) are not considered for the time being.

3.2. Model Constraint

For the lithium battery distribution business of Company A, the path planning model is constructed with the objective of minimizing distribution costs and transportation risks. The multi-objective hazardous material vehicle path problem can reflect the trade-off between cost as well as risk and is closer to the actual research situation. The mathematical notations used in this study are listed in Table A1. The specific constraints are explained and taken as follows:
(1)
Distribution cost
The distribution cost includes variable cost and time window penalty cost, where the variable cost includes the fuel cost and maintenance cost of each operating vehicle. The variable cost is related to the unit transportation cost of the vehicle and the transportation distance; the longer the distance the vehicle travels, the higher the fuel cost required. Assuming there are K ¯ vehicles serving all lithium battery exchange stations, the variable cost of one of the vehicles in the distribution process can be expressed as f 1 :
f 1 = C 1 i N j N k K x i j k d i j   i N , k K
where C 1 denotes the transportation cost per unit of distance of the distribution vehicle, d i j denotes the distance from lithium battery exchange station i to j , and x i j k denotes whether the distribution vehicle k drives from lithium battery exchange station i to j .
In the actual process of distributing lithium batteries, different lithium battery exchange stations have different requirements for the delivery service time. If the vehicle completes the delivery within the expected time of the lithium battery exchange station, the penalty cost is 0. If the vehicle arrives earlier than the expected service time of the lithium battery exchange station, the vehicle must wait at the lithium battery exchange station and has to pay an additional waiting cost. If it arrives later than the desired delivery time of the lithium battery exchange station, the exchange station imposes a penalty cost to encourage the delivery vehicle to complete the service within the desired service time. In this study, a soft time window constraint was applied to the penalty cost incurred by the delivery vehicle k arriving at the lithium battery exchange station j, as shown in Equation (2).
f 2 = { C 2 ( E T j T j k ) , T j k < E T j     0 , E T j < T j k < L T j C 3 ( T j k L T j ) , L T j <   T j k
where [ E T j , L T j ]   denotes the best service time range required by the lithium battery replacement station j . T j k denotes the moment the delivery vehicle k arrives at the lithium battery swap station   j . ( E T j T j k ) denotes the length of time that the vehicle   k   arrives early at the lithium battery exchange station j . C 2   denotes the unit penalty cost incurred for arriving earlier than the optimal service time required by the lithium battery exchange station   j . ( T j k L T j )   denotes the delayed arrival of vehicle   k at the lithium battery exchange station j for the length of time. C 3 denotes the unit penalty cost incurred for arriving later than the optimal service time required by the lithium battery exchange station   j .
In summary, the distribution cost of the distribution vehicle is shown in Equation (3).
f 3 = f 1 + f 2
(2)
Transportation risk
Transportation risk, i.e., the impact of possible accidents during transportation on people, property, and the environment, depends on the rate of transportation accidents and the consequences of the impact of accidents [52]. In the vehicle path problem of hazardous materials transportation, the risk factors are mostly random and uncertain, which makes the process of quantification complicated, and no unified evaluation model has emerged to address this so far. Measuring transportation risk by personnel risk is still a widely used method by scholars [53]. The equation is: Personnel risk = Probability of accident × Number of people affected within a given distance [54], defined as Equation (4).
R i j = p i j P i j
where R i j   denotes the risk of people in the section of road from the lithium battery exchange stations i to j . p i j   denotes the probability of an accident from lithium battery exchange stations   i to j . P i j   denotes the number of people that can be affected in the area in case of an accident in the section of road from lithium battery exchange stations i to   j . The specific definition is given in Equation (5).
P i j = 2 d i j λ ρ i j
where d i j   denotes the distance from lithium battery exchange stations i to   j . λ denotes the radius of the area that can be affected in case of an accident in the section of road from lithium battery exchange stations i to   j . ρ i j denotes the population density from the lithium battery exchange stations i to j. The values of p i j and ρ i j are randomly generated similarly as in [43].   p i j   is taken in the range [1, 2], ρ i j   is taken in the range [300, 600] (in people), and the value of λ is 0.1 km.

3.3. Multi-Objective Model

The objective of this study is to minimize distribution cost and transportation risk, which is part of the multi-objective model. The solution methods for multi-objective problems mainly include the weighting method, hierarchical solution method, and multi-objective evolutionary method (e.g., NSGA-II). Referring to the literature [48], in this study, distribution cost and transportation risk are transformed into a single-objective function using a weighting method. The problem of setting the weighting coefficients of it can be determined based on the importance of distribution cost and transportation risk of company A (the weighting coefficients of both need to satisfy the summation equal to 1).
Target model:
m i n Z = ω 1 C 1 i N j N k K x i j k d i j + ω 1 [ C 2 i N j N k K x i j k m a x { ( E T j T j k ) , 0 }   + C 3 i N j N k K x i j k m a x { ( T j k L T j ) , 0 } ]               + ω 2 i N j N R i j
ω 1 + ω 2 = 1
Load capacity constraint:
f w a g o n i k + j N x i j k ( p i v i ) L , i , j N , k K
Vehicle number constraints:
i N k K x O i k K  
i N k K x O i k = j N k K x j O k
Demand non-splittable constraints:
i = 1 N ¯ x i j k = y j k , j N , k K
j = 1 N ¯ x i j k = y i k , i N , k K
Delivery time continuity constraint:
T j k = T i k + t i j + s i k , i , j N , k K
Variable Constraints:
x i j k { 0 , 1 } , i , j , k
y i k { 0 , 1 } , i , k
Sub-tour elimination constraint:
i N j N k K x i j k | S | 1 , S { 1 , 2 , , N ¯ + 1 } , 2 | S | N ¯
Constraint (6) indicates that the total distribution cost and transportation risk are minimized, and consists of three items, which are the distribution cost, the penalty cost for early or late delivery time, and the transportation risk. The second term of max (x, 0) indicates that the penalty cost is 0 if the vehicle arrives within the best service time required by the lithium battery exchange station, otherwise, either early or late arrival will incur the corresponding penalty cost. Constraint (7) indicates the weight relationship, and the sum of the two equals 1. Constraint (8) indicates that the actual weight of the vehicle k after completing the task of delivering or picking up lithium batteries at each lithium battery exchange station cannot exceed the maximum weight of that vehicle. Constraint (9) ensures that the number of vehicles departing from the centralized charging station does not exceed the maximum number of vehicles. Constraint (10) ensures that the vehicle departs from and eventually returns the centralized charging station, and the number of departure vehicles is equal to the number of return vehicles. Constraints (11) and (12) indicate that each lithium battery exchange station has and can only be delivered by one vehicle, and there must be a path connected to it when the demand is not split and the lithium battery exchange station is served by the same vehicle. Constraint (13) ensures continuity in the distribution process. Constraints (14) and (15) represent variable constraints. Constraint (16) means that the vehicle only visits each lithium battery exchange station once.

4. Ant Colony Genetic Hybrid Algorithm Design and Verification

4.1. Algorithm Design

To address the problems of traditional heuristics such as failure to converge and the tendency to fall into local optimality, optimizing a single heuristic algorithm or combining multiple heuristics into a hybrid metaheuristic algorithm can improve the solving ability. From reviewing the literature, it can be seen that the two algorithms, ant colony and genetic, have a wide research base and have relative advantages in solving vehicle path optimization problems. The ant colony algorithm can produce good quality solutions with a small number of iterations, while the genetic algorithm has a good global search capability, which can effectively solve the problems faced by the ant colony algorithm in the local optimization problem. Therefore, this study proposes the ant colony genetic algorithm to solve the model.
The basic design idea is shown in Figure 3; the ant colony algorithm is used as the main body, and the feasible solution is first output as a new population by the algorithm. Then, the fitness function is used to evaluate the new population; that is, the objective function is used to judge the quality of the individuals in the population. Since the objective function in this study is to minimize the delivery cost and risk, the inverse of the objective function is taken as the value of the fitness function. The solution corresponding to the minimum value of the objective function is chosen. Then, the solution is inserted into the genetic algorithm; the crossover and variation operators of the genetic algorithm are used to further optimize the better solution obtained by the ant colony algorithm, so as to expand the search space. In the case of satisfying the termination condition, the convergence speed of the algorithm is accelerated to improve the quality of the solution.
These are some explanations for the specific steps:
(1)
Initialization parameters. Set the initial value of the iteration number iter = 1 and the maximum iteration number iter_max = 100. The definitions and values of other parameters ( m , , β , ρ , Q , P c , P m ) are shown in Table 4.
(2)
Pheromone updating. Set the initial pheromone of all locations to 1, so that the ants have the same probability of crawling to each location. Calculate the length path traversed by the ants, record the current optimal solution, and update the pheromone.
(3)
Transfer of ants. When choosing the next place to visit, the ants will use the pheromone concentration on each connection path as a reference. P i j k ( t ) denotes the probability that ant k moves from point i to j at time t.
P i j k ( t ) = { [ τ i j ( t ) ] α · [ η i j ( t ) ] β s allow k [ τ i s ( t ) ] α · [ η i s ( t ) ] β , s allow k 0 , s allow k
η i j ( t )   denotes the heuristic function, the value is the reciprocal of the distance between points i and j, which represents the expected degree of ants i   transferring from point i to j. allow k ( k = 1 , 2 , , m ) denotes the set of all locations except the departure location of ant k, including (n − 1) elements. It can be seen from Equation (18) that after the ants release the pheromone, the pheromone concentration between the sites decreases with time.
{ τ i j ( t + 1 ) = ( 1 ρ ) τ i j ( t ) + Δ τ i j Δ τ i j = k = 1 n Δ τ i j k , 0 < ρ < 1
Δ t i j denotes the sum of the pheromone concentrations released by all ants on the way from point i to j. The pheromone released by ants was studied through the ant density system, the concentration of pheromone released in this model was a constant value Q.
(4)
Load capacity check of the visiting location: if the load constraints are met, continue to visit, otherwise re-select the next place to be visited. The sites visited by the ants are added to the taboo list until the ants have visited all the sites.
(5)
Evaluate the new population: In this study, the fitness function was used to evaluate the new population judging the quality of individuals in the group through the objective function. Since the objective function in this study is to minimize the delivery cost and transportation risk, the inverse of the objective function is used as the value of the fitness function.
(6)
Crossover and mutation: according to the crossover probability P c , some genes on the chromosomes corresponding to the two individuals are crossed to generate a new individual. Then, according to the mutation probability P m , the gene on the chromosome is mutated. Through crossover and mutation, new individuals can be generated to increase population diversity.

4.2. Validity Verification

Based on the Solomon arithmetic example set, 15 rows of the dataset from the R class test set were selected to validate the model as well as the algorithm effectiveness. The original data is improved by borrowing the data generation of pickups and deliveries from the literature [55]. In this case, the basic data in the example are kept unchanged, and the data of delivery quantity d i and pickup quantity p i are generated according to the following rules: let the demand of customer i in the original example be   G i , the coordinates of station i are ( x i , y i ), and let r i = min( x i / y i , y i / x i ), then d i   =   G i r i , p i =   G i (1 − r i ). The details of the coordinate system, coordinate position, delivery volume, pickup volume, time window, and service time corresponding to each station are shown in Table 5. The specific calculation of personnel risk refers to Equations (4) and (5). Referring to the method in [43], relevant information such as the accident rate between points, the radius of the affected area when the accident occurs, and the population density, are generated. The accident rate between points is randomly generated in the range [1, 2], the population density between points is randomly generated in the range [300, 600] (in people) (for specific values, see Table A2 and Table A3 in Appendix A). Referring to the analysis of parameters in [56], the settings of other parameters required are shown in Table 5.
In order to compare the results in a more stable and scientific manner, each algorithm is run ten times, and the average value of the weighted function value is output (see Table 6). The platform for running the algorithm is a laptop with an Intel(R) Core(TM) i7-8565U CPU.
As can be seen from Table 6, when the weights are combined as (0.8, 0.2), (0.5, 0.5), (0.4, 0.6), and (0.8, 0.2), the average value of the ten-run results of the improved ant colony algorithm (ACO-GA) is the smallest. When the weight combination is (0.6, 0.4), the weighting function value of ACO-GA is the second smallest. It can be seen that the ACO-GA proposed in this paper can obtain better objective function values than the benchmark algorithm in most cases.

5. Empirical Analysis

5.1. Case Introduction and Preprocessing

Company A is a lithium battery sales and operation and R&D company, and its location is in Beijing. Company A’s stores are located in several districts in Beijing; nine stores plus one warehouse of Company A were selected for this study. There are three vehicles in the warehouse, which are used to provide the delivery service for each store. Figure 4 shows the locations of Company A’s warehouses and stores. Location 0 is the warehouse of Company A, and locations 1–9 are the stores. The specific latitude and longitude information is given in Table A4.
Since the distance data between each store are required for the calculation of the distribution cost and transportation risk, the latitude and longitude between two points needs to be transformed into the actual distance. According to Equation (17) [57], the transformed distance matrix is shown in Table 7, where it is assumed that B2 denotes the latitude of Store 1, C2 denotes the longitude of Store 1, D2 denotes the latitude of Store 2, E4 denotes the longitude of Store 2, F2 denotes the actual distance between Store 1 and Store 2, and R = 6371.373 (km) denotes the radius of the Earth.
F 2 = R · arcos [ cos ( B 2 ) · cos D 2 · cos ( C 2 E 4 ) + sin ( B 2 ) · sin ( D 2 ) ]
Company A’s delivery vehicle has a maximum loading capacity of 50 pieces, and the vehicle travels at a constant speed of 60 km/h. The warehouse provides services 24 h a day, and the pickup requirements, service hours and time windows of each store are shown in Table 8.

5.2. Parameter Combination Analysis

Determining the appropriate parameter values not only reduces the program running time, but also finds the optimal value of the objective function. Since the algorithm in this study is a hybrid algorithm, it is necessary to perform sensitivity analysis on the main parameters of the ant colony algorithm and the genetic algorithm to determine the optimal parameter combination. Parameters such as α, β, and ρ in the ant colony algorithm have great impact on the performance of the algorithm. Regarding the parameter settings of the ant colony algorithm, it is generally believed that the larger the values of α and β, the larger the computational effort and the longer the program run time. In the case that satisfactory solutions can be obtained, it is recommended that α and β be relatively small values. Generally, values of α between 1 and 2, β between 1 and 5, and ρ between 0 and 1 will achieve a solution with better results.
The initial values of the parameters were set as follows: population size m = 10, information heuristic factor α = 1, expectation heuristic factor β = 5, information volatility factor ρ = 0.75, total pheromone release Q = 10 (constant, the total amount of pheromone released by ants in one cycle), P c = 0.5, and P m = 0.1. Under the condition that other parameter values remain unchanged, only one parameter value is changed for analysis. If some optimal parameters have been determined in the previous step, we proceed to the sensitivity analysis of the next parameter based on the existing optimal combination. In this section, we set the weights as: ω 1 = ω 2 = 0.5 ; indicating that the company places equal emphasis on delivery costs and transportation risks. After ten repeated tests, objective function values and iteration times of different combinations are compared, and a more appropriate parameter combination is finally selected.
(1)
Information heuristic factor α and the expectation heuristic factor β
From Table 9 and Table 10, it can be seen that while keeping other parameters constant, the larger the values of α and β, the faster the algorithm converges. The results are that the optimal values of α and β are 2 and 5, respectively, at which time the weighted objective function value and the number of iterations to achieve convergence are minimal.
(2)
Information volatility factor ρ
As can be seen from Table 11, when ρ is relatively small, the pheromone volatilizes slowly, and the algorithm has strong global search ability, but requires a larger number of iterations to achieve convergence. When ρ is relatively large, the pheromone volatilizes quickly, the convergence speed is fast, and the global search ability of the algorithm is weak, making it is easy to fall into the local optimum. After comprehensive consideration, the value of ρ was chosen to be 0.8 in this study. At this time, the value of the weighted objective function value is minimal and convergence can be achieved in 36 iterations.
(3)
Crossover operator P c and variational operator P m
It can be seen from Table 12 that while keeping the value of the variational operator P m   constant, P c is taken as 0.2, the weighted objective function value and the number of iterations to reach convergence are minimal.
It can be seen from Table 13 that while keeping the value of the crossover operator P c constant, and P m is taken as 0.6, the weighted objective function value and the number of iterations to reach convergence are minimal. Although, the transportation risk and running time of the model are not minimal when P c = 0.6. But, based on the total cost (weighted value of the objective function), we determine the optimal P c = 0.6.
Therefore, the algorithm parameters in this study will be taken as follows: m = 10, α = 2, β = 5, ρ = 0.8, Q = 10, P c = 0.2,   P m = 0.6.

5.3. Path Preference Analysis

In order to explore how to choose the best route solution under different preferences, different weights are set for distribution cost and transportation risk for total cost accounting, where the size of the weights represent the importance of the decision makers, and a larger weight indicates that the decision makers want the goal to be smaller. Since the cost and risk have different orders of magnitude and units, they are standardized to ensure the uniformity of the scale in the weighting process. When setting weights for distribution cost and transportation risk, the principle that the sum of the two weights is equal to 1 is observed. By setting different weights for distribution cost and transportation risk, indicating different transportation preferences of enterprises, the optimal path is shown in Table 10, and the route planning schematic is shown in Figure 5.
According to Table 14, companies will have five different options when setting different weights on distribution costs and transportation risks. When an enterprise is facing the decision of both distribution cost and transportation risk, the more the enterprise cares, the greater the corresponding weight value, and the resulting value of this part will be small. The smaller the weight value, the lower the degree of concern of the enterprise, and the resulting value of this part will be larger.

6. Conclusions

In view of the practical problems of unsafe home charging, insufficient number of charging piles, high purchase cost of lithium batteries, and difficulty in charging distribution vehicles, this paper proposes a battery replacement mode of “centralized charging + unified distribution” for small electric vehicles based on the lithium battery leasing and distribution business of company A.
The proposed dual-objective model and improved ant colony algorithm (ACO-GA) can provide an optimal path that takes into account the distribution cost and transportation risk when solving the transportation path problem of hazardous goods such as lithium batteries. This can also provide reference for other enterprises that intend to develop power exchange businesses. When verifying the effectiveness of the improved algorithm, different weight combinations are used to verify the effectiveness of ACO-GA on the R-type dataset in the Solomon test set, which make the results more stable. After determining the optimal parameter combination through sensitivity analysis, we set different weight coefficients when calculating the weighted objective function value, which meet the distribution needs of enterprises with different risk preferences.
The study also has several limitations, which we plan to improve in our future work. Firstly, in terms of algorithm improvement, it can be combined with more intelligent algorithms suitable for VRPSPDSTW to carry out a variety of algorithm variants. Secondly, on the scale of the calculation example, a larger-scale data set can be further selected for experiments. Finally, in the comparison of algorithms, comparisons with other deep learning and reinforcement learning-based algorithms can be added in the future.

Author Contributions

Investigation, H.M.; Software, H.M.; Supervision, L.C. and F.L.; Writing—original draft, L.C. and Y.W.; Writing—review and editing, L.C., Y.W. and F.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Social Science Fund of China (Grant No. 20BJY185), the Beijing Social Science Foundation (Grant No. 19ZDA12, 18GLB022,18GLA009, and 21XCB005), and Beijing Collaborative Innovation Center for Intelligent Logistics System.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data presented in this study are available in the tables and Appendix A of the paper.

Conflicts of Interest

The authors declare no conflict of interest.

Appendix A

Table A1. Table of notations.
Table A1. Table of notations.
Sets and Parameters
O Centralized charging station
N = { i | i = 1 , 2 , , N ¯ } The number of lithium battery exchange stations
K = { k | k = 1 , 2 , , K ¯ } The number of delivery vehicles at centralized charging stations
d i j The   distance   from   station   i     to     j for lithium battery exchange
t i j Delivery   time   for   lithium   battery   exchange   station   i     to     j
s i k Service   hours   of   delivery   vehicles   k     at   battery   exchange   station   i
v i Lithium   battery   exchange   station   i delivery volume
p i Pickup   volume   at   lithium   battery   exchange   station   i
[ E T i , L T i ] The   best   service   time   range   required   by   the   lithium   battery   exchange   station     i
C 1 Delivery vehicle unit distance transportation cost
C 2 Unit penalty cost for early arrival
C 3 Unit penalty cost for being late
L Maximum load capacity of the vehicles.
R i j Personnel   risk   from   lithium   battery   exchange   station   i     to     j
p i j Accident   probability   from   lithium   battery   exchange   station   i   to   j
P i j The   number   of   people   affected   from   lithium   battery   exchange   station   i   to   j
λ The radius of influence of the accident
ρ i j Population   density   of   lithium   battery   exchange   station   i   to   j
ω 1 Weighting factor for delivery cost
ω 2 Weighting factor for transportation risk
T i k The   time   the   delivery   vehicle   k     arrives   at   the   lithium   battery   exchange   station     i
f w a g o n i k The   load   of   the   delivery   vehicle   k   when   it   arrives   at   the   lithium   battery   exchange   station     i
x i j k = { 0 , 1 } delivery   vehicle   k   heading   to   j   from   the   lithium   battery   exchange   station     i
y i k = { 0 , 1 } lithium   battery   exchange   station   i   is   serviced   by   the   vehicle   k
Table A2. Annual transport accident rate of road section.
Table A2. Annual transport accident rate of road section.
0123456789101112131415
001.131.761.441.511.451.841.221.641.061.481.901.431.271.011.39
11.1301.281.331.791.331.351.901.541.211.801.561.871.021.481.14
21.761.2801.501.361.811.261.371.081.591.311.291.011.671.171.91
31.441.331.5001.561.281.751.091.371.611.351.931.261.431.611.69
41.511.791.361.5601.251.311.451.811.441.561.271.021.421.631.43
51.451.331.811.281.2501.301.811.471.391.931.461.381.401.041.49
61.841.351.261.751.311.3001.211.081.321.691.641.501.831.051.73
71.221.901.371.091.451.811.2101.871.211.101.341.681.351.451.88
81.641.541.081.371.811.471.081.8701.391.051.491.331.401.351.08
91.061.211.591.611.441.391.321.211.3901.331.471.451.051.461.39
101.481.801.311.351.561.931.691.101.051.3301.011.591.391.041.30
111.901.561.291.931.271.461.641.341.491.471.0101.381.291.901.07
121.431.871.011.261.021.381.501.681.331.451.591.3801.281.421.47
131.271.021.671.431.421.401.831.351.401.051.391.291.2801.171.03
141.011.481.171.611.631.041.051.451.351.461.041.901.421.1701.29
151.391.141.911.691.431.491.731.881.081.391.301.071.471.031.290
Table A3. Average density of people affected by transport sections (people/km).
Table A3. Average density of people affected by transport sections (people/km).
0123456789101112131415
00572338574490329383464588590347592588446540342
15720538588497310555581504528523418497351512309
23385380547509395586310432414530539356447434494
35745885470497348335450588402476367526376452510
44904975094970377553376545373579405359375485442
53293103953483770527526414470322316459534581339
63835555863355535270459349481379496507525435325
74645813104503765264590332589300533546561325420
85885044325885454143493320474465343556487405454
95905284144023734704815894740447589556487405454
103475235304765793223793004654470406547304312350
115924185393674053164965333435894060534324579533
125884973565263594595075465565565475340582563465
134463514473763755345255614874873043245820429355
145405124344524855814353254054053125795634290452
153423094945104423393254204544543505334653554520
Table A4. Coordinates of distribution center and store locations.
Table A4. Coordinates of distribution center and store locations.
NumberStore NameLongitudeLatitude
0Company A
warehouse
116.65087340.154191
1Store 1116.39426540.200878
2Store 2116.44138739.810124
3Store 3116.56334539.769598
4Store 4116.65320140.334977
5Store 5116.86453839.805573
6Store 6115.72125639.577671
7Store 7116.24419339.923892
8Store 8116.56929739.878586
9Store 9115.99027639.71676

References

  1. China Bicycle Association. Available online: http://www.china-bicycle.com/ (accessed on 3 February 2022).
  2. Huajing Intelligence Network. Available online: https://www.huaon.com/channel/trend/797511.html (accessed on 14 April 2022).
  3. GB 17761-2018; E-Bike Safety Technical Specifications. Standards Press of China: Beijing, China, 2018.
  4. Shuai, C.Y.; Yang, F.; OuYang, X.; Xu, G. Data-driven demand forecasting for electric bicycle power exchange. Transp. Syst. Eng. Inf. 2021, 21, 173–179. [Google Scholar]
  5. Ministry of Transport of the People’s Republic of China. Report on Sustainable Transport in China. Available online: https://www.mot.gov.cn/zhuanti/transport/chengguowj/202110/P020211027595442474938.pdf (accessed on 3 February 2022).
  6. Fire and Rescue Department Ministry Emergency Management. Available online: https://www.119.gov.cn/article/46TiYamnnrs (accessed on 3 February 2022).
  7. Dai, J.C. Comparison of electric vehicle whole vehicle charging mode and centralized charging + power exchange mode. Power Demand Side Manag. 2014, 16, 57–60. [Google Scholar]
  8. Dantzig, G.B.; Ramser, J.H. The truck dispatching problem. Manag. Sci. 1959, 6, 80–91. [Google Scholar] [CrossRef]
  9. Pérez-Rodríguez, R.; Hernández-Aguirre, A. A hybrid estimation of distribution algorithm for the vehicle routing problem with time windows. Comput. Ind. Eng. 2019, 130, 75–96. [Google Scholar] [CrossRef]
  10. Konstantakopoulos, G.D.; Gayialis, S.P.; Kechagias, E.P. Vehicle routing problem and related algorithms for logistics distribution: A literature review and classification. Oper. Res. 2022, 22, 2033–2062. [Google Scholar] [CrossRef]
  11. Solomon, M.M. Algorithms for the vehicle routing and scheduling problems with time window constraints. Oper. Res. 1987, 35, 254–265. [Google Scholar] [CrossRef] [Green Version]
  12. Koskosidis, Y.A.; Powell, W.B.; Solomon, M.M. An optimization-based heuristic for vehicle routing and scheduling with soft time window constraints. Transp. Sci. 1992, 26, 69–85. [Google Scholar] [CrossRef]
  13. Chiang, W.C.; Russell, R.A. A metaheuristic for the vehicle-routing problem with soft time windows. J. Oper. Res. Societ. 2004, 55, 1298–1310. [Google Scholar] [CrossRef]
  14. Bae, H.; Moon, I. Multi-depot vehicle routing problem with time windows considering delivery and installation vehicles. Appl. Math. Model. 2016, 40, 6536–6549. [Google Scholar] [CrossRef]
  15. Ye, Y.; Zhang, H.Z. Wolfpack algorithm for solving vehicle path problems with time windows. Highw. Traffic Technol. 2017, 34, 100–107. [Google Scholar]
  16. Sun, J.C.; Li, D. Open vehicle path problem based on improved firefly algorithm. Pract. Underst. Math. 2018, 48, 182–190. [Google Scholar]
  17. Wu, L.B.; He, Z.J.; Chen, Y.J.; Wu, D.; Cui, J.Q. Brainstorming-Based Ant Colony Optimization for Vehicle Routing With Soft Time Windows. IEEE Access 2019, 7, 19643–19652. [Google Scholar]
  18. He, M.L.; Wei, Z.X.; Wu, X.H.; Peng, Y.T. An improved ant colony based algorithm for solving vehicle path problems with soft time windows. Comput. Integr. Manuf. Syst. 2021, 1–14. Available online: https://kns.cnki.net/kcms/detail/detail.aspx?dbcode=CAPJ&dbname=CAPJLAST&filename=JSJJ20210723002&uniplatform=NZKPT&v=EfDktvpY4SAJDRUlkmtxMlt5c853-MY6QWqILRG8ZdYmDW_TBi_7fAm7iPg60Br8 (accessed on 3 February 2022).
  19. Yu, Y.; Wang, S.; Wang, J.; Huang, M. A branch-and-price algorithm for the heterogeneous fleet green vehicle routing problem with time windows. Transp. Res. Part B Methodol. 2019, 122, 511–527. [Google Scholar] [CrossRef]
  20. Wang, J.; Weng, T.; Zhang, Q. A two-stage multiobjective evolutionary algorithm for multiobjective multidepot vehicle routing problem with time windows. IEEE Trans. Cybern. 2018, 49, 2467–2478. [Google Scholar] [CrossRef] [PubMed]
  21. Song, M.; Li, J.; Han, Y.; Han, Y.-Y.; Liu, L.; Sun, Q. Metaheuristics for solving the vehicle routing problem with the time windows and energy consumption in cold chain logistics. Appl. Soft Comput. 2020, 95, 106561. [Google Scholar] [CrossRef]
  22. Goetschalckx, M.; Jacobs-blecha, C. The vehicle routing problem with back hauls. Eur. J. Oper. Res. 1987, 42, 39–51. [Google Scholar] [CrossRef]
  23. Min, H. The multiple vehicle routing problem with simultaneous delivery and pick up points. Transp. Res. Part. A Gen. 1989, 23, 377–386. [Google Scholar] [CrossRef]
  24. Reil, S.; Bortfeldt, A.; Mönch, L. Heuristics for vehicle routing problems with backhauls, time windows, and 3D loading constraints. Eur. J. Oper. Res. 2018, 266, 877–894. [Google Scholar] [CrossRef]
  25. Koç, Ç.; Laporte, G.; Tükenmez, İ. A review of vehicle routing with simultaneous pickup and delivery. Comput. Oper. Res. 2020, 122, 104987. [Google Scholar] [CrossRef]
  26. Kalayci, C.B.; Kaya, C. An ant colony system empowered variable neighborhood search algorithm for the vehicle routing problem with simultaneous pickup and delivery. Expert Syst. Appl. 2016, 66, 163–175. [Google Scholar] [CrossRef]
  27. Islam, M.A.; Gajpal, Y.; ElMekkawy, T.Y. Hybrid particle swarm optimization algorithm for solving the clustered vehicle routing problem. Appl. Soft Comput. 2021, 110, 107655. [Google Scholar] [CrossRef]
  28. Goksal, F.P.; Karaoglan, I.; Altiparmak, F. A hybrid discrete particle swarm optimization for vehicle routing problem with simultaneous pickup and delivery. Comput. Ind. Eng. 2013, 65, 39–53. [Google Scholar] [CrossRef]
  29. Avci, M.; Topaloglu, S. An adaptive local search algorithm for vehicle routing problem with simultaneous and mixed pickups and deliveries. Comput. Ind. Eng. 2015, 83, 15–29. [Google Scholar] [CrossRef]
  30. Avci, M.; Topaloglu, S. A hybrid metaheuristic algorithm for heterogeneous vehicle routing problem with simultaneous pickup and delivery. Expert Syst. Appl. 2016, 53, 160–171. [Google Scholar] [CrossRef]
  31. Ni, L.; Liu, K.P.; Tu, Z.G. Urban courier co-delivery path optimization considering simultaneous pickup and delivery. J. Chongqing Univ. 2017, 40, 30–39. [Google Scholar]
  32. Chen, X.Q.; Hu, D.W.; Yang, Q.Q.; Hu, H.; Gao, Y. Improved ant colony algorithm for multi-objective simultaneous delivery vehicle path problem. Control Theory Appl. 2018, 35, 1347–1356. [Google Scholar]
  33. Ma, Y.F.; Yan, F.; Kang, K.; Li, Z.M. Uncertain simultaneous pickup delivery vehicle path problem and particle swarm algorithm research. Oper. Res. Manag. 2018, 27, 73–83. [Google Scholar]
  34. Ren, T.; Luo, T.Y.; Gu, Z.H.; Hu, Z.Q.; Jia, B.B.; Xing, L.N. Urban logistics common distribution path optimization considering simultaneous pickup and delivery. Comput. Integr. Manuf. Syst. 2022, 1–18. Available online: https://kns.cnki.net/kcms/detail/detail.aspx?dbcode=CAPJ&dbname=CAPJLAST&filename=JSJJ20220315000&uniplatform=NZKPT&v=RxFrS6Ig2zID5j-2re_ca_HUwv4q5cjDEfQHzf8ervf8NyXtmHojlHxPnWaoENOr (accessed on 3 February 2022).
  35. Wang, L.Y.; Song, Y.B. Multiple Charging Station Location-Routing Problem with Time Window of Electric Vehicle. J. Eng. Sci. Technol. Rev. 2015, 8, 190–201. [Google Scholar]
  36. Wang, C.; Mu, D.; Zhao, F.; Sutherland, J.W. A parallel simulated annealing method for the vehicle routing problem with simultaneous pickup–delivery and time windows. Comput. Ind. Eng. 2015, 83, 111–122. [Google Scholar] [CrossRef]
  37. Zhang, Q.H.; Wu, S.P. Modeling of simultaneous delivery vehicle path problem with time window and modal solution algorithm. Comput. Appl. 2020, 40, 1097–1103. [Google Scholar]
  38. Zhang, X.F.; Hu, R.; Qian, B. Superheuristic distribution estimation algorithm for solving simultaneous delivery vehicle path problems with soft time windows. Control. Theory Appl. 2021, 38, 1427–1441. [Google Scholar]
  39. Hornstra, R.P.; Silva, A.; Roodbergen, K.J.; Coelho, L.C. The vehicle routing problem with simultaneous pickup and delivery and handling costs. Comput. Oper. Res. 2020, 115, 104858. [Google Scholar] [CrossRef]
  40. Yan, J.; Chang, L.; Wang, L.L.; Zhao, T. Algorithm for solving the simultaneous pickup and delivery vehicle path problem with time windows. Ind. Eng. 2021, 24, 72–76. [Google Scholar]
  41. Ahkamiraad, A.; Wang, Y. Capacitated and multiple cross-docked vehicle routing problem with pickup, delivery, and time windows. Comput. Ind. Eng. 2018, 119, 76–84. [Google Scholar] [CrossRef]
  42. Lagos, C.; Guerrero, G.; Cabrera, E.; Moltedo, A.; Johnson, F.; Paredes, F. An improved particle swarm optimization algorithm for the VRP with simultaneous pickup and delivery and time windows. IEEE Lat. Am. Trans. 2018, 16, 1732–1740. [Google Scholar] [CrossRef]
  43. Zhang, S.Z.; Chen, T.T.; Sun, R.T.; Bai, X.; Gao, C. Path optimization of dangerous goods transportation vehicles considering cargo capacity. J. Wuhan Univ. Technol. (Inf. Manag. Eng. Ed.) 2020, 42, 290–297. [Google Scholar]
  44. Erkut, E.; Ingolfsson, A. Transport risk models for hazardous materials: Revisited. Oper. Res. Lett. 2005, 3, 81–89. [Google Scholar] [CrossRef]
  45. Xin, C.L.; Meng, J.; Zhang, J.W.; Feng, Q.R. A review of hazardous materials transportation route optimization problems. Chin. J. Saf. Sci. 2018, 28, 102–107. [Google Scholar]
  46. Kara, B.Y.; Verter, V. Designing a road network for hazardous materials transportation. Transp. Sci. 2004, 38, 188–196. [Google Scholar] [CrossRef] [Green Version]
  47. Zografos, K.G.; Androutsopoulos, K.N. A heuristic algorithm for solving hazardous materials distribution problems. Eur. J. Oper. Res. 2004, 152, 507–519. [Google Scholar] [CrossRef]
  48. Androutsopoulos, K.N.; Zografos, K.G. Solving the bicriterion routing and scheduling problem for hazardous materials distribution. Transp. Res. Part. C Emerg. Technol. 2010, 18, 713–726. [Google Scholar] [CrossRef] [Green Version]
  49. Chai, X.; He, R.C.; Ma, C.C. Multi-objective optimization of hazardous materials transportation vehicle path problem. Chin. J. Saf. Sci. 2015, 25, 84–90. [Google Scholar]
  50. Zhang, M.; Wang, N.M. Research on path optimization of dangerous goods transportation vehicles for major accident avoidance. Oper. Res. Manag. 2018, 27, 1–9. [Google Scholar]
  51. Li, L.; Deng, Y.T.; Mou, L.L. Research on optimization of dangerous goods transportation path based on improved NSGA-II algorithm. Saf. Environ. Eng. 2021, 28, 111–117. [Google Scholar]
  52. Erkut, E.; Verter, V. Modeling of transport risk for hazardous materials. Oper. Res. 1998, 46, 625–642. [Google Scholar] [CrossRef] [Green Version]
  53. Teng, Y.; Sun, L.J.; Zhou, Y.X. A multi-model vehicle path optimization method considering the risk of hazardous materials transportation. Syst. Eng. 2020, 38, 93–102. [Google Scholar]
  54. Kazantzi, V.; Kazantzis, N.; Gerogiannis, V.C. Risk informed optimization of a hazardous material multi-periodic transportation model. J. Loss Prev. Process. Ind. 2011, 24, 767–773. [Google Scholar] [CrossRef]
  55. Zhou, R.; Shen, W.L.; Liu, M.Z.; Zhao, H. Hybrid discrete particle swarm optimization algorithm for integrated vehicle path problems with time windows for loading and unloading. China Mech. Eng. 2016, 27, 494–502. [Google Scholar]
  56. Ye, Z.W.; Zheng, Z.B. Configuration of Parameters α, β and ρ in Ant Algorithm. Geomat. Inf. Sci. Wuhan Univ. 2004, 29, 597–601. [Google Scholar]
  57. Li, Z.X.; Li, J.X. Quickly Calculate the Distance between Two Points and Measurement Error Based on Latitude and Longitude. Geomat. Spat. Inf. Technol. 2013, 36, 235–237. [Google Scholar]
Figure 1. Schematic diagram of the delivery process.
Figure 1. Schematic diagram of the delivery process.
Sustainability 14 09883 g001
Figure 2. Operation flow of “centralized charging + unified distribution” power exchange model.
Figure 2. Operation flow of “centralized charging + unified distribution” power exchange model.
Sustainability 14 09883 g002
Figure 3. Flow chart of ant colony genetic hybrid algorithm.
Figure 3. Flow chart of ant colony genetic hybrid algorithm.
Sustainability 14 09883 g003
Figure 4. Store location map of Company A.
Figure 4. Store location map of Company A.
Sustainability 14 09883 g004
Figure 5. Distribution route diagram.
Figure 5. Distribution route diagram.
Sustainability 14 09883 g005
Table 1. Existing studies on VRPTW.
Table 1. Existing studies on VRPTW.
LiteratureProblemAlgorithmObjective
Bae et al. [14]Multi-depot vehicle routing problem with time windows (MDVRPTW)Heuristic algorithm and a hybrid genetic algorithmMinimize the total relevant costs
Ye et al. [15]VRPTWWolf pack algorithmMinimum total transportation cost
Sun et al. [16]Multi-depot open vehicle routing problem with soft time windows (MDOVRPTW)Improved glowworm swarm optimization (GSO)Minimum total cost
Wu et al. [17]VRPSTWOptimized ACOMinimum total transportation cost
He et al. [18]VRPSTWOptimized ACOMinimize delivery costs
Yu et al. [19]Heterogeneous fleet green vehicle routing problem with time windows (HFGVRPTW)Multi-vehicle approximate dynamic programming (MVADP)The branches and computational time
Wang et al. [20]Multi-depot VRPTW (MDVRPTW)Two-stage multi-objective evolutionary algorithmMinimize the number of vehicles, total travel distance, makespan, total waiting time, and total delay time
Song et al. [21]VRPTW in cold chain logisticsImproved artificial fish swarm (IAFS)Minimize the fixed cost and the energy consumptions
Table 2. Existing studies on VRPSPD.
Table 2. Existing studies on VRPSPD.
LiteratureApplication ScenariosAlgorithmObjective
Goksal et al. [28]VRPSPDOptimized PSOExtend the algorithm for solving VRPSPD
Avci et al. [29,30]VRPSD with multiple vehicle modelsHybrid Local Search AlgorithmExtend the algorithm for solving VRPSPD
Ni et al. [31]Multiple courier companies jointly deliverOptimized ACOMinimize total delivery costs
Chen et al. [32]Dual-objective VRPSPD model considering both vehicle capacity and distance constraintsOptimized ACOMinimize the maximum length difference routes and minimize transportation cost
Ma et al. [33]Uncertain simultaneous pickup and delivery vehicle routing problemOptimized PSOLowest operating costs and highest customer satisfaction
Ren et al. [34]The problem of picking up and delivering orders in the same cityOptimized GAMinimize total cost
Table 3. Existing studies on VRPSPDTW.
Table 3. Existing studies on VRPSPDTW.
LiteratureAlgorithmObjective
Wang et al. [36]Simulated annealing (SA) algorithmMinimize the routing cost
Zhang, Q.H et al. [37]Memetic algorithmMinimum the number of vehicles and shortest vehicle travel path
Zhang, S.Z et al. [38]Optimized TSAMinimize the total cost including time penalty cost
Hornstra et al. [39]Adaptive large neighborhood search (ALNS)Minimize the total processing cost
Yan et al. [40]K-means-ACOMinimize the total travel cost during return and delivery service
Ahkamiraad et al. [41]A hybrid of the genetic
algorithm and particle swarm optimization (HGP)
Minimize the transportation and fixed costs
Lagos et al. [42]PSOMinimize the
total distance of the paths and serving customers’ demands
Table 4. Algorithm parameter settings.
Table 4. Algorithm parameter settings.
ParametersParameter MeaningParameter Value
m Population size10
Information heuristic factor1
β Expectation heuristic factor5
ρ Information volatility factor0.75
Q Total pheromone release10
P c Crossover probability0.5
P m Mutation probability0.1
C 1 Delivery vehicle unit distance transportation cost3 CNY/km
C 2 Unit penalty cost for early arrival20 CNY/h
C 3 Unit penalty cost for being late 30 CNY/h
λ The radius of influence of the accident0.1 km
Table 5. Detailed information table for each location.
Table 5. Detailed information table for each location.
Serial NumberCustomer
Coordinates
Delivery
Quantity (Pieces)
Pickup
Quantity (Pieces)
Time
Window (min)
Service Time (min)
0(35, 35)
1(41, 49)82[161, 171]10
2(35, 17)34[50, 60]10
3(55, 45)112[116, 126]10
4(55, 20)712[149, 159]10
5(15, 30)1313[34, 44]10
6(25, 30)30[99, 109]10
7(20, 50)23[81, 91]10
8(10, 43)27[95, 105]10
9(55, 60)151[97, 107]10
10(30, 60)88[124, 134]10
11(20, 65)48[67, 77]10
12(50, 35)136[63, 73]10
13(30, 25)194[159, 169]10
14(15, 10)137[32, 42]10
15(30, 5)17[61, 71]10
Table 6. Algorithm result comparison.
Table 6. Algorithm result comparison.
Algorithm ω 1 ω 2 Number of Iterations
0.8–0.20.6–0.40.5–0.50.4–0.60.2–0.8
ACO-GA1008.23889.22789.14708.29556.4235
ACO1015.13865.88801.90725.51560.1945
GA1471.991232.841200.591007.75767.5868
SAA1547.081377.721202.06983.86728.1289
Table 7. Distance matrix between stores of Company A (kilometers).
Table 7. Distance matrix between stores of Company A (kilometers).
Stores No.0123456789
0022.442.243.420.142.8102.343.031.472.3
1 043.650.126.559.489.233.338.863.8
2 011.461.136.166.721.113.339.9
3 063.326.175.332.212.149.3
4 061.511.557.451.288.8
5 0100.954.626.575.3
6 059.180.027.8
7 028.231.6
8 052.6
9 0
Table 8. Detailed information table of each store.
Table 8. Detailed information table of each store.
SerialStore NameDeliveryPickupTimeService Time
Number Quantity
(Pieces)
Quantity
(Pieces)
Window
(min)
(min)
0Company A
warehouse
1Store 1102[7:30–7:45]8
2Store 2710[7:45–8:00]10
3Store 363[8:10–8:35]5
4Store 456[7:05–7:20]5
5Store 5414[8:15–8:25]10
6Store 6147[8:30–9:00]10
7Store 7116[7:55–8:10]9
8Store 863[7:50–8:15]6
9Store 9142[8:00–8:30]7
Table 9. Results of the effect of α value on the ant colony genetic hybrid algorithm.
Table 9. Results of the effect of α value on the ant colony genetic hybrid algorithm.
αβAverage Cost
(CNY)
Average RiskWeighted Objective Function Value (CNY)Number of Iterations to Reach ConvergenceTime (s)
0532.305.7819.048911.70
0.5537.845.7621.803212.91
1534.915.9920.457912.57
2521.375.4113.88711.73
Table 10. Results of the effect of β value on the ant colony genetic hybrid algorithm.
Table 10. Results of the effect of β value on the ant colony genetic hybrid algorithm.
αβAverage Cost
(CNY)
Average
Risk
Weighted Objective Function ValueNumber of Iterations to Reach ConvergenceTime (s)
2041.658.6824.902013.83
2233.086.2519.663817.60
2434.335.6620.00616.20
2532.895.8719.38511.86
Table 11. Results of information volatility factor ρ on hybrid ant colony genetic algorithm.
Table 11. Results of information volatility factor ρ on hybrid ant colony genetic algorithm.
ρAverage Cost (CNY)Average RiskWeighted Objective
Function Value
Number of Iterations to Reach ConvergenceTime (s)
0.236.115.6920.905213.78
0.433.885.6719.784212.03
0.638.175.9422.065712.10
0.832.365.9019.133613.29
Table 12. Results of P c on the ant colony genetic algorithm.
Table 12. Results of P c on the ant colony genetic algorithm.
P c P m Average Cost
(CNY)
Average
Risk
Weighted Objective Function ValueNumber of Iterations to Reach ConvergenceTime (s)
0.2128.115.8516.982316.11
0.4129.535.9817.755613.06
0.6128.295.7717.037114.64
0.8129.645.9217.789512.35
Table 13. Results of P m on the ant colony genetic hybrid algorithm.
Table 13. Results of P m on the ant colony genetic hybrid algorithm.
P c P m Average Cost
(CNY)
Average
Risk
Weighted Objective Function ValueNumber of Iterations to Reach ConvergenceTime (s)
10.233.755.8919.828813.23
10.430.695.4618.083113.76
10.628.855.7917.321414.32
10.828.985.8517.421712.75
Table 14. Path selection preference results with different weights.
Table 14. Path selection preference results with different weights.
StrategyWeightsOptimal Path
ω 1 ω 2 Vehicle 1Vehicle 2Vehicle 3Normalized
Objective Function Value
Distribution CostShipping
Risks
Programs
Preference
A0.20.80→4→1
→2→3→8→5→0
0→7→9
→6→0
0.7660.145.40Very concerned
about risk
B0.40.60→1→4
→8→2→7→0
0→3→5
→6→0
0→5→3
→2→0
0.4729.348.67More concerned
about risk
C0.50.50→4→1
→7→0
0→8→3
→2→9→0
0→5→6
→0
0.8355.178.08Intentional risk
and cost
D0.60.40→4→1
→7→2→8→0
0→5→3
→9→0
0→6→00.4229.887.85More concerned about cost
E0.80.20→4→1
→7→2→8→0
0→3→5
→9→0
0→6→00.8028.408.51Very concerned about cost
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Chen, L.; Ma, H.; Wang, Y.; Li, F. Vehicle Routing Problem for the Simultaneous Pickup and Delivery of Lithium Batteries of Small Power Vehicles under Charging and Swapping Mode. Sustainability 2022, 14, 9883. https://doi.org/10.3390/su14169883

AMA Style

Chen L, Ma H, Wang Y, Li F. Vehicle Routing Problem for the Simultaneous Pickup and Delivery of Lithium Batteries of Small Power Vehicles under Charging and Swapping Mode. Sustainability. 2022; 14(16):9883. https://doi.org/10.3390/su14169883

Chicago/Turabian Style

Chen, Lei, Haiyan Ma, Yi Wang, and Feng Li. 2022. "Vehicle Routing Problem for the Simultaneous Pickup and Delivery of Lithium Batteries of Small Power Vehicles under Charging and Swapping Mode" Sustainability 14, no. 16: 9883. https://doi.org/10.3390/su14169883

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop