Logistics Network Distribution Optimization Based on Vehicle Sharing

Abstract

The development of the sharing economy has provided new ideas for a vehicle-sharing urban logistics network cooperative distribution strategy. In view of the lack of dispatching capacity or transportation capacity of logistics enterprises with multiple distribution centers, this paper proposes a vehicle-sharing urban logistics network cooperative distribution strategy. Based on the comprehensive consideration of a multi-distribution center, multi-model, rental vehicle, load, speed, fuel consumption, and other factors, the calculation method of vehicle energy consumption is introduced, the network collaborative distribution model with vehicle sharing is established, and an adaptive genetic algorithm combined with a scanning algorithm is designed. Finally, the validity and reliability of the mathematical model and algorithm are validated and analyzed by an example. The research results show that vehicle sharing can improve the efficiency of the distribution network and effectively reduce costs.

1. Introduction

With the vigorous development of electronic logistics and urban express delivery, especially holiday promotions, the transportation capacity of the logistics and distribution industry has become increasingly demanding. How to adjust the transportation capacity of enterprises to adapt to the dramatic changes in logistics demand has become crucial. Therefore, the logistics system optimization strategy of networked collaborative distribution, timely sharing, and rental vehicles has become important research content to solve this problem.

The vehicle routing problem has been developed for more than 50 years. Many scholars have researched it in depth, and the factors considered are more and more comprehensive. Compared with the single depot problem, the multi-depot vehicle routing problem (MDVRP) becomes more complex, especially in the case of sharing and leasing of vehicles and vehicle type selection, which is more difficult to solve effectively. Wang et al. [1] proposed a new hybrid algorithm to solve multi-depot VRP with time windows considering two-dimensional packing constraints; Bae et al. [2] studied multi-depot VRP with time windows and applied it to the distribution of electronic products delivery and designed a mixed integer programming model and heuristic genetic algorithm to solve it; and Tang et al. [3] established a soft time window. The mathematical model of a multi-depot and multi-vehicle routing problem was established, and an improved ant colony optimization algorithm was proposed to solve the problem. Baniasadi et al. [4] proposed a transformation technique for the clustered generalized traveling salesman problem with applications to logistics. Adelzadeh et al. [5] studied the multi-depot vehicle routing problem on the basis of considering the fuzzy time window. Yang et al. [6] solved vehicle routing problems with fuzzy requirements, multi-center, and open triple constraints; and Jin et al. [7] proposed an improved differential evolution algorithm and applied it to a multi-distribution center logistics vehicle scheduling problem. The demand-separable vehicle routing model proposed by Pan et al. [8] can effectively improve the vehicle loading rate and reduce vehicle cost. Jiang et al. [9] used the composite neighborhood discrete firefly algorithm to deal with the problem of asymmetric multi-vehicle vehicle routing in agricultural machinery maintenance, and Ge et al. [10] used a cloud genetic algorithm to study shared vehicles to reduce vehicle distribution distance and improve distribution efficiency. Liu et al. [11] proposed a hybrid genetic algorithm to study vehicle sharing and leasing to meet the lack of capacity. Kuo et al. [12] constructed a model where the factors of loading cost were considered. Li et al. [13] considered the influence of speed, load, and distance on distribution cost in the vehicle routing problem for the first time. Luca et al. [14] balanced inventory and path selection in an integrated manner to reduce inventory costs and solve multi-warehouse inventory path problems. Yong et al. [15] discussed the multi-vehicle green vehicle routing problem and designed an improved ant colony algorithm to solve this problem. Tu et al. [16] proposed the two-tiered Thiessen polygon method, based on the meta-heuristic method of the graph, to solve this problem on a large scale. Alinaghian et al. [17] studied the multi-vehicle multi-compartment vehicle routing problem. Guo et al. [18] proposed three hybrid algorithms to solve the multi-vehicle routing problem in a single depot.

At the same time, research progress in ride-sharing has also been analyzed and discussed. Furuhata et al. [19] proposed a classification for understanding the key factors in existing ride-sharing systems. Stiglic et al. [20] discussed the benefits of using rendezvous points to improve the performance of ride-sharing systems. When travelers are willing to walk to rendezvous points from other locations, drivers tend to be able to carry more independent passengers without many additional stops. Lee and Savelsbergh [21] established an integer programming model for the outsourcing distribution problem and proposed a heuristic algorithm to solve it. Hipólito et al. [22] described a static carpooling problem aimed at assigning drivers a group of viable passengers. Agatz et al. [23] systematically summarized the optimization challenges and relevant operational research models in the development of ride-sharing technology. The study of Agatz et al. [24] investigated the feasibility of dynamic carpooling, in which drivers announce their travel schedule shortly before departure to match and share their travel schedule with independent passengers along the way, and proposed a multi-cycle solution method for real-time updates. The scholar described the problem as a set partition problems and proposed an accurate solution based on column generation.

Some scholars tried to use the existing traffic flow of the residual capacity for the carriage of goods; Ghilas et al. [25] and Masson et al. [26] discussed the possibility and potential benefits of using public transportation in freight transportation. Li et al. [27] considered the situation of using taxis to jointly transport packages with passengers and proposed a heuristic solution strategy to insert the package request into the existing taxi routes. Due to the flexibility of passengers to get in and out of the taxi, the taxi may stop several times to pick up the package. Archetti et al. [28] studied the case for Walmart’s vision of having in-store customers deliver goods ordered by online customers, creating an interesting new variant of the vehicle routing problem called the Accidental Driver Routing Problem (VRPOD). Suh et al. [29] discussed people and goods transportation, such as resource sharing, and studied the concepts and solutions to improve urban traffic; the results of the survey showed that traditionally irrelevant mode share transportation resources have great potential (for example, trams can be used for freight). Alp et al. [30] studied the concept of crowdsourced delivery, which aims to use excess capacity to assist distribution on journeys that are already occurring. Fatnassi et al. [31] studied the integration of passenger transportation and freight transportation in the context of an urban logistics automated transportation system.

Through comprehensive analysis of the above literature, we found that comprehensive consideration of vehicle type, load, speed, and capacity demand changes has become a realistic demand of logistics enterprises; at the same time, the problem of inadequate transport capacity caused by the holiday economy is worth attention. The structure of this article is as follows: Section 2 describes the research problem; in Section 3, we establish a vehicle routing model based on multiple constraints. In Section 4, an information matrix is proposed to solve the difficulty of chromosome coding, and an adaptive genetic algorithm is improved by using a scanning algorithm and a two-sided successive correction algorithm. In Section 5, a case analysis of the model and algorithm are carried out, and an adaptive genetic algorithm is tested from different aspects. In Section 6, the conclusions of this paper are presented.

2. Problem Description

The research object of this paper is a logistics enterprise, which has many distribution centers; a logistics network is optimized through efficient collaboration and resource sharing across multiple facilities and service cycles, and among them, multiple service periods can be determined according to different facilities and customers of time window types. The actual network optimization not only considers the centralized transportation between multiple logistics facilities but also considers the rational utilization of logistics resources between multiple service cycles. Each distribution center has enough goods and a variety of vehicles of different numbers. The coordinates and requirements of each customer are known and served only once. Distribution vehicles start from the distribution center only once in a single decision-making cycle and eventually return to the distribution center. The strategy of networked collaborative distribution among multi-distribution centers means all customers are not confined to a single distribution center. When the demand for transportation capacity of a distribution center increases sharply, they can share the vehicles in the nearby capacity-rich distribution center. When the capacity of all distribution centers is insufficient, they can rent the required external vehicles. It is expected that through the cooperation between multiple facilities established in the process of network optimization, the sharing of customer service and transportation resources will be realized, thereby improving transportation efficiency and optimization and rational allocation of resources. Multi-distribution centers operate as shown in Figure 1.

3. Model Establishment

3.1. Fuel Consumption Cost Calculations

The optimization goal of this paper is the lowest fuel consumption, so the calculation formula of fuel consumption should be given first. The symbols used in this article are defined in

Table 1. Model symbol description.

Symbol	Explanation
D	Set of distribution centers
C	Set of customers
K0	Set of customer vehicles owned by distribution center (total)
K	Set of vehicles that can provide distribution services to all customers
wijkn	The load of vehicle k with n type in the process of driving from node i to node j
tis	Delivery service time provided by customer i
Qkn	Maximum load of vehicle k with type n
vnk	Average speed of vehicle type n
wn	The unit fixed cost of a self-owned vehicle of type n, including car purchase, maintenance, driver’s salary, etc.
gkn	When vehicle k of type n runs as a rented vehicle, the value is 1; otherwise it is 0.
Gkn	When the vehicle K of type n is the vehicle owned by the distribution center, the value is 1; otherwise it is 0.
N	Set of vehicle types for distribution vehicles
S	Set of branch elimination constraints
T	Charging time for rental vehicles
Krd	Total number of vehicles leased in distribution center D
Rn	Rental per unit time for vehicles of type n, in which Rn > Wn
w0n	No-load capacity of vehicle type n
qi	Demand for customer i
cf	Unit fuel consumption cost
Tkn	The total time spent on a single delivery service for a vehicle k with type n
xijkn	When the vehicle k of type n drives from node i to j, the value is 1; otherwise it is 0.
yikn	Vehicle k with n type completes the assignment task of node i with a value of 1; otherwise it is 0.

.

In order to solve the problem of fuel consumption during vehicle operation, the paper refers to the relevant theory of vehicle energy consumption calculation in references [15,21,22]. The energy consumption (unit: L) and fuel consumption (unit: CNY) of arc $(i,j)$ can be expressed as follows:

$$p_{ij}\approx {\alpha }_{ij}(w_0+w_{ij})d_{ij}+{\beta }_k{v}_{ij}{}^2{d}_{ij}$$

(1)

$$f{c}_{ij}=c_f\ast p_{ij}=[{\alpha }_{ij}(w_0+w_{ij})+{\beta }_k{v}_{ij}^2]c_f{d}_{ij}$$

(2)

where ${\alpha }_{ij}$ is a constant related to vehicle driving condition; ${\beta }_k$ is a constant related to vehicle type $k$; $w_0$ (unit: kg), $w_{ij}$ (unit: kg), $v_{ij}$ (unit: V/m), and $d_{ij}$ (unit: m) are, respectively, the net weight of vehicles, vehicle load, vehicle speed, and distance in a specific section; and $c_f$ (unit: CNY/liter) is the fuel price. According to the relevant data in reference [22], ${\alpha }_{ij}$ ≈ 3.10 × 10⁻⁵ and ${\beta }_k$ ≈ 5.13 × 10⁻⁵ are calculated. This paper assumes that the speed is 100 km/h and the oil price $c_f$ is 7.9 CNY/liter.

3.2. Mathematical Model of Networked Cooperative Distribution

In this section, we describe the improvement problem of the traditional VRP, namely the networked distribution problem of vehicle sharing, considering the vehicle sharing factors and network optimization characteristics. The symbols used in this article are defined in Table 1. According to the research problems, this paper establishes a mathematical model of network collaborative distribution, in which the objective function is as follows:

$$\begin{array}{l}\min \begin{array}{c}Z={\displaystyle \sum _{i\in D\cup C}{\displaystyle \sum _{j\in D\cup C,j\ne i}{\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{x}_{ijkn}\times c_f\times f{c}_{ijkn}}}}}\end{array}\\ \begin{array}{cc}\begin{array}{cc}& \end{array}& \end{array}+\frac{T_{kn}}{T}\left({\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{R}_n\times g_{kn}}}+{\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{w}_n\times G_{kn}}}\right)\end{array}$$

(3)

$${\displaystyle \sum _{i\in D\cup C}{\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{x}_{ijkn}}}}=1,\forall j\in D\cup C$$

(4)

$${\displaystyle \sum _{j\in D\cup C}{\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{x}_{ijkn}}}}={\displaystyle \sum _{j\in D\cup C}{\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{y}_{jkn}}}},\forall i\in D\cup C$$

(5)

$${\displaystyle \sum _{j\in D\cup C}\begin{array}{c}{\displaystyle \sum _{i\in D\cup C,i\ne j}{q}_j{x}_{ijkn}}\end{array}}\le Q_{kn},\forall k\in K,n\in N$$

(6)

$${\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}\left(g_{kn}+G_{kn}\right)}}\le {\displaystyle \sum _{d\in D}{K}_{rd}}+K_0$$

(7)

$${\displaystyle \sum _{i\in D}{\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{x}_{ijkn}}}}={\displaystyle \sum _{i\in D}{\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{x}_{jikn}}}},\forall j\in D\cup C$$

(8)

$${\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}{q}_j{x}_{ijkn}}}\le w_{ijkn}\le {\displaystyle \sum _{k\in K}{\displaystyle \sum _{n\in N}(Q_{kn}-q_j)x_{ijkn}}},\forall \left(i,j\right)\in D\cup C$$

(9)

$$T_{kn}={\displaystyle \sum _{j\in D\cup C}{t}_{js}\times y_{jkr}}+{\displaystyle \sum _{j\in D\cup C}{d}_{ij}\times x_{ijkn}/v_{kn}},\forall k,n,i\in D\cup C$$

(10)

$${\displaystyle \sum _{i\in S}{\displaystyle \sum _{j\in S}{x}_{ijkn}>1}},\forall S\in C,k\in K,n\in N$$

(11)

$$x_{ijkn}\in \left\{0,1\right\},\forall \left(i,j\right)\in D\cup C,\forall k\in K,\forall n\in N$$

(12)

$$y_{ikn}\in \left\{0,1\right\},\forall i\in C,\forall k\in K,\forall n\in N$$

(13)

$$G_{kn}\in \left\{0,1\right\},\forall k\in K,\forall n\in N$$

(14)

$$g_{kn}\in \left\{0,1\right\},\forall k\in K,\forall n\in N$$

(15)

where the objective function (3) represents the total cost consisting of fuel consumption, vehicle rent, and fixed running cost of the vehicle; constraint (4) indicates that each customer is served by only one vehicle; constraint (5) indicates that the vehicle serving the customer must start from that customer; constraint (6) indicates that the total demand of the customer serving the vehicle is not equal to that of the customer serving the vehicle; constraint (7) means that the number of vehicles available in the distribution center does not exceed the total number of vehicles owned and rented by the distribution center; constraint (8) means that the distribution vehicles return to the distribution center after departure from the distribution center; constraint (9) determines the load of the vehicle k with type n from node i to j in operation; constraint (10) denotes the total time consumed by a vehicle to complete its service; constraint (11) denotes subway elimination constraints; and constraints (12) to (15) are constraints of 0–1 variables.

4. Design Adaptive Genetic Algorithm

The multi-distribution center vehicle routing problem (MDVRP) is an NP-hard problem. When considering the factors of vehicle type, load, vehicle sharing, and leasing, it is difficult to obtain a satisfactory solution by using a traditional solving algorithm. When the number of customers increases rapidly, it is difficult for general intelligent algorithms to obtain good solutions quickly. Aiming at this dilemma, this paper designs an improved adaptive genetic algorithm by referring to the advantages of the wide search range of a genetic algorithm, effective classification of customers by a scanning algorithm, and efficient optimization of local distribution lines by a two-sided successive correction algorithm. The improved adaptive genetic algorithm has strong robustness and is suitable for parallel processing. Combined with genetic operators, it can improve the global search ability and convergence rate of the algorithm.

4.1. Chromosome Coding Design

There are many factors that need to be considered when designing algorithms for VRP problems: firstly, when the distribution center calls the vehicle, which type of vehicle will be selected for which customer service; secondly, when the distribution center needs to rent a vehicle, how does the chromosome distinguish between its own and renting a vehicle; thirdly, when the vehicle in the distribution center is shared, how will it determine the direction and model number of the vehicle shared by any distribution center on the chromosome; and finally, in the process of algorithm programming, how to determine the optimal allocation scheme of customers and vehicles in the constraints of load changes, vehicle models, and differences in one’s own and rented vehicles is key.

The customers’ chromosomes are all coded in decimals. In order for the chromosome to contain information such as distribution center, model, rental vehicle, and fuel consumption change, this paper proposes the concept of an information matrix. By decoding the initial chromosome, the corresponding information matrix can be obtained, and then a detailed distribution scheme can be obtained. In the process of decoding a chromosome into an information matrix, some rules need to be followed. Each distribution center prioritizes the use of its own vehicles and can only call the vehicle to the nearest capacity-rich distribution center if its own capacity is insufficient. When the distribution center is rich in capacity, it should first meet its own distribution tasks before sharing the vehicles with other distribution centers. When the entire distribution center shares the vehicle, the vehicle with the right model is only rented when the capacity is insufficient. The distribution center vehicle and the rental vehicle are both from a distribution center and finally return to the nearest distribution center. The end customer in each path of the chromosome is connected to the nearest distribution center, and the starting customer is connected to the distribution center closest to the distance at which the vehicle can be delivered. Vehicles are allowed to be redundant, but empty vehicles are not allowed to be delivered.

Based on the above analysis, the solution of the scheme is represented by natural number coding, with the customer number as the element of the solution, and 0 as the distribution center. Each solution can be divided into several routes, which must contain all customer points. Each route contains the distribution center and at least one customer point. The routes of the solution are separated by distribution center 0. For example, 0-1-2-3-4-5-0-6-7-0 represents a feasible solution at this scale.

For example, there are three distribution centers, 12 customers, and the company has three models each; the vehicles’ net weights and maximum loads are, respectively, 1, 2, 3 and 4, 6, 10 (unit: ton). The rental company has one of three models. A chromosome can be interpreted as shown in

Table 2. Chromosome decoding.

Number	Contents	Information Matrix
1	Customer Number (Chromosome) C	3	2	1	9	5	6	11	8	10	4	12	7
2	Corresponding to customer needs Q	q1	q2	q3	q4	q5	q6	q7	q8	q9	q10	q11	q12
3	Departure distribution center D	1	/	/	2	/	/	3	/	/	3	/	/
4	Return to the distribution center D	/	/	1	/	/	1	/	/	2	/	/	3
5	Whether to rent a vehicle (0 No, 1 Yes)	0	0	0	0	0	0	0	0	0	1	1	1
6	Vehicle model (load/ton)	4	4	4	6	6	6	10	10	10	6	6	6
7	Goods weight (unit/ton)	w1	w2	w3	w4	w5	w6	w7	w8	w9	w10	w11	w12
8	Gross vehicle weight (unit/ton)	W1	W2	W3	W4	W5	W6	W7	W8	W9	W10	W11	W12

. The detailed information of the vehicle can be obtained by interpreting the obtained information matrix, as shown in

Table 3. Interpretation of the distribution plan.

Route	Departure Distribution Center	Service Customer Order	Return to the Distribution Center	Vehicle Model (Load/ton)	Whether to Rent a Vehicle
Route 1	1	3->2->1	1	4	No
Route 2	2	9->5->6	1	6	No
Route 3	3	11->8->10	2	10	No
Route 4	3	4->12->7	3	6	Yes

. If the fitness function and the information matrix are called, this can obtain detailed information on the delivery cost, time, fuel consumption, and so on.

4.2. Population Initialization

In order to construct the chromosome structure described in the text, firstly, the natural number arrangement of any one customer group is obtained by a scanning algorithm, that is, the corresponding chromosome, and the above behavior is repeated to obtain N chromosomes (assuming the population size is N).

The specific steps to obtain the initial population by the scanning algorithm are as follows:

Step 1: Obtain the coordinates of the multi-distribution center and each customer and use the center of gravity of the multi-distribution center as the virtual total distribution center to serve as the pole;

Step 2: Use any customer and pole connection as the polar coordinate of the polar axis. Based on the polar axis, scan all nodes clockwise or counterclockwise and sort by scanning order, for multiple customers at the same angle. Then, according to the distance, the distance is sorted from small to large, and after one round of scanning, the customer order is obtained as a corresponding chromosome;

Step 3: Shift a certain angle clockwise to a customer and repeat step 2 to finally obtain a population of size N.

4.3. Genetic Operator Design

In order to make the individuals with high fitness in each generation inherit, the individuals with low fitness have a large probability of cross mutation, and the algorithm adopts adaptive rules. The adaptive crossover probability P_c and the mutation probability P_m production function are, respectively,

$$P_c=pc-(2pc-1)\sin \frac{\pi (f-f_{\min })}{f_{\max }-f_{\min }}$$

(16)

$$P_m=1-0.9\sin \frac{\pi (f-f_{\min })}{2(f_{\max }-f_{\min })}pm$$

(17)

where $pc$ and $pm$ are the cross mutation probability given by the algorithm, respectively, $f_{\max }$ and $f_{\min }$ are the maximum fitness value and the minimum fitness value, respectively, in each generation, and $f$ is the fitness value of the individual.

Selection operation: In order to eliminate the individuals with low fitness and ensure that the outstanding individuals are inherited, this paper adopts the gambling disc selection method.

Crossover: In order to expand the search scope of the algorithm and ensure that the excellent genes are not destroyed, this paper uses a two-point crossover operation. The specific steps are as follows: first, select two chromosomes as father 1 and father 2; secondly, randomly generate two natural numbers r1 and r2 (r1 < r2) that do not exceed the length of the chromosome for father 1; again, obtain father 1. The order of all gene segments between the gene positions r1 and r2 is the father 2, this arrangement is to cover the gene between r1 and r2 in the father 1 to become the son 1; finally, a similar operation is performed on the father 2 and son 2.

Mutation operation: In order to ensure the random search ability of the algorithm and the diversity of the population, this paper uses a two-point mutation operation. First, two natural numbers r3 and r4 that do not exceed the length of the chromosome are randomly generated. Second, two of the genes in chromosomes r3 and r4 in the chromosome requiring mutation operation are determined; finally, the two genes are exchanged to generate a new chromosome. The specific operation of cross mutation is shown in Figure 2.

4.4. Fitness Function

The fitness function evaluates the merits of each chromosome and selects excellent individuals through the fitness function to perform other operations, so as to obtain better individuals. The fitness function in this paper is set as the reciprocal of the value of the objective function.

4.5. Termination Conditions

When the algorithm reaches the set maximum number of iterations or the population is not in evolution, the entire operation is terminated, and the specific scheme and delivery cost of the optimal vehicle transportation obtained by this iteration are displayed.

5. Analysis of Examples

5.1. Example Data

This study is based on the relative data of a logistics company. The customer and distribution center information and vehicle type information are shown in

Table 4. Customer and distribution center information.

Node Number	Coordinate/km	Demand /t	Node Number	Coordinate/km	Demand/t	Node Number	Coordinate/km	Demand/t
1	[37,52]	1.2	21	[62,42]	0.6	41	[10,17]	3.2
2	[49,49]	3.5	22	[42,57]	3.6	42	[21,10]	3.7
3	[52,64]	2.1	23	[16,57]	2.1	43	[5,64]	1.6
4	[20,26]	1.4	24	[8,52]	1.5	44	[30,15]	3.8
5	[40,30]	2.6	25	[7,38]	3.3	45	[39,10]	2.7
6	[21,47]	2.0	26	[27,68]	1.2	46	[32,39]	0.5
7	[17,63]	3.6	27	[30,48]	4.0	47	[25,32]	3.0
8	[31,62]	3.8	28	[43,67]	3.8	48	[25,55]	3.9
9	[52,33]	1.6	29	[58,48]	2.9	49	[48,28]	2.3
10	[51,21]	0.5	30	[58,27]	2.4	50	[56,37]	3.5
11	[42,41]	3.9	31	[37,69]	3.5	51	[20,37]	0
12	[31,32]	3.4	32	[38,46]	1.7	52	[37,48]	0
13	[5,25]	3.8	33	[46,10]	2.8	53	[48,20]	0
14	[12,42]	2.6	34	[61,33]	3.1
15	[36,16]	1.5	35	[62,63]	2.2
16	[52,41]	4.0	36	[63,69]	0.6
17	[27,23]	0.8	37	[32,22]	1.4
18	[17,33]	4.6	38	[45,35]	3.9
19	[13,13]	1.4	39	[59,15]	1.9
20	[57,58]	3.3	40	[5,6]	3.7	Total demand	130

and

Table 5. Vehicle information.

Vehicle Model	Distribution Center D Own Number of Vehicles (Units)			The Number of Vehicles That Can Be Rented (Units)	Maximum Load (ton/T)	Vehicle Net Weight	Private Vehicle Fixed Fee (yuan/hour)	Rental Vehicle Fee (yuan/hour)
	d = 1	d = 2	d = 3
Type1	0	1	2	3	14	5	25	40
Type2	1	2	0	3	10	4	20	35
Type3	2	1	2	3	8	3	15	20

. In Table 3, the numbers are 1–50 for customer information, and the nodes 51–53 are coordinate information of distribution centers 1, 2, and 3.

5.2. Calculation Results

This paper used the MATLAB2016 software design algorithm program, with a set group size of 100, 300 iterations, and an initial cross mutation probability of 0.9 and 0.1, respectively. The results are shown in

Table 6. Final optimization plan.

Route	Departure Distribution Center	Service Customer Order	Return Distribution Center	Vehicle Model	Vehicle Rental	Travel Distance	Transit Time	Fixed Cost		Variable Costs	Total Cost
								Own vehicle	Rental vehicle
1	2	1->28->31->26->48	2	14	No	63.58	47.68	19.87	/	38.26	58.13
2	2	22->8->6->46	2	10	No	64.30	48.23	16.08	/	31.87	47.94
3	2	32->20->35->36->3	2	10	No	71.88	53.91	17.97	/	36.32	54.29
4	2	11->2	2	8	No	31.27	23.46	5.86	/	14.35	20.21
5	3	16->21->29	2	8	No	59.64	44.73	11.18	/	28.02	39.20
6	3	10->49->9->50->34->30	3	14	No	48.16	36.12	15.05	/	28.95	44.00
7	3	45->33->39	3	8	No	46.47	34.85	8.71	/	21.56	30.28
8	3	15->44->42->17->37->5	3	14	No	72.56	54.42	22.68	/	42.53	65.21
9	1	47->41->13	1	10	No	56.93	42.70	14.23	/	29.62	43.85
10	1	4->19->40	1	8	No	70.83	53.12	13.28	/	31.34	44.62
11	1	18->12	1	8	No	31.12	23.34	5.83	/	13.93	19.77
12	1	25->14->23	1	8	Yes	55.36	41.52	/	13.84	25.08	38.92
13	1	7->43->-24>	1	8	Yes	69.79	52.34	/	17.45	32.19	49.63
14	2	27->38	2	8	Yes	42.11	31.59	/	10.53	19.24	29.76
					Total	784.00	588.00	150.74	41.82	393.25	585.82

and Figure 3. From the chart, the design algorithm could automatically classify customers, share, and rent vehicles properly. In the optimization result, the total driving distance was 784 km, the total time of the vehicle was 588 min, the shortest time of line 11 was 23.34 min, and the longest time of line 8 was 54.42 min. The total cost of this program was CNY 585.82, of which the fixed cost was CNY 192.56, and the total rental cost was CNY 41.82; the variable cost was CNY 393.25, and the loading rate of each vehicle was higher.

5.3. Comparative Analysis of Algorithms

In order to further verify the effectiveness of the proposed algorithm, the improved adaptive genetic algorithm was compared with the genetic algorithm and the simulated annealing algorithm, respectively. Taking 65 customer location nodes as examples, the calculation results are shown in

Table 7. Comparison of algorithm performance.

	Genetic Algorithm		Simulated Annealing Algorithm		Improved Adaptive Genetic Algorithm
	The Average Solution	Optimum Solution	The Average Solution	Optimum Solution	The Average Solution	Optimum Solution
Non-Vehicle Sharing	1836.21	17885.10	1779.57	1741.85	1718.06	1709.92
Vehicle Sharing	1362.98	1347.26	1330.18	1312.23	1296.7	1290.42

.

After calculation, in the non-sharing mode of vehicles, the three algorithms all used seven common vehicles to complete the transportation of 65 customer points, and the average and optimal solutions obtained by the improved adaptive genetic algorithm were better than the genetic algorithm and simulated annealing algorithm. The cost of the average solution obtained by the improved adaptive genetic algorithm was 6.73% lower than that of the genetic algorithm and 3.73% lower than that of the simulated annealing algorithm.

In the vehicle sharing mode, due to the same matching algorithm, the matching rate of cooperative vehicles in the same situation was the same, and there were 65 customers completed by cooperative vehicles. The three algorithms all used seven ordinary vehicles to complete the distribution of the remaining customer orders. The optimal and average solutions obtained by the improved adaptive genetic algorithm were still derived from the genetic algorithm and the simulated annealing algorithm.

6. Conclusions

Aiming at the problem of lack of dispatching ability or transportation capacity of logistics enterprises in multiple distribution centers, a network cooperative distribution strategy of vehicle sharing was proposed. Based on the comprehensive consideration of multi-distribution center, multi-model, rental vehicle, load, speed, fuel consumption, and other factors, the calculation method of vehicle energy consumption was introduced, and the network collaborative distribution model with vehicle sharing was established on the basis of considering multi-constraint conditions. Since the model is a strong NP problem, and it is difficult for the algorithm to encode chromosomes due to multiple constraints, the concept of an information matrix was proposed to simplify the coding difficulty. At the same time, the scanning algorithm and two-edge successive correction algorithm were used to improve the genetic algorithm to enhance the searching ability and convergence speed of the algorithm. Finally, the reliability and applicability of the model and algorithm were verified by numerical experiments. The results showed that vehicle sharing can improve the efficiency of the distribution network and effectively reduce costs, the proposed model and solution strategy are feasible, and the algorithm has obvious advantages in solution accuracy and efficiency. This research can be used as a reference for logistics enterprises to implement vehicle sharing and leasing strategies. The service time window, dynamic change factors of customer information, and energy consumption will be considered in further research, and the advantages of a tabu search algorithm, ant colony algorithm, particle swarm optimization algorithm, and other algorithms will be used for reference to study the problem.