Analysis of A Fresh Food E-commerce Enterprise’s Terminal Distribution Network Site Selection

. In view of the current prosperity of China’s fresh market transactions, fresh e-commerce enterprises in the end of the construction of the logistics network of the pain point problem, this study of fresh e-commerce in the “last mile" distribution of site planning issues. This paper focuses on a fresh food e-commerce enterprise’s end distribution network and explores the key inﬂuenc-ing factors in fresh food product distribution from three dimensions: timeliness, economy and service quality. And then establishes a mathematical model for selecting community stores for end distribution of fresh food e-commerce enterprises with the objectives of minimizing input costs and maximizing customer satisfaction. The model takes into account the costs of building and operating community stores and the transportation costs of distribution. In the model introduces three types of penalty costs due to site selection variability and transforms the model from a dual-objective to a single-objective model. The study takes the example of store location selection in 35 neighborhoods in region B by enterprise A and designs an immune optimization algorithm to solve the model. Based on the results of the algorithm, we verify the rationality and validity of the model, and obtain the optimal solution that ensures customer satisfaction and minimizes all costs.


Introduction
In recent years, China's fresh food market continues to expand, and a large number of fresh food e-commerce companies have emerged to meet the real demand for fresh food "at home". The construction of the end logistics network of fresh produce e-commerce companies directly affects the timeliness of delivery, determines the efficiency of product transportation and has a direct impact on customer satisfaction, making the construction of the end logistics network a pain point for many companies. In this paper, we analyze the site planning of fresh food in the "last mile" distribution from two perspectives: transportation cost and customer satisfaction of fresh food enterprises.
Many current Chinese scholars have conducted analysis and research on the location of centers for fresh produce distribution. Li Mengxun et al. established the dimension of freshness as a premise, and Cheng Xinfeng et al. considered the penalty cost, and then developed algorithms for example analysis [1,2]. Bing Su et al. and Xianfu Zhang and Yan Qi et al. established a model and related analytical solution based on customer time satisfaction [3][4][5].
In addition to considering the siting problem of centers for fresh produce distribution, a multi-objective model was used by Tienxia Zhang et al [6]. Shaoxun Chen et al. addressed the interrelationship between firm location cost and consumer preference by building a twolevel planning site selection model [7]. Jing Chen, Dong and Ming studied the quantity set distribution strategy based on the flat spoilage characteristics of fresh fruit and vegetable products [8]. Deng, Hui and Zhang, Zhibin combined the advantages of ant colony algorithm to propose an optimization scheme for fresh from end distribution path [9]. Wang, Dao-Ping et al.'s study customers choose fresh products will focus on price and quality, and customers will compete with each other [10].
In international studies, P. Amorim et al, K. Govindan et al, Bortolini M et al. have contributed to the research between distribution methods and product characteristics [11][12][13]. Byung Dug Song et al. focused on the study of customer satisfaction improvement [14][15][16]. Hsu et al. studied the vehicle path problem with time window for fresh produce delivery and constructed a model with the optimization objective of minimizing the total delivery cost [17]. Shukla faced the fresh produce cold chain logistics problem and considered the impact of distribution path on cost [18]. Rizet, Chopra et al. pointed out that delivery times are influenced by both the mode of transportation and the configuration of the distribution network [19].
According to the summary of previous studies, the influencing factors in the problem of end-of-pipe distribution network for fresh produce are shown in table 1.

Service quality
Location of community stores Convenience and safety of the entire purchase process The scale and facilities of the store Service quality of store staff Considering the impact of each factor on customer satisfaction and overall cost, it is found that the location of community stores, the path from logistics center to community stores, and the number of community stores will have a direct and important impact in the problem of fresh produce end distribution location analysis through literature analysis and collation.
The main factor of fresh produce delivery is the delivery time, and the siting model will consider the transportation cost of enterprises while fully ensuring the realization of customer satisfaction. Based on the key influencing factors, the model is finally transformed from a dual objective to a single objective for analysis and solution.
This paper contributes to finding the key influencing factors in fresh produce distribution from three dimensions: timeliness, economy and service quality, considering not only the costs of building and operating community stores for e-commerce enterprises and the transportation costs of distribution [1], but also focusing on the three types of penalty costs due to site selection variability, with the goal of minimizing input costs and maximizing customer satisfaction for fresh produce e-commerce enterprises. A mathematical model for siting com-munity stores for end-delivery of fresh produce e-commerce enterprises is established. This study enriches the research kernel of the fresh produce delivery site selection problem and provides references for considering more influencing factors in subsequent studies.
The Section 1 of this paper introduces relevant domestic and international studies, the Section 2 constructs the model, the Section 3 conducts data processing and analysis for the example, the Section 4 analyzes and summarizes the results of the example and draws conclusions, and points out the shortcomings and outlook for the study.

Model Analysis
The logistics network model established in this paper is used to solve the siting problem of the end distribution points of fresh produce. To study the site selection problem of a fresh produce distribution network consisting of logistics center -community stores -customers, the basic structure of the distribution network is shown in figure. 1:

Notation
Parameter Symbols K = {k|k = 1, 2, 3...m} : set of logistics centers. I = {i|1, 2, 3...n} : set of community stores, distribution ends. J = { j|1, 2, 3...o} :set of consumers. X ki : weight of goods from logistics center k to community store i per year, unit (kg). X i j : weight of goods picked up by customer j from community store i per year, unit (kg). d ki : distance from logistics center k to community store i unit (km). d i j : distance from community store i to consumer j, unit (km). V 1 : average vehicle speed from logistics center to community store, unit (km/min). V 2 : average speed from consumer to community store, unit (km/min). C ki : unit transportation cost of transportation vehicle from logistics center k to community store i,unit (yuan/kg.km). C i j : unit travel cost of customer j to community store i,unit (yuan/kg.km). C i : annual construction and operation cost of fresh e-commerce stores. B : total annual investment budget of fresh produce e-commerce, unit (yuan). θ : fresh product freshness sensitivity factor. https://doi.org/10.1051/e3sconf/202340902004 X k : the maximum handling capacity of logistics center k, unit (kg). X i : maximum handling capacity of community store i, unit (kg). D i : annual demand of customer j, unit (kg). β : proportion of consumer demand decline for each percentage point of fresh product freshness decline. c :unit price of fresh product, unit (yuan).  y i j = 0, customerjdon tpickupgoodsfromstorei; 1,customerjpickupgoodsfromstorei;

Modeling
(1) Objectives There are two goals in constructing the end distribution network model: one is to ensure customer satisfaction; the other is to ensure that the total cost paid by the fresh produce e-commerce company is kept at a minimum while meeting the necessary conditions.
Considering the real situation, this paper chooses the time satisfaction function that can better reflect the customer's tolerance level of waiting time under the time window to reflect customer satisfaction. The expression of time satisfaction function is as follows.
Then the customer satisfaction (TS) objective function is: 2.The lowest total cost of fresh produce e-commerce. The fresh produce end distribution network model is designed mainly to maximize the interests of the enterprise. In order to ensure that the economic benefits of the enterprise are as large as possible, in addition to ensuring customer satisfaction and expanding the market, the benefits can also be ensured by controlling investment costs and reducing expenses. The main costs that fresh produce e-commerce companies need to consider are the transportation costs from logistics centers to community stores, the operation and construction costs of community stores, and various penalty costs.
1)Transportation Costs k∈K i∈I 2)Construction and operation costs 3)Penalty Costs Penalty cost of the first part: the penalty cost of loss of freshness due to long transportation time during transportation from logistics center to community stores.
The second part of the penalty cost: the penalty cost incurred by the community store transporting to the consumer with reduced freshness.
The third part of the penalty cost: the cost of travel for consumers to move to the community store.
i∈I j∈J Therefore, the objective function with the lowest total cost is: 3. Conversion of dual to single objective. The terminal distribution network design model has two main objectives, the first objective is to maximize customer satisfaction, and the second objective is to ensure the lowest possible total cost for the e-commerce enterprise. The total cost minimization in the model includes the penalty cost, which is mainly due to customer satisfaction and is part of the embodiment of the first objective of the model to ensure maximum customer satisfaction. Therefore, the first objective function in the model is converted into a constraint and a basic value is set, which can also largely ensure the achievement of this objective. Then the dual-objective model can be converted to a single-objective model.
The conversion of customer satisfaction maximization is as follows. A customer service level parameter ∂ is set and customer satisfaction can be basically satisfied when T S = i∈I j∈J T i j * y i j /o > ∂.
(2)Constraint Analysis The number of community stores to be built cannot be more than the alternative number n, and greater than 1.
The total weight of goods dispatched by the logistics center does not exceed its maximum handling capacity, and the total weight of goods received by community stores and delivered to consumers. i∈I x ki ≤ x k , ∀k ∈ K.
Each community store only plays the role of fresh food transit, the total mass of goods transported in from the logistics center is the same as the total mass of goods transported to the customer by the store.
The demand of each consumer is known and a consumer can only go to one community store to pick up goods, so the quantity of goods transported to the customer by the community should be equal to the customer's demand.
Construction and operation costs should not be higher than the expected investment value.
Only when the construction of community store i is determined, the weight of goods from logistics center k to community store i is not zero, and similarly the quality of goods from community store i to delivery to customer j is not zero. At the same time, x i j to be meaningful, then customer j must pick up the goods from community store i.
x i j < M * y i j , ∀i ∈ I. i∈I The target value of customer satisfaction should be greater than the customer service level parameter.
To sum up, the final model can be obtained. In the final model, the above Eq. (8) is the final objective function, and the Eq. (9)-Eq. (20) are the constraints. In addition, two "0-1" variables need to be supplemented as new constraints.

Example Description
Based on the established site selection decision model and algorithmic process, the community store site selection of the terminal distribution network of realistic fresh food e-commerce enterprises is selected as an empirical case. According to the existing conditions and data entry of community coordinates and fresh food demand, assuming that the number and location of logistics centers have been determined, E of 35 cells is selected as the community store, and the remaining 35-E cells are used as consumption points, and the geographical location of cells is taken from Baidu map latitude and longitude query to pick up the coordinate system data.
All relevant parameters in the model in Section 2 are referred to the relevant literature [1], and the parameters are set as in table 2 The coordinates of fresh produce logistics center 1 and logistics center 2 are shown in table 3.

Algorithm Introduction
The community store location problem in this paper mainly uses the immune optimization algorithm, which is an emerging intelligent algorithm based on the theory of immunology. It is inspired by the biological immune system and uses the diversity generation and maintenance mechanism of the immune system to maintain the diversity of the population, which overcomes the problem of "early maturity" in the general optimization search process and allows the global optimal solution to be found eventually [10]. The immune algorithm uses a population search strategy, the same as the genetic algorithm, which emphasizes the exchange of information in the population. The structure of the algorithm goes through a cycle of "generating the initial population -calculating the evaluation criteria -exchanging individual information between populations -generating new populations" to have a higher probability of obtaining the global optimal solution in the end.

Solve the Algorithm and Analyze the Results
For the established model, the immune algorithm is firstly used to solve it. From the parameter setting, it is known that the annual construction cost of the logistics e-commerce plan is 2 million yuan, and the community stores that can be built by the logistics enterprise should be less than 7 under the restrictive condition of the total construction cost. And because the maximum handling capacity of each community store is limited, the condition is not satisfied when the number of stores is 3, so the number of community stores should be chosen to be greater than 3. After solving for E is 4, 5 and 6 respectively with the immune algorithm, the optimal site selection scheme under different conditions is derived.
(1) When the number of community stores is 6.
When setting the community store to 6, the solution yields the community store site selection diagram as shown in figure.

2:
The results showed that neighborhoods 3, 7, 11, 10, 9 and 8 were selected as community delivery stores, while the remaining neighborhoods were used as consumption points.
(2) When the number of community stores is 5. When setting the community store to 5, the solution yields the community store site selection diagram shown in figure. 3: The results showed that neighborhoods 23, 20, 3, 10, and 7 were selected as community delivery stores, and the rest of the neighborhoods were used as consumption points.
(3) When the number of community stores is 4.
When the community store is set to 4, the solution yields the community store site selection diagram shown in figure. 4: The results show that neighborhoods 34, 3, 7, and 8 are selected as community distribution stores, and the remaining neighborhoods are used as consumption points.
The maximum demand for each community store was calculated when the number of community stores was 6, 5 and 4, respectively, as shown in table 4 below.
The sum of the maximum demand for each community store in the three scenarios is less than the 650-ton limit for the community store and therefore satisfies the limit of the maximum handling capacity of the community store.
Consumer satisfaction was calculated by Public Notice table 2, based on the distance from each community to the community store when E is equal to 6, 5 or 4, respectively, the pickup time, and the satisfaction of each campus under the time window. The unit of distance (m)   T E=6 = 84.00% > ∂ = 75% Therefore, when 4, 5 and 6 community stores are selected, the solution results not only satisfy the limit of the maximum handling capacity of each community store, but also can satisfy the conditional limit of customer satisfaction, and the above three site options are feasible. the case of E is 6 not only has high construction cost, but also the satisfaction is lower than E is 5, and the low satisfaction will cause the penalty cost of this part of the community store to the district to increase. Assumption A, the variability of costs consumed from the logistics center to the store is minimal. Therefore, under assumption A, the lowest total cost option is selected by first not considering E is 6 as an alternative and only considering the options of E is 4 and E is 5 under the assumption that A holds. If A does not hold, the total cost of E is equal to 6 is calculated, and then the three options are compared again.

Total Costing
By calculating the cost of transportation from logistics center i to community store j denoted as C1, the penalty cost of reduced product freshness caused by this transportation process denoted as C2, the construction cost of the store denoted as C3, the travel cost from the community to the community store denoted as C4, and the penalty cost of reduced freshness of this part of the product denoted as C5. the total cost of the two scenarios with E is 5 and E is 4 can be obtained as follows table 5 and table 6.

Analysis of Algorithm Results
From the calculation results, it can be seen that the total annual transportation cost from the logistics center to the community store is $115,707 when building 5 community stores, and the total annual transportation cost from the logistics center to the community store is $114,9882 when building 4 logistics centers. Assumption A, the variability of costs consumed from the logistics center to the store is minimal. The difference is small, so the assumption A holds, so there is no need to calculate the option of building 6 community stores. Building 5 community stores costs 10.53% more than building 4 community stores, so the final choice is to build 4 community stores. Building 4 stores, the customer satisfaction of the time window is only about 80%, but it has reached the condition limit in the model. Therefore, the final optimal solution is to select 4 community stores: Store 3, Store 7, Store 8, and Store 34. logistics center 1 transports fresh products to Store 7 and Store 34; logistics center 2 transports products to Store 3 and Store 8.

Conclusions and Future Research
This paper first analyzes the problems of fresh food delivery in real life, and then identifies three main influencing factors in fresh food delivery. This paper constructs a mathematical model of community store location with the dual objectives of maximizing customer satisfaction and minimizing total cost. The site selection model introduces three penalty costs in addition to the basic construction and operation costs. Penalty cost of the first part: the penalty cost of loss of freshness due to long transportation time during transportation from logistics center to community stores. The second: the penalty cost incurred by the community store transporting to the consumer with reduced freshness. The third: the cost of travel for consumers to move to the community store. And comprehensive calculation of the total cost of fresh agricultural products e-commerce enterprises. By simplifying the model, the paper achieves the conversion work from dual objective to single objective.
The immune algorithm is designed for the initial site selection analysis of the logistics model, and the site selection model is solved for the arithmetic example, and the cost variability consumed from the logistics center to the store is verified to be small under all the constraints and comparing the alternatives. Finally, the best solution is found that achieves the lowest total cost while ensuring customer satisfaction.
The model is constructed based on a number of idealistic assumptions, without considering the individual characteristics of each type of fresh produce, the complexity of road conditions, and the variability between stores in different communities. Based on customer satisfaction, e-commerce companies should also consider the total cost not just the various costs spent on the books, but also the various penalty costs that occur after construction, and only by fully considering the realistic needs can fresh produce logistics develop and provide better services.