Research on Express Crowdsourcing Task Allocation Considering Distribution Mode under Customer Classification

Abstract

In order to promote the sustainable development of crowdsourcing logistics and control the cost of crowdsourcing logistics while improving the quality of crowdsourcing services, this paper proposes a courier crowdsourcing task allocation model that considers delivery methods under customer classification, with the optimization objective of minimizing the total cost of the crowdsourcing platform. This model adopts two delivery modes: home delivery by crowdsource couriers and pickup by customers. Customers can freely choose the express delivery method according to their actual situation when placing orders, thus better meeting their needs. Based on the customer’s historical express-consumption data, the entropy weight RFM model is used to classify them, and different penalty functions are constructed for different categories of customers to reduce the total delivery cost and improve the on-time delivery of efficient and potential customers. And a Customer Classification Genetic Algorithm (CCGA) was designed for simulation experiments, which showed that the algorithm proposed in this study significantly improved the local search ability, thereby optimizing the delivery task path of express crowdsourcing. This improvement not only improves the delivery timeliness for efficient and potential customers, but also effectively reduces the total delivery cost. Therefore, the research on parcel crowdsourcing task allocation based on customer classification reduces the cost of crowdsourcing delivery platforms and improves customer satisfaction, which has certain theoretical research value and practical-application significance.

1. Introduction

Crowdsourcing logistics, as a flexible and cost-effective distribution model, is becoming an important part of the modern logistics industry. However, to realize the sustainable development of crowdsourcing logistics, continuous improvement of service quality, and effective cost control are two key factors. First, high-quality service is the key to improving customer satisfaction, and crowdsourcing logistics needs to ensure the timeliness, accuracy and safety of distribution to meet customer expectations. Second, cost control is extremely important for crowdsourcing logistics to achieve sustainable development. Effective cost control can improve the economic efficiency of crowdsourcing logistics and ensure the sustainability of the business. In order to achieve an effective balance between customer satisfaction and cost control, crowdsourcing logistics platforms not only need to actively collect and analyze customer feedback to continuously improve their services, but also adjust their cost structure to adapt to market changes.

With the development of the internet and the improvement of residents’ consumption level, e-commerce shopping has increasingly become the most common way of shopping. The express delivery industry is facing many challenges, especially during the promotion period of e-commerce websites, where the transportation volume of various express delivery companies has surged, and packages are severely delayed at delivery points. The number of couriers does not match the number of individual transactions. Our self-operated logistics teams are far from meeting the exponential growth of online express delivery volume, and as such, crowdsourcing logistics has emerged. As an emerging logistics method, many scholars have conducted research on crowdsourcing logistics, mainly focusing on its feasibility, cost-effectiveness, risk management, technological innovation, and legal policies. Research has shown that crowdsourcing logistics does have unique advantages compared to traditional logistics [1,2]. Seghezzi and Mangiaracina (2022) [3] estimate that the cost of achieving customer satisfaction through crowdsourcing logistics and traditional last mile delivery services will be lower. Therefore, the “last mile” transportation problem of express delivery can be solved through crowdsourcing logistics. Essentially, crowdsourcing is a sharing economy that shares the time and energy of idle individuals in society. It is a rational allocation and effective utilization of idle labor, and a good supplement to China’s current express delivery industry.

When customers choose express delivery services, they have shifted from the initial price orientation to the current service quality orientation, and the timeliness and customization requirements of express delivery are becoming increasingly strict. However, the current situation of express delivery has not been greatly improved, and most of them ignore the historical consumption behavior factors of customers. The delivery mode of express delivery cannot meet the needs of customers well, often resulting in low delivery efficiency. Peng et al. (2021) [4] defined the user satisfaction utility as composed of user preference utility, delay waiting utility, and task completion expectation, and proposed a crowdsourcing task allocation method based on user satisfaction utility. However, he did not distinguish between consumers, and in previous research on delivery issues, most of them ignored the consumer behavior factors of customers, assuming that the value of customers to crowdsourcing platforms is the same, which is obviously not realistic. This work adopts targeted strategies for customers by studying their previous consumption behaviors, which is conducive to maintaining customer loyalty and long-term development of enterprises. Yu et al. (2020) [5] studied the real-time distribution route optimization problem considering customer classification, and divided it into three categories from the perspective of customer consumption behavior. Different degrees of penalty costs were imposed on three types of customers who violated time window constraints to reflect the importance that enterprises attach to different customers. But the author only considered the situation of one distribution center, and this paper aims to study the vehicle routing problem between multiple distribution centers based on actual express delivery situations. The problem becomes more complex when the couriers receive the package at the corresponding distribution center before delivering. On 1 March 2024, the newly revised “Measures for the Administration of the Express Delivery Market” stipulated the phenomenon of “delivery without notice” for express delivery: without the consent of the customer, reception of the express delivery on behalf of the customer is not allowed, and delivering the express delivery to intelligent express boxes, express service stations, or other express end service facilities without authorization is also not allowed. But there are also situations where delivery personnel need to call each customer to inquire, resulting in high time costs or the inability to contact customers. Especially in the crowdsourcing environment, the lack of professional training for crowdsourcing delivery personnel can easily overlook customer needs and lead to arbitrary work. In actual delivery, the ability of crowdsourced delivery personnel to deliver on time, in compliance with regulations, and safely is indeed a major challenge for express crowdsourcing. Specifically, this work solves the following problems:

(1) Considering the differences in customer value, the entropy weighted RFM model is used to classify customers into efficient customers, potential customers, and marginal customers based on their historical consumption behavior, and different penalty functions are established to enable customers to obtain distribution services that match their value, rather than treating customers uniformly in previous research.

(2) In the whole process from online order to express receipt, the crowdsourcing platform adds two delivery modes of home delivery by crowdsource couriers and pickup by customers; customers can choose according to the actual situation, which simplifies the process of express delivery.

In this work, crowdsourcing logistics is applied to the express delivery industry. Considering the customer’s consumption behavior, the customers are classified, and the customer classification genetic algorithm (CCGA), considering the distribution mode under the customer classification, is established to solve the problem. The efficiency of the algorithm and the influence of the time window on the route planning decision and distribution performance of distribution enterprises are investigated. The rest of this article is as follows: Section 2 reviews the relevant literature. Section 3 introduces the method of customer classification, the penalty function of different customer categories, and the crowdsourcing task allocation model considering distribution mode under customer classification. Section 4 designs a genetic algorithm (CCGA) based on customer classification. Section 5 gives the experimental results under different algorithms and discusses the meaning of the results. In Section 6, the contributions of this work are summarized and suggestions for future research are put forward.

2. Literature Review

2.1. Crowdsourcing Logistics

Carbone et al. (2017) [6] studied crowdsourcing logistics is a new way to provide logistics services by utilizing mobile applications, web platforms, and individuals’ idle logistics resources and capabilities, which can solve the “last mile” transportation problem of express delivery by utilizing local couriers. Archett et al. (2016) [7] combined crowdsourcing logistics and professional logistics, and studied the classic 6-capacity vehicle routing problem with the coexistence of temporary and professional drivers. Seghezzi et al. (2021) [8] designed a heuristic algorithm to minimize the total transportation cost. Chen et al. (2020) [9] proposed a new framework called Crowd Express to use cabs to transport parcels in a crowdsourcing way, and experiments show that this can simultaneously reduce costs and improve the speed of courier delivery. The above research indicates that crowdsourcing logistics is indeed a practical and feasible delivery method. This article extends the crowdsourcing model for delivery in the express delivery field to the express delivery field, and allocates express delivery tasks through a crowdsourcing platform to improve delivery efficiency and reduce platform costs

2.2. Customer Classification

Most of the current research on the courier crowdsourcing task allocation problem ignores the factor of customer consumption behavior; that is, it assumes that different customers have the same value to the courier company, which is obviously not in line with the reality. In 1956, Wendell et al. [10] put forward the concept of “segmentation”; through the segmentation of customers, the development of appropriate marketing strategies could effectively reduce the cost of enterprises. After that, many scholars segmented customers according to their purchasing trends and proposed different marketing strategies after segmenting customer groups [11,12,13,14]. Elham et al. (2023) [15] explored the relative importance of market segmentation when evaluating consumers’ online purchasing behavior. The data were collected through a standard AHP questionnaire and the results of the study showed that age, gender, and marital status were the most critical determinants of online consumer behavior and group influence, adaptability, and brand loyalty were found to be the least important factors, thus encouraging companies to target their consumers based on the necessary categories as a way of stimulating consumers to shop online.

A number of scholars have also studied the methods of customer classification. Christy et al. (2021) [16] refer to the transaction data of the company’s customers over a specific period of time, perform effective segmentation based on the RFM model (recency, frequency and monetary), use the traditional K-means algorithm and fuzzy C-means algorithm to cluster the customers, so as to identify the potential customers of the company, to increase the company’s revenue and make marketing strategies for each segment. Hosseini et al. (2015) [17] used the RFM model and K-means clustering method to categorize customers. The innovation of this literature is that it incorporates the time and trend in customer value changes and improves the accuracy of prediction based on customer’s past behavior. Kansal et al. (2023) [18] proposed a Heterogeneous Integrated Learning Approach for Multi-Level Default Risk Rating Strategy (MLDRR) and constructed a new Credit Customer Eight Segmentation Model (ESM) to segment credit customers into eight important categories such as highest risk customers, potentially risky customers, target customers, etc. Wu et al. (2011) [19] developed a soft clustering approach to categorize online customers based on their cross-category purchasing data. Anitha et al. (2022) [20] proposed a new method for predicting customer value, RFM/P, which combines product and customer marketing perspectives by considering both customer and product perspectives. The overall customer value is derived by estimating and aggregating the customer value corresponding to each product, which eliminates the need to choose between product and customer perspectives to provide a more complete overview of the company’s future cash flows. Based on the traditional RFM model, Xu (2024) [21] introduces the NES theory and constructs a user profiling algorithmic model of RFM + NES in order to study the potential and persistence of customers and provide new ideas for customer relationship management. The above literature classifies customers from different perspectives and improves classification models and methods, indicating that adopting different strategies for classifying customers can better improve the efficiency of enterprises. This paper improves the RFM model by using the entropy weight RFM model to classify customers.

2.3. Crowdsourcing Task Allocation

Tong et al. (2020) [22] stated that task allocation, quality control, incentive mechanism design, and privacy protection are four core algorithmic problems in the field of crowdsourcing. The task allocation problem of express crowdsourcing studied in this article refers to assigning the tasks of express crowdsourcing to crowdsourced couriers and planning reasonable delivery routes for them. In recent years, scholars at home and abroad have expanded the research on crowdsourcing task allocation with the goals of maximizing satisfaction, minimizing platform cost, etc. Song et al. (2020) [23] believe that task allocation should be executed in real time based on partial information and propose an online greedy algorithm which finds new tasks or staff with less cost. Li et al. (2021) [24] study crowdsourcing task allocation based on employer net profit and the employee satisfaction problem. Zhao et al. (2017) [25] proposed a trust assessment model that considers contextual information and the requester’s needs, and designed an effective trust-oriented worker selection algorithm. Bask et al. (2018) [26] addressed the inadequacy of task assignment in modeling dynamic fairness metrics, and introduced a new crowdsourced delivery task assignment strategy. The strategy considers fairness to workers while maximizing the task allocation ratio. Wang et al. (2020) [27] proposed the maximum quality and minimum cost task allocation problem using the MQC-GAC algorithm to optimize task allocation. Due to the relatively low skill level and credibility of workers in crowdsourcing platforms, there may exist a large number of unassignable tasks and taskless workers; Jiang et al. (2023) [28] designed a worker cooperation model to provide an improved strategy for crowdsourcing platforms, which facilitates cooperation and mutual assistance among workers, thus saving budgets and improving the overall task assignment success rate. Peng et al. (2022) [29] considered the service quality of workers and transformed the task allocation problem into a combinatorial optimization problem through delayed matching. They used an improved firefly swarm algorithm to solve the task allocation problem, effectively improving the positive incentive amplitude and incentive fairness for workers. Wang et al. (2022) [30] proposed a new multi-objective cold-chain logistics distribution optimization method based on customer level classification. By categorizing customers based on their value levels, priority delivery points can be determined. Based on the objective functions of maximizing customer satisfaction and minimizing delivery costs, a logistics delivery path model was constructed. The experimental results showed that compared with traditional methods, the above method effectively reduced the delivery cost of cold-chain logistics, and the delivery time was more reliable, improving customer satisfaction. This paper also aims to minimize the cost of crowdsourcing platforms, allocate tasks for express crowdsourcing, and design the CCGA to solve them to accelerate delivery time and save costs.

In summary, the above literature is helpful in solving a series of problems in the delivery of express delivery enterprises, and has enlightening significance for the research on express crowdsourcing delivery in this study. However, there is still a lack of systematic optimization research on the task allocation problem of express crowdsourcing from the perspective of customers. And this article considers the impact of different customers on crowdsourcing platforms, dividing customers into three categories based on their historical express delivery consumption behavior. The difference from the above literature is that it applies crowdsourcing logistics to the express delivery industry and studies the task allocation problem of express crowdsourcing considering delivery methods based on customer classification. Unlike previous studies, which only adopted a single delivery mode, in order to improve the drawbacks of the crowdsourcing delivery mode for express delivery, two alternative delivery modes have been added to the previous single delivery mode, namely delivery to home and self-pickup, to reduce the cost of the crowdsourcing logistics platform and improve customer satisfaction. This also provides a relevant scientific basis for the development of express crowdsourcing logistics, and has certain reference value for similar logistics problems.

3. Problems and Models

3.1. Problem Description and Assumptions

During the whole process from online ordering to express receipt, each customer has two delivery modes: home delivery or pickup. If the customer chooses the home delivery mode, they need to choose the time window for receiving the express delivery conveniently, and the courier will send the express delivery to the customer according to the receiving address filled by the customer; if the customer chooses the pickup mode, the courier will deliver the express to the express station near the receiving address, and the customer will go to the express station to retrieve it on their own. After the crowdsourcing couriers are assigned an order by the platform, they need to complete the delivery service according to the delivery order and delivery path planned by the platform to the corresponding delivery point. The courier must first pick up the goods at the corresponding distribution center before delivery. The crowdsourcing task assignment result diagram is shown in Figure 1. Tasks 1, 2, 7, 9, and 10 belong to Distribution Center 1. Tasks 3, 4, 5, 6, 8, 10, and 12 belong to Distribution Center 2. Tasks 6, 11, 13, and 14 belong to Distribution Center 3. The task delivery sequence of courier 1 is {1, 2, 3}, and its delivery route is {Distribution Center 1, Task 2, Task 1, Distribution Center 2, Task 2}. The task delivery sequence of courier 2 is {4, 5, 6}, and its delivery route is {Distribution Center 2, Task 4, Task 5, Task 6}. The task delivery sequence of courier 3 is {7, 8, 9, 10}, and its delivery route is {Distribution Center 1, Distribution Center 2, Task 9, Task 10, Task 8, Task 7}. The task delivery sequence of courier 4 is {11, 12, 13, 14}, and the delivery route is {Distribution Center 3 Task 14, Task 11, Distribution Center 2, Task 12, Task 13}.

This paper analyzes the data from October 2023 to December 2023 on the trading platform of a certain branch of the Shanghai D Express Company. In the traditional RFM model, R represents the time of the most recent transaction, F represents the frequency of transactions during this period, and M represents the amount of transactions during this period. In the model of crowdsourcing delivery in this article, these three indicators are given new meanings. R represents delivery receipt recency, which is the latest time the customer signed for the express delivery. F represents the frequency of express delivery receipts, which refers to the total number of times customers receive express delivery during this period. M represents the amount of express delivery freight, which is the freight paid by the customer for signing for the express delivery during this period. The comparison of the indicators between the traditional RFM model and the express entropy weight RFM model is shown in

Table 1. Comparison of Indicator Meanings between Traditional RFM Model and Express Entropy Weight RFM Model.

Model	R	F	M
Traditional RFM model	Recency	Frequency	Monetary
Entropy weighted RFM model	Express delivery receipt recency	Express delivery receipt frequency	Express delivery monetary fee

.

The traditional RFM model assumes that the weights of the R, F, and M indicators are the same. However, in reality, the impacts of R, F, and M on crowdsourcing platforms are different. In order to classify customers more accurately, this article evaluates the importance of each indicator in the RFM model. The method suggests the following actions: assign corresponding weights to each indicator of the traditional RFM model using the entropy weight method [31], establish a new entropy weight RFM model, and classify customers based on this model. This method combines the advantages of entropy weight method and RFM model, and different customer values can be reflected through the different proportional allocations of these three indicators, which can more accurately evaluate customer value and behavior, and provide strong support for business decision-making, thereby improving customer satisfaction and business growth.

In the calculation results based on SPSS26.0 software, express delivery monetary fee has the highest weight, accounting for 65.5% of the proportion, indicating that the customer’s express freight amount has significant importance in this research analysis. Next is express delivery receipt frequency, accounting for 29.099% of the total, indicating that the frequency of customers’ express delivery receipts is more important in this research analysis. However, the weight of express delivery receipt recency is the lowest, only 5.401%, indicating that the importance of this indicator in research analysis is relatively small, as shown in

Table 2. Weight of RFM model indicators.

Term	Information Entropy Value e	Information Utility Value d	Weight (%)
express delivery receipt recency	0.991	0.009	5.401
express delivery receipt frequency	0.954	0.046	29.099
express delivery monetary fee	0.897	0.103	65.5

.

This study will assign weights to these three indicators based on the entropy weight method and update them with weighted weights. According to the updated data, we use the K-means clustering method to classify customer categories. In Python 3.12.6 software, the elbow method is first used to determine the optimal clustering value k. As the number of clusters k increases, the squared error SSE will naturally gradually decrease. The point where SSE begins to slow down is found as the optimal value k. According to the elbow diagram in Figure 2, it can be seen that increasing the number of clusters no longer significantly reduces clustering errors after k = 3. Therefore, in this paper, k = 3 is taken to cluster customers into three categories.

By using the entropy weight method to determine the weights, it can be concluded that express delivery monetary fee has the greatest impact on the express crowdsourcing platform, bringing more profits to the platform. Express delivery receipt frequency also has a significant impact on the platform, while express delivery receipt recency has a relatively small impact on the platform. Therefore, different proportions of these three indicators can be used to reflect different customer values. By observing Figure 3, Figure 4 and Figure 5, the following conclusions can be drawn: Cluster 1 customers have the highest express delivery monetary fee, moderate express delivery receipt frequency, and relatively low express delivery receipt recency, indicating that Cluster 1 customers are efficient customers; Cluster 2 customers have the highest express delivery receipt frequency, while the express delivery monetary fee and the express delivery receipt recency are moderate, indicating that cluster 2 customers are potential customers; the express delivery receipt frequency and the express delivery monetary fee for cluster 0 customers are relatively small, while the express delivery receipt recency is relatively high, indicating that cluster 0 customers are edge customers. Based on these results, this study categorizes customers into three types: efficient customers, potential customers, and edge customers.

This paper considers the scenario of multiple distribution centers, multiple express stations, and multiple couriers. In actual express delivery scenarios, when a customer places an order, their task location, time window, and other relevant information will be obtained by the crowdsourcing platform and pushed to the corresponding crowdsourcing couriers. A crowdsourcing courier will be assigned different tasks, and they usually use the same transportation tools to start delivery from their respective locations. Due to the nature of crowdsourcing, they do not need to return to the distribution center. In the process of building a model, in order to ensure that the model can be effectively solved, the following reasonable assumptions need to be made based on the actual situation.

(1)

When the crowdsourcing platform assigns tasks to couriers, the customer category, task location, task time window, delivery method selected by the customer, courier location, courier station location, and delivery center location are all known.

(2)

Each crowdsourced delivery person starts from their own starting position and completes the crowdsourced delivery task.

(3)

Each customer point is served by only one delivery person, who can only arrive and depart once, but each crowdsourced delivery person can serve multiple customer points.

(4)

The delivery speed of each crowdsourced deliveryman is the same and constant, without any traffic congestion.

(5)

Crowdsourced couriers will not refuse delivery tasks assigned by crowdsourcing platforms.

(6)

The service time of crowdsourced couriers at the distribution center is the same as the service time of home delivery and pickup delivery methods.

(7)

If arriving at the customer point earlier or later than the time window, there will be opportunity costs or penalty costs.

(8)

For the convenience of describing the model, it is assumed that each crowdsourced delivery person will not return to their initial position after completing the delivery.

3.2. Denotational Description

3.2.1. Denotational

In this question, it is assumed that there are multiple distribution centers and courier stations: distribution center $h\in H$, H $=\left\{1, 2, 3, \cdots , h\right\}$. Express station node $u\in U$, U $=\left\{1, 2, 3, \cdots , u\right\}$. Initial location node of crowdsourcing couriers $c\in C$, C $=\left\{1, 2, 3, \cdots , c\right\}$. Express crowdsourcing customer nodes $n\in N$, N $=\left\{1, 2, 3, \cdots , n\right\}$. Crowdsourced delivery vehicle serial number $k\in K$, K $=\left\{1, 2, 3, \cdots , k\right\}$. The delivery vehicles have the same load capacity, represented by $q_k$. The weight of each customer’s crowdsourcing task is represented as $w_n$. The time window for customer n is represented as $\left[e_n, l_n\right]$; $e_n$ represents the earliest service time requested by customer, n, $l_n$ represents the latest service time requested by customer n. The distance between node n and node m is represented by the Euclidean distance $d_{\mathit{nm}}$. The unit transportation cost of delivery vehicles is represented by ${\rho }_k$. Vehicle speed is represented by $v_k$. The time when vehicle k arrives at node m is represented by $t_{\mathit{mk}}^1$. The waiting time of vehicle k at customer node n is denoted by $t_{\mathit{nk}}^w$. The service time of vehicle k at node m is represented by $s_{\mathit{mk}}$. The time taken for the delivery vehicle to travel from node n to node m is $t_{\mathit{nm}}$. The time when vehicle k leaves node m is represented by $t_{\mathit{mk}}^2$. $f_n$ is the delivery method chosen by the customer; if home delivery is chosen, then $f_n$ = 0; If pickup is chosen, then $f_n$ = 1.

3.2.2. Variable

$x_{\mathit{nmk}}$: 0–1 variable; if vehicle k travels from node n to node m, then $x_{\mathit{nmk}}$ = 1; otherwise, it is 0.

$y_{\mathit{nk}}$: 0–1 variable; if vehicle k completes the delivery task of customer node n, then $y_{\mathit{nk}}$ = 1; otherwise, it is 0.

All relevant symbols involved in the model are listed in

Table 3. Relevant parameter values.

Parameter	${\mathit{q}}_{\mathit{k}}$ (kg)	${\mathit{v}}_{\mathit{k}}$ (km/h)	${\mathit{\rho }}_{\mathit{k}}$ (CNY /km)	$\mathit{A}$ (CNY /min)	${\mathit{\beta }}_1$ (CNY /min)	${\mathit{\beta }}_2$ (CNY)	${\mathit{b}}_{\mathit{k}}$ (CNY/veh)	${\mathit{M}}_{\mathit{p}}$ (CNY)
Value	180	30	0.5	0.1	0.3	0.2	20	1000

.

3.3. Penalty Function of Customer Classification

According to the entropy weight RFM model, three different types of customers are clustered based on the indicators of customer historical express delivery receipt recency R, express delivery receipt frequency F, and express delivery monetary fee M. To maintain high-quality customer resources and the long-term interests of the enterprise, the time penalty function for different categories of customers should be different. Efforts should be made to meet the expected time window of efficient customers to increase platform efficiency; therefore, the penalty value for efficient customers is higher than that for potential and edge customers.

This paper assumes that if the crowdsourced delivery person arrives earlier than the earliest service delivery time requested by the customer, they need to wait until the earliest service delivery time requested by the customer. During this process, opportunity costs will be incurred. This study assumes that the early arrival penalty coefficients for the three types of customers are the same and are represented uniformly by $\alpha$. If the crowdsourced couriers arrive within the customer’s expected time window, there will be no time penalty cost. If the crowdsourced delivery person is later than the latest service time requested by the customer, different time penalty costs will be incurred based on the customer’s different categories.

For crowdsourcing platforms, efficient customers have the highest amount and frequency of express delivery receipts, which is of great significance for the long-term development of the platform and is an indispensable customer resource. Therefore, if a crowdsourced delivery person exceeds the latest service time requested by efficient customers, it will impose a significant penalty on the crowdsourced delivery person to avoid reducing the satisfaction of efficient customers. The penalty function for efficient customers is represented by $F_1$. If the crowdsourced couriers exceed the latest service time required by efficient customers, a significant penalty cost will be incurred, represented by $M_p$. The efficient customer time penalty function is shown in Equation (1).

$$F_1=\left\{\begin{array}{l}\alpha \cdot {\displaystyle \sum _{n\in N}{\displaystyle \sum _{k\in K}\max (e_n-t_{nk}^1,0)}}, \hspace{1em}{t}_{nk}^1<e_n\\ 0,e_n\le t_{nk}^1\le l_n \\ M_P, t_{nk}^1>l_n\end{array}\right.$$

(1)

Potential customers are an indispensable type of customer for the continued development and growth of crowdsourcing platforms. They have a large amount of express delivery receipts and a high frequency of express delivery receipts, and have the potential to become efficient customers. Therefore, the platform should also try its best to meet the needs of potential customers and stimulate consumption. The time penalty function for potential customers is represented by $F_2$. If the crowdsourced couriers exceed the latest service time requested by potential customers, a linear penalty cost of ${\beta }_1$ will increase over time. Potential customers incur the time penalty function as shown in Equation (2).

$$F_2=\left\{\begin{array}{l}\alpha \cdot {\displaystyle \sum _{n\in N}{\displaystyle \sum _{k\in K}\max (e_n-t_{nk}^1,0)}}, \hspace{1em}{t}_{nk}^1<e_n\\ 0,e_n\le t_{nk}^1\le l_n \\ {\beta }_1 \cdot {\displaystyle \sum _{n\in N}{\displaystyle \sum _{k\in K}\max (t_{nk}^1-l_n,0)}}, \hspace{1em}{t}_{nk}^1>l_n\end{array}\right.$$

(2)

The amount and frequency of express delivery receipts for edge customers are relatively small, and their contribution to the platform is low. Therefore, the late arrival penalty cost for this type of customer is set to a fixed value. The penalty function for edge customers is represented by $F_3$, and for crowdsourced couriers who arrive later than the latest service time requested by edge customers, a fixed value of ${\beta }_2$ is used to represent their late arrival time cost. Edge customers incur a time penalty function as shown in Equation (3).

$$F_3=\left\{\begin{array}{l}\alpha \cdot {\displaystyle \sum _{n\in N}\max }(e_n-t_{nk}^1,0), \hspace{1em}{t}_{nk}^1<e_n\\ 0,e_n\le t_{nk}^1\le l_n \\ {\beta }_2, t_{nk}^1>l_i\end{array}\right.$$

(3)

3.4. Cost Analysis

3.4.1. Fixed Costs

Fixed costs refer to a series of expenses related to delivery tasks, including vehicle depreciation and maintenance costs. These costs are related to whether the vehicles perform delivery tasks and the number of vehicles performing the tasks [32], and are not related to other factors such as the actual distance traveled by the vehicles. The cost of each vehicle is a fixed value, so this paper uses $b_k$ to represent the fixed cost per unit vehicle. The fixed cost of delivery is shown in Equation (4).

$$R_1={\displaystyle \sum _{k\in K}{b}_k}$$

(4)

3.4.2. Distance Cost

Due to the presence of multiple distribution centers, express stations, and crowdsourced couriers in this study, the distance cost of delivery refers to the cost from the previous node to the next node, which is directly proportional to the distance traveled for delivery. The distance cost is shown in Equation (5).

$$R_2={\displaystyle \sum _{n\in M}{\displaystyle \sum _{m\in M,n\ne m}{\displaystyle \sum _{k\in K}{x}_{nmk}}}}\cdot d_{nmk}\cdot {\rho }_k$$

(5)

3.4.3. Time Penalty Cost

Time penalty cost refers to the additional cost incurred during the actual delivery process due to factors such as natural environment and traffic conditions, which may cause the goods to be unable to be delivered within the scheduled time window. If the crowdsourced couriers responsible for delivery do not arrive within the expected time window of the customer, punishment will be imposed on the crowdsourced couriers. In this paper, Equation (6) is used to represent the time penalty cost. Equation (7) represents that each customer has only one category attribute. If the customer category of the task belongs to efficient customers, then $a_n$ is 1, otherwise it is 0; if the customer category of the task belongs to potential customers, then $b_n$ is 1; otherwise, it is 0; if the customer category of the task belongs to edge customers, then $c_n$ is 1, otherwise it is 0.

$$R_3={\displaystyle \sum _{n\in N}{\displaystyle \sum _{m\in N,m\ne n}{\displaystyle \sum _{k\in K}(a_n\cdot F_1+b_n\cdot F_2}}}+c_n\cdot F_3)\cdot x_{nmk}$$

(6)

$$a_n+b_n+c_n=1$$

(7)

3.5. Model Construction

This study aims to establish a crowdsourcing task allocation model for express delivery considering delivery methods under customer classification, with the goal of minimizing the total platform cost. The specific details are as follows:

$$\min R=R_1+R_2+R_3$$

(8)

$$s.t.{\displaystyle \sum _{n\in N,n\ne m}{x}_{nmk}}={\displaystyle \sum _{m\in N,m\ne n}{x}_{nmk}}=y_{nk},\forall k\in K$$

(9)

$${\displaystyle \sum _{k\in K}{y}_{nk}}=1,\forall n\in N$$

(10)

$${\displaystyle \sum _{n\in N}{w}_n}\cdot y_{nk}\le q_k,\forall k\in K$$

(11)

$$t_{mk}^1+t_{nk}^w+s_{mk}\le t_{mk}^2,\forall k\in K,\forall m\in M$$

(12)

$$t_{mk}^1=t_{nk}^2+x_{nmk}\cdot t_{nm},\forall k\in K,\forall n,m\in M$$

(13)

$$t_{nm}={\displaystyle \frac{d_{nm}}{v_k}},\forall k\in K,n,m\in M$$

(14)

$${\displaystyle \sum _{k\in K}{\displaystyle \sum _{m\in M}{x}_{nmk}}=k,\forall n\in C}$$

(15)

$$t_{nk}^w=\max (0,e_n-t_{nk}^1),\forall k\in K,\forall n\in N$$

(16)

$$x_{mnk}=x_{muk}=1,\forall m\in M,\forall n\in N,\forall u\in U,\forall k\in K,f_n=1$$

(17)

Equation (8) is the total cost objective function of the model, including fixed costs, penalty costs, and time costs; Equation (9) indicates that each vehicle can only access the same customer node once in a delivery task to avoid duplicate services; Equation (10) represents that each customer node can only be serviced by one vehicle once, ensuring the uniqueness of the service; Equation (11) indicates that the total weight of the goods delivered by the crowdsourced couriers cannot exceed the maximum load capacity of the crowdsourced couriers’ vehicles; Equation (12) represents the calculation method for the time when vehicle k leaves node m; Equation (13) represents that the arrival time of the subsequent node is equal to the departure time of the predecessor node plus the transportation time between the two; Equation (14) represents the crowdsourced couriers starting from their own location and heading to the next node for delivery; Equation (15) defines that the transportation time of a vehicle is equal to the distance traveled by the vehicle divided by its speed; Equation (16) describes the waiting time generated by the vehicle arriving at the customer node ahead of schedule; Equation (17) represents that if customer n chooses the pickup delivery method, their package will be delivered to the courier station.

4. Solving Algorithm

The express crowdsourcing task allocation model established in this work is an NP-hard problem. The adaptive crossover and mutation genetic algorithm and simulated annealing algorithm are used to overcome the premature phenomenon. The genetic algorithm (GA) was first proposed by John Holland. It is a heuristic algorithm that simulates Darwin’s theory of evolution and the survival of the fittest mechanism in nature to search for global optimal solutions. GA transforms the problem-solving process into a process similar to the crossover and mutation of chromosome genes in biological evolution, usually including basic operations such as selection, crossover, and mutation. However, a single heuristic algorithm has certain limitations. For example, GA has the defects of a weak local search ability, premature convergence, and an easy-to-fall-into local optimal solution when solving large-scale complex optimization problems. The genetic algorithm and simulated annealing algorithm are combined to accept the poor solution with a certain probability, so as to strengthen the ability to jump out of the local optimum. In addition, Dai et al. (2013) [33] proved the effectiveness of the genetic simulated annealing algorithm combining GA and SA. Therefore, the algorithm flow chart is shown in Figure 6, and the specific design idea is as follows:

4.1. Genetic Manipulation

4.1.1. Coding

The coding in this work is divided into three steps: the first step is to generate the first n columns of chromosomes, representing the number of tasks (0-n); in the second step, the tasks are randomly arranged. In the third step, if the number of couriers is m, m-1 separators are inserted into the corresponding tasks. For example, there are two distribution centers A and B, two couriers, and five tasks (tasks 1 and 3 belong to the distribution enter A, tasks 2, 4, and 5 belong to the distribution enter B). The chromosome code is 321645. After decoding this gene sequence, the delivery routes are A-3-B-2-1, B-4-5, indicating that the delivery route of courier 1 is distribution enter A-task 3-distribution enter B-task 2-task 1, and the delivery route of courier 2 is distribution enter B-task 4-task 5.

Initial population and fitness function is a set of path schemes are randomly generated by the above coding method. The smaller the total cost function is, the better the scheme is. Therefore, the reciprocal of the objective function is selected as the fitness function.

4.1.2. Selection

In this work, the roulette selection operation is used to randomly generate any real number between [0, 1] to determine the number of times each chromosome is selected.

4.1.3. Crossover

Using partial matching crossover, two positions are randomly selected, and the region between two points is defined as the cross-matching region, and the genes in the region are exchanged and matched. Different from the traditional crossover method, instead of directly exchanging the crossover segment of the chromosome, the chromosome that needs to be crossed is added to the first segment of the other chromosome, and then the repeated part of the other chromosome is removed one by one to obtain a new chromosome. In this work, the crossover coefficient is set to $\mathit{sin}(t/T_{max})$. The crossover coefficient increases first and then decreases with the number of iterations. At the beginning, the smaller crossover coefficient is maintained, and better individuals in the parent generation are retained. In the middle period, a larger crossover coefficient is maintained and the search range is expanded, and a smaller crossover speed is maintained in the later period. With this, we can find the best individual among the better individuals. A cross process demonstration is shown in Figure 7

4.1.4. Variation

A gene was randomly selected as the mutation region, and the gene in the mutation region was inverted and inserted into the chromosome to complete the mutation operation. The coefficient of variation was also taken as $\mathit{sin}(t/T_{max})$. The inverted insertion process is shown in Figure 8.

4.2. Simulated Annealing Operation

After a series of operations such as selection, crossover, and mutation of the genetic algorithm, the original population is formed into a new generation of population. At this time, the improved adaptive Metropolis criterion is introduced to correct the new generation of population in the following two cases to obtain the target population. The Metropolis criterion is often expressed as the following model:

$$p_z=\left\{\begin{array}{ccc}1& if& E(x_{after})<E(x_{before})\\ \exp (-{\displaystyle \frac{E(x_{after})-E(x_{before})}{kT}})& if& E(x_{after})\ge E(x_{before})\end{array}\right.$$

(18)

(1) If the energy value (function value) of the individual j in the population after selective crossover and mutation is less than the energy value (function value) of the individual i before the operation, the individual is retained unconditionally and all individuals are traversed.

(2) If the energy value (function value) of the new individual j is greater than the individual energy value (function value) before the operation, the individual ${ p}_z=\mathit{exp}[-{\displaystyle \frac{\Delta E}{kt}}]$ is selected according to the probability formula of the difference between the two states. If $p>random(0, 1)$, then the new individual j after genetic operation is selected, otherwise the original state is still maintained to select individual i, which is the Metropolis criterion, where k is the Boltzmann constant.

5. Experimental Results and Analysis

Throughout the entire delivery task, the crowdsourced courier strives to arrive and complete the pickup and delivery tasks within the time window specified by the customer. The designed model and algorithm need to determine the corresponding pickup and delivery order, optimize the delivery path, and minimize the cost of the crowdsourcing platform. This paper uses Matlab R2021a software to solve the delivery plan. The experimental parameters were set to three distribution centers A, B, and C, six crowdsourced couriers, and 47 crowdsourcing tasks. The crowdsourced courier had the same service time of 2 min per package at the distribution center, express station, and customer. Other model parameters are shown in

Table 4. Distribution center and couriers’ station information.

Serial Number	x (km)	y (km)	Time Window (min)
A	10	15	6:00–22:00
B	15	38	6:00–22:00
C	42	14	6:00–22:00
1	30	60	6:00–22:00
2	64	22	6:00–22:00
3	15	25	6:00–22:00

. The crossover probability $p_c$ and mutation probability $p_m$ of the experimental parameters are $\sin ({\displaystyle \frac{t\pi }{T_{\max }}})$, and the termination iteration number G is 500.

The information of distribution centers and courier stations is shown in

Table 5. Location information of crowdsourced couriers.

Serial Number	x (km)	y (km)
1	14	19
2	19	42
3	46	18
4	45	53
5	39	21
6	16	13

. There are a total of 3 distribution centers A, B, C, and 3 courier stations. In order to facilitate crowdsourcing delivery and meet customers’ pickup time needs, a large time window has been set for distribution centers and courier stations. The location and comprehensive service quality start information of the crowdsourced couriers are shown in

Table 6. Optimal delivery route.

Delivery Route
Courier 1: A->8->17->C->34->14->B->23->24-> Express station 2(40)->21->22
Courier 2: B->15->7->28->26->16
Courier 3: B->4->36->6-> Express station 1(39)-> Express station 1(46)->C-> Express station 2(43)-> Express station 2(41)
Courier 4: B-> Express station 1(38)-> Express station 1(47)->1->33->25->20
Courier 5: C->9-> Express station 3(44)-> Express station 3(45)->B-> Express station 3(42)->12->A->32->27->5->37->31->30->35->3->29
Courier 6: B->13->10->A->2->11->18->19

. The earliest start time for crowdsourced couriers is 9 a.m.

The delivery task information assigned to crowdsourcing express by the crowdsourcing platform is shown in

Table 7. Comparison between CCGA and STWGA.

	Experiments Number	Cost (CNY)	Number of Timeout Tasks			Average Delay Time (min)
			Efficient	Potential	Edge	Efficient	Potential	Edge
CCGA	1	542.80	0	0	9	0	0	61.86
	2	569.29	0	0	5	0	0	13.39
	3	557.61	0	0	9	0	0	43.03
	4	549.87	0	0	6	0	0	39.68
	5	584.63	0	0	6	0	0	46.92
STWGA	1	695.21	3	3	5	8.79	6.88	12.65
	2	590.71	3	2	0	4.29	4.55	0
	3	590.75	5	2	0	9.70	3.41	0
	4	549.77	4	2	0	15.69	1.30	0
	5	612.79	3	3	3	5.37	4.17	10.15

. Tasks 1–37 have a selection of home delivery, tasks 38–47 have pickup mode selected. Delivery tasks that require pickup will be delivered to the nearest delivery station based on the principle of proximity. Then, crowdsourcing tasks 38, 39, 46, and 47 are placed at courier station 1, crowdsourcing tasks 40, 41, and 43 are placed at courier station 2, and crowdsourcing tasks 42, 44, and 45 are placed at courier station 3. The customer categories are shown in Table 7. Category 1 represents efficient customers, category 2 represents potential customers, and category 3 represents marginal customers.

After 50 experiments, the optimal delivery plan based on customer classification was finally obtained, as shown in

Table 8. Comparison between CCGA and RPA.

Algorithm	Cost	Efficient Customer Delay Time	Potential Customer Delay Time	Edge Customer Delay Time
CCGA	572.24	0.00	0.00	31.28
RPA	6473.45	81.96	68.30	94.33

, and the delivery route map is shown in Figure 9. From the population evolution trend in Figure 10, it can be seen that the CCGA designed in this paper has convergence. In terms of convergence speed, around 400 generations have converged to the optimal value, and the population eventually tends to be excellent.

In order to verify the effectiveness of the algorithm, this study adopted two different optimization algorithms, CCGA and GA, and compared the solutions obtained from them. From Figure 11, it can be observed that using the algorithm proposed in this paper to continuously adjust and improve the path results in a lower total cost for the crowdsourcing logistics platform.

By comparing the iterative data of the two algorithms, as shown in Figure 12 and Figure 13, it can be seen that in the process of finding the optimal solution, the CCGA found the optimal solution earlier and broke through the local optimal solution problem of traditional genetic algorithms. The average optimal value of the CCGA is CNY 575.59, while the average optimal value of the GA is CNY 684.59. The CCGA has improved the average path convergence value by 15.92% compared to the GA algorithm. Therefore, the CCGA in this article has stronger optimization ability and improved algorithm efficiency

One can compare and analyze the CCGA designed in this article with the crowdsourcing platform that does not adopt a customer classification strategy, using the same time window delivery plan algorithm (STWGA) to verify the rationality and effectiveness of the model. In the STWGA, the penalty cost for all customers is calculated based on the penalty function for potential customers. The comparative analysis of the two schemes is shown in

Table A1. Symbol Explanation Table.

Symbol	Illustrate
h	Distribution center, $h\in H$, H $=\left\{1, 2, 3, \cdots , h\right\}$
u	Delivery station node, $u\in U$, U $=\left\{1, 2, 3, \cdots , u\right\}$
c	Initial location node for crowdsource couriers, $c\in C$, C $=\left\{1, 2, 3, \cdots , c\right\}$
n	Express crowdsourcing customer nodes, $n\in N$, N $=\left\{1, 2, 3, \cdots , n\right\}$
M	All node sets, M = $H\cup U\cup C\cup N$
k	Crowdsourced delivery vehicle serial number, $k\in K$, K $=\left\{1, 2, 3, \cdots , k\right\}$
$q_k$	Delivery vehicle load capacity
$w_n$	The weight of each customer’s crowdsourcing task for express delivery
$e_n$	The earliest service time requested by customer n
$l_n$	The latest service time requested by customer n
$d_{\mathit{nm}}$	The distance between node n and node m
${\rho }_k$	unit transport costs
$v_k$	The driving speed of vehicle k
$t_{\mathit{mk}}^1$	The time when vehicle k arrives at node m
$t_{\mathit{nk}}^w$	The waiting time of vehicle k at customer node n
$s_{\mathit{mk}}$	Service time of vehicle k at node m
$t_{\mathit{nm}}$	The time it takes for the delivery vehicle to travel from node n to node m
$t_{\mathit{mk}}^2$	The moment when vehicle k leaves node m
$f_n$	The delivery method chosen by the customer; if home delivery is chosen, then $f_n$ = 0; if pickup is chosen, then $f_n$ = 1
$x_{\mathit{nmk}}$	Variable of 0–1; if vehicle k travels from node n to node m, then $x_{\mathit{nmk}}$ = 1, otherwise it is 0
$y_{\mathit{nk}}$	Variable of 0–1; if vehicle k completes the delivery task of customer node n, then $y_{\mathit{nk}}$ = 1; otherwise, it is 0
$\alpha$	Customer’s early arrival penalty coefficient
$F_1$	Time penalty cost for efficient customers
$M_p$	Punishment cost for late arrival time of efficient customers
${\beta }_1$	Penalty coefficient for late arrival time cost of potential customers
$F_2$	Time penalty cost for potential customers
${\beta }_2$	Late arrival time cost for edge customers
$F_3$	Edge customer time penalty cost
$b_k$	Fixed cost per unit vehicle
$a_n$	If the customer category of the task belongs to efficient customers, then $a_n$ is 1; otherwise, it is 0
$b_n$	If the customer category of the task belongs to potential customers, then $b_n$ is 1; otherwise, it is 0
$c_n$	If the customer category of the task belongs to edge customers, then $c_n$ is 1; otherwise, it is 0
$R_1$	fixed cost
$R_2$	distance cost
$R_3$	Time penalty cost
R	Total cost of crowdsourcing platform

. It can be seen that the total cost of STWGA is lower than that of CCGA, and the number of overtime tasks for efficient and potential customers is less. In the five experiments in the table, the average total delay of efficient customers was 0 min. The delivery plan using STWGA resulted in an average total delay of 8.77 min for efficient customers, and an average total delay of 3.85 min for potential customers. The total cost of our algorithm was reduced by 7.73% compared to STWGA. Although occasionally the CCGA has a slightly higher cost than the STWGA, in the long run, CCGA is more conducive to retaining customers with significant contributions and promoting the development of enterprises.

We compare and analyze the CCGA with the crowdsourcing platform’s Random Path Delivery algorithm (RPA), conducting 50 experiments each. The comparative analysis of the two schemes is shown in

Table A2. Express delivery task information.

Serial Number	Coordinate (x, y)	Time Window (min)	Distribution Center	Weight (kg)	Customer Category
1	(41, 49)	9:35–13:35	B	1	1
2	(35, 17)	9:40–13:40	A	1	2
3	(55, 45)	9:25–13:25	B	4	3
4	(10, 43)	9:15–13:15	B	1	1
5	(55, 60)	9:00–13:00	A	2	3
6	(30, 60)	9:30–13:30	B	1	1
7	(20, 65)	9:50–13:50	B	1	2
8	(50, 35)	9:45–13:45	A	1	2
9	(30, 25)	9:30–13:30	C	4	2
10	(15, 10)	9:30–13:30	B	2	2
11	(30, 5)	9:45–13:45	B	1	3
12	(10, 20)	9:10–13:10	C	2	2
13	(50, 30)	9:15–13:15	B	1	2
14	(20, 40)	9:25–13:25	C	1	2
15	(15, 60)	9:40–13:40	B	2	2
16	(45, 65)	9:10–13:10	B	5	1
17	(45, 20)	9:15–13:15	C	1	2
18	(45, 10)	9:25–13:25	B	7	3
19	(55, 5)	9:50–13:50	A	1	2
20	(65, 35)	9:40–13:40	B	2	2
21	(65, 20)	9:30–13:30	B	1	3
22	(45, 30)	9:30–13:30	C	1	1
23	(35, 40)	9:45–13:45	B	3	3
24	(41, 37)	9:40–13:40	B	1	2
25	(64, 42)	9:40–13:40	B	1	3
26	(40, 60)	9:10–13:10	B	1	3
27	(31, 52)	9:25–13:26	A	1	3
28	(35, 69)	9:25–13:25	B	1	2
29	(53, 52)	9:20–13:20	A	3	1
30	(65, 55)	9:30–13:30	A	1	3
31	(63, 65)	9:40–13:40	C	1	3
32	(2, 60)	9:15–13:15	A	2	3
33	(45, 53)	9:30–13:30	B	1	1
34	(39, 21)	9:10–13:10	A	1	2
35	(59, 49)	9:05–13:05	B	3	3
36	(14, 47)	9:35–13:35	B	1	1
37	(59, 64)	9:45–13:45	A	1	3
38	(34, 64)	9:30–13:30	B	1	1
39	(24, 69)	9:45–13:45	B	2	2
40	(54, 39)	9:45–13:45	A	1	2
41	(34, 29)	9:25–13:25	C	4	2
42	(19, 14)	9:45–13:45	B	1	2
43	(34, 9)	9:45–13:45	B	1	3
44	(14, 24)	9:30–13:30	C	6	2
45	(9, 34)	9:05–13:05	B	1	2
46	(24, 44)	9:35–13:35	C	2	2
47	(19, 64)	9:35–13:35	B	1	2

. The CCGA in this article will greatly reduce the total cost of crowdsourcing platforms, and the delivery delay time for various customers will also be significantly improved. Therefore, the algorithm in this article is superior to random algorithms, and it also indicates that it is more effective in crowdsourcing delivery

6. Conclusions

This work studies the problem of express crowdsourcing task allocation considering distribution mode under customer classification. Based on the customer’s historical express delivery consumption behavior, the entropy weight RFM model is used to classify them into three categories: efficient customers, potential customers, and edge customers. Different time window penalty cost functions are set for different customers to improve the expected time window satisfaction rate of efficient customers, thereby improving overall customer satisfaction and maintaining the high-quality customer resources of distribution enterprises. At the same time, considering the two distribution methods of home delivery and pickup, the optimization model is established with the minimum total cost of the crowdsourcing logistics platform as the objective function. The experimental results show that the CCGA designed in this work can significantly reduce the total cost of the crowdsourcing platform compared with the traditional GA. The algorithm can quickly converge to the optimal value in the early stage, and performs stably in many experiments, which verifies the rationality of the model and the effectiveness of the algorithm. Compared with STWGA, CCGA hardly violates the time window of efficient customers. For potential customers, the number of crowdsourcing tasks violated is also lower, indicating that establishing different time penalty costs based on different customers is beneficial for improving delivery timeliness and reducing platform costs. Compared with RPA, CCGA has greatly improved the delivery cost and delivery timeliness, indicating that it is necessary to allocate the express crowdsourcing tasks and plan the delivery route. CCGA has a significant effect of shortening the distance, saving costs and improving timeliness.

In summary, this work can provide decision support for express crowdsourcing task problems, and has certain reference value for maintaining high-quality customer resources of enterprises, improving distribution efficiency and customer satisfaction. In actual delivery, it has improved the situation where crowdsource couriers arrived at the delivery location but could not contact customers, increased delivery efficiency, and provided customers with more diverse options for receiving goods, thereby increasing customer satisfaction. However, this work also has some limitations. For example, it only considers the minimum total cost of logistics crowdsourcing distribution platforms from an economic point of view, without considering emergencies in the transportation process, such as extreme weather, road congestion, etc. The model lacks certain dynamics, and further research can be carried out in the future.