Store replenishment decision method and system based on data analysis

Abstract

The application relates to the technical field of supply chain management, in particular to a store replenishment decision method and system based on data analysis. The method comprises the following steps: acquiring multiple dimension data of each commodity in each store at each time in multiple historical supply periods to construct a data matrix of commodities and stores; adopting a collaborative clustering algorithm to cluster the data matrix to obtain multiple commodity clusters and multiple store clusters; adopting a global optimization algorithm to globally optimize an initial global distribution matrix constructed in advance to obtain an optimized global distribution matrix; and obtaining a constrained total supply amount; based on the constrained total supply amount, randomly generating an initial population, locally optimizing the initial population, and obtaining an optimal replenishment scheme of any commodity to different stores and an optimal replenishment scheme set. The application can efficiently obtain a commodity replenishment scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of supply chain management. More specifically, the present invention relates to a store replenishment decision-making method and system based on data analysis. Background Art

[0002] Store replenishment is a key operational link in modern chain retail networks, and the accuracy and timeliness of its decision-making are directly related to the sales performance and profitability of the entire enterprise. However, due to complex factors such as market demand fluctuations, promotional activities, and seasonal changes, store replenishment decision-making inevitably faces challenges. After a replenishment decision-making error, the ability to quickly adjust and optimize the inventory structure is of crucial significance for increasing sales, reducing inventory costs, reducing goods losses, and improving customer satisfaction.

[0003] Existing store replenishment decision-making methods mainly include the model method.

[0004] The model method calculates the replenishment quantity by establishing a mathematical model and theoretically is not affected by subjective factors, having higher decision-making potential. Among them, the model method based on global optimization algorithms, such as genetic algorithms, aims to find the global optimal replenishment plan that minimizes the total supply chain cost.

[0005] For example, the Chinese patent with the publication number CN119761968B discloses a method applied to warehouse replenishment planning, including: constructing a warehouse replenishment and storage location collaboration model, where the goal of the warehouse replenishment and storage location collaboration model is to minimize the total energy consumption of warehouse transportation equipment and the load of warehouse channels; solving the warehouse replenishment and storage location collaboration model based on an improved non-dominated sorting elite genetic algorithm to obtain the optimal population solution and the corresponding lowest energy consumption value and channel usage difference coefficient. The above method utilizes the genetic algorithm model.

[0006] However, when the number of stores and products is huge, as the number of stores and products increases, the efficiency of general genetic algorithms is extremely low, and it is often difficult to obtain the optimal solution within the limited business decision-making time. Summary of the Invention

[0007] In order to solve the problem of low decision-making efficiency caused by the increase in the number of stores and products, the present invention provides solutions in the following two aspects.

[0008] In a first aspect, the present invention provides a data analysis-based store replenishment decision-making method, comprising: acquiring multi-dimensional data of each product at each time point within multiple historical supply cycles in each store to construct a product-store data matrix; clustering the data matrix using a collaborative clustering algorithm to obtain multiple product clusters and multiple store clusters; globally optimizing a pre-constructed initial global allocation matrix using a global optimization algorithm to obtain an optimized global allocation matrix, wherein the global allocation matrix includes the replenishment ratio allocated from each product cluster to each store cluster; and the fitness function and the correlation index during global optimization. And the replenishment ratio at each iteration is positively correlated; the correlation index is the ratio of the mean of the overlapping elements of any product cluster and any store cluster in the data matrix to the mean of all elements in the data matrix; the estimated total replenishment quantity of any product to all stores is multiplied by the replenishment ratio of its product cluster and the corresponding store cluster as the constrained total supply quantity; based on the constrained total supply quantity, an initial population is randomly generated, and a local optimization algorithm is used to locally optimize the initial population to obtain the optimal replenishment scheme for any product to replenish different stores, and to obtain the set of optimal replenishment schemes for all products to replenish store clusters.

[0009] Furthermore, the multi-dimensional data of each product in each store includes total sales, average supply quantity, and average number of days in stock; each element of the data matrix is ​​the weighted sum of the multi-dimensional data of the product and the store after standardization processing.

[0010] Furthermore, the fitness function is also positively correlated with the historical importance weight, which is the total sales volume of any product cluster within a set historical time period divided by the total sales volume of all products within the set historical time period.

[0011] Furthermore, the fitness function is a maximization objective function, which includes the sum of the products of each replenishment ratio in the global allocation matrix, the historical importance weight of the corresponding product cluster, and the correlation indicators between the corresponding product cluster and the store cluster.

[0012] Furthermore, the maximization objective function is also positively correlated with the transportation efficiency index between the corresponding commodity cluster and store cluster. The transportation efficiency index is the reciprocal of the ratio of the average transportation distance from the central warehouse of the commodity cluster to each store in the store cluster to the maximum transportation distance from all warehouses to all stores plus 1.

[0013] Furthermore, the global optimization process includes: randomly obtaining an initial global allocation matrix, concatenating all elements of the initial global allocation matrix into a one-dimensional real-number encoded chromosome to generate an initial population containing multiple chromosomes; constructing a fitness function; and performing selection, crossover, and mutation operations based on the fitness function until the termination condition is met to obtain the optimized global allocation matrix.

[0014] Furthermore, the estimated total replenishment quantity is obtained from the planned total replenishment quantity of all goods that should be supplied to all stores during the current supply cycle.

[0015] Furthermore, the fitness function value of the local optimization algorithm is positively correlated with the product of the correlation index of the corresponding product cluster and store cluster and the replenishment quantity, and negatively correlated with the absolute value of the difference between the replenished inventory and the maximum inventory.

[0016] Furthermore, both the global optimization algorithm and the local optimization algorithm are genetic algorithms.

[0017] In a second aspect, the present invention provides a data analysis-based store replenishment decision system, comprising: a processor; and a memory storing computer instructions for data analysis-based store replenishment decisions, wherein when the computer instructions are executed by the processor, the system performs the aforementioned data analysis-based store replenishment decision method.

[0018] The beneficial effects of this invention are as follows: The solution of this invention groups massive amounts of goods and stores into groups using a collaborative clustering algorithm, significantly reducing computational complexity. Then, a "global-first, local-later" optimization strategy is adopted, first allocating macro-level resources between clusters, and then performing refined replenishment quantity calculations within each cluster. Compared to existing technologies, this method greatly improves decision-making efficiency and accuracy, and can quickly generate optimal replenishment plans that balance sales, inventory, and logistics costs.

[0019] By fusing multi-dimensional data to obtain a data matrix, calculations can be simplified.

[0020] This invention determines the weight by using the ratio of the total sales volume of a product cluster to the total sales volume of all products, so that replenishment resources can be more rationally tilted towards products with high contribution.

[0021] The aforementioned maximization objective function ensures that product clusters with good historical sales performance and high relevance to the target store receive higher allocation priority, making the optimization results closer to the goal of maximizing actual business value.

[0022] When determining the specific replenishment quantity of goods, it is possible to find the best balance between meeting market demand and avoiding inventory backlog or stockouts, thereby achieving refined control over the inventory level of individual goods and improving inventory turnover rate. Attached Figure Description

[0023] Figure 1 This illustration schematically shows a flowchart of the steps of the data analysis-based store replenishment decision-making method in this embodiment; Figure 2 The diagram illustrates the structure of the data analysis-based store replenishment decision system in this embodiment. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0025] like Figure 1 As shown, the data analysis-based store replenishment decision-making method in this embodiment includes the following steps: Step S1: Obtain product data for different products across multiple historical supply cycles in different stores.

[0026] In this embodiment, based on the company's internal sales and supply data, product data for different products in different stores over multiple historical supply cycles is obtained; the product data consists of multiple dimensions, including total sales, average supply volume, and average number of days in stock.

[0027] The aforementioned historical supply cycle can be one month or one quarter, etc.; the number of historical supply cycles can be 10.

[0028] In this embodiment, after acquiring product data for different products in different stores, a data matrix of products and stores is constructed. The multi-dimensional data of each product in each store includes total sales, average supply quantity, and average number of days in stock; each element of the data matrix is ​​the weighted sum of the multi-dimensional data of products and stores after standardization.

[0029] In the data matrix of products and stores, the rows are product numbers and the columns are store numbers. The values ​​are a comprehensive interaction strength index used to quantify the degree of business association between a specific product and a specific store.

[0030] The interaction strength index is calculated by weighting and summing the standardized data from different dimensions of the product data. For example, it can be used to weight and sum the three core business indicators of sales performance, supply stability, and inventory fulfillment rate. This provides a more comprehensive and accurate picture of the product's overall performance in the store, offering a more solid and reliable data foundation for subsequent clustering and optimization, thereby improving the accuracy of the final decision.

[0031] The standardization process in this embodiment can employ the Z-score method or the maximum / minimum value normalization method.

[0032] It should be noted that the weights in the above weighted summation can be adjusted according to the business focus. For example, if more emphasis is placed on sales performance, the weight of total sales volume can be set higher. For instance, the weight of total sales volume can be 0.5, and the weights of average supply quantity and average days in stock can be 0.25 respectively.

[0033] Step S2: The data matrix is ​​clustered using a collaborative clustering algorithm to obtain the clustered product clusters, store clusters, and correlation indicators between product clusters and store clusters.

[0034] Co-clustering algorithm is an ensemble learning method that combines the concepts of clustering and classification. It aims to perform clustering and classification tasks simultaneously by assigning data samples to different clusters and performing classification within each cluster, thus achieving collaborative clustering and classification. The specific steps are as follows: first, a clustering step is performed to divide the data samples into different clusters; then, within each cluster, a classification step is performed on the samples in that cluster.

[0035] It should be noted that the collaborative clustering algorithm used in this embodiment mainly uses the clustering step and does not focus on the classification step.

[0036] In this embodiment, the process of obtaining the related indicators is as follows: First, the data matrix is ​​processed using a co-clustering algorithm, which simultaneously performs clustering operations on the rows (product sets) and columns (store sets) of the matrix to obtain m product clusters and n store clusters.

[0037] Each of the m product clusters has a set of products with similar sales patterns or attributes; each of the n store clusters has a set of stores with similar customer groups or geographical locations.

[0038] Secondly, obtain multiple cross-sub-data matrices for m product clusters and n store clusters, and calculate the ratio of the average of all elements in each cross-sub-data matrix to the average of all elements in the data matrix, which serves as the correlation index A for the corresponding cross-sub-data matrix.

[0039] Among them, the correlation index characterizes the correlation strength between product clusters and store clusters; when the correlation index of the cross sub-data matrix corresponding to product cluster i and store cluster j is... When the value is greater than 1, it indicates that the correlation between product cluster i and store cluster j is higher than the average level, which is a business model that needs to be closely monitored.

[0040] The aforementioned cross-data sub-matrix is ​​a part of the data matrix, representing the intersection of the product cluster and the store cluster within the data matrix. Therefore, the ratio of the average of all elements in each cross-data sub-matrix to the average of all elements in the data matrix is ​​calculated as the correlation index A for the corresponding cross-data sub-matrix. In other words, the correlation index is the ratio of the average of the overlapping elements of any product cluster and any store cluster in the data matrix to the average of all elements in the data matrix.

[0041] Furthermore, the correlation index A can be used to construct an m-row n-column correlation matrix, where the rows represent the number of product clusters, the columns represent the number of store clusters, and the elements in the corresponding rows and columns represent the correlation index between the product clusters and the store clusters.

[0042] Step S3: Construct an initial global allocation matrix and use a global optimization algorithm to optimize the initial global allocation matrix to obtain an optimized global allocation matrix.

[0043] In this embodiment, both the global optimization algorithm and the local optimization algorithm are genetic algorithms.

[0044] In this embodiment, each element in the initial global allocation matrix represents the proportion of the total replenishment quantity of the i-th product cluster allocated to the j-th store cluster.

[0045] To ensure the integrity of the allocation, each row of the initial global allocation matrix must satisfy the constraint that the sum of all elements in each row must be 1, that is, every element in the matrix must be equal to 1. It is a percentage value between 0 and 1.

[0046] It should be noted that the optimization objective in this embodiment is to find an optimal global allocation matrix P with m rows and n columns.

[0047] Specifically, the process of using a genetic algorithm to globally optimize the initial global allocation matrix to obtain the optimized global allocation matrix is ​​as follows: a. Initialization: Randomly generate an initial population consisting of multiple chromosomes.

[0048] In this embodiment, all elements in the initial global allocation matrix P are concatenated into a one-dimensional vector, which serves as a complete real-number encoded chromosome.

[0049] Each chromosome is normalized during generation to ensure that the decoded matrix satisfies the constraint that the sum of its rows is 1.

[0050] b. Construction of fitness function: For each chromosome in the population, construct a fitness function to calculate its fitness value.

[0051] The fitness function in this embodiment for: ; in, Let represent the replenishment ratio to be optimized for product cluster i and store cluster j, where represents the percentage of replenishment quantity for product cluster i allocated to store cluster j. Let m be the historical importance weight of the i-th product cluster, m be the total number of product clusters, and n be the total number of store clusters. This is the correlation index between product cluster i and store cluster j. During the optimization process, it is used to guide the algorithm to prioritize exploring highly correlated replenishment patterns in historical data.

[0052] in, The calculation method is as follows: the total sales volume of product cluster i in the historical set time period is divided by the total sales volume of all products in the historical set time period.

[0053] This historical importance weight is introduced to ensure that the algorithm prioritizes more important product clusters in historical data when allocating resources.

[0054] The historical time period includes multiple historical supply cycles. For example, the number of multiple historical supply cycles is less than or equal to 10.

[0055] The fitness function described above represents the maximum expected overall system efficiency during each iteration of the iterative process.

[0056] In another embodiment, transportation efficiency can also be considered, that is, transportation efficiency is introduced into the fitness function. In this case, the fitness function... for: ; in, Let be the transportation efficiency index for product cluster i and store cluster j.

[0057] Among them, transportation efficiency indicators It is the reciprocal of the sum of the ratio of the average transportation distance from the central warehouse of commodity cluster i to each store in store cluster j and the maximum transportation distance from all warehouses to stores, and 1.

[0058] To reduce computational complexity, a central warehouse concept is used. This means that even if multiple goods within a product cluster i are scattered across different warehouses, the central warehouse for product cluster i can be chosen by selecting the geographically central point or designating the core warehouse with the highest shipment volume. The transportation efficiency index has a range of (0, 1], with values ​​closer to 1 indicating higher transportation efficiency.

[0059] c. Through standard genetic algorithm operations such as selection, crossover, and mutation, the population is iteratively evolved until the iteration stops, and an optimized global allocation matrix is ​​obtained.

[0060] The iteration stopping condition mentioned above is reaching the maximum number of iterations or convergence. Convergence specifically means that, in T consecutive iterations (e.g., T=3), the norm of the variation amplitude of the elements of the global allocation matrix (e.g., the Frobenius norm) is less than a preset minimum threshold. (For example, 10 to the power of negative fourth). This indicates that the allocation strategy has stabilized.

[0061] In this embodiment, the chromosome with the highest fitness value during the entire iteration process is selected and decoded into an m x n matrix, which is used as the optimized global assignment matrix.

[0062] The fitness function constructed above comprehensively considers replenishment ratio, importance of product clusters, transportation efficiency, and business correlation strength.

[0063] The aforementioned global optimization algorithm focuses on using data from historical supply cycles to obtain the allocation of each product to stores, providing data support for the allocation of products to stores in the current supply cycle.

[0064] Step S4: In this embodiment, after obtaining the optimized global allocation matrix, the constrained total supply of any product in any store cluster is obtained based on the replenishment ratio in the global allocation matrix and the estimated total replenishment quantity of any product to all stores.

[0065] Specifically, the process for obtaining the total supply constraint is as follows: Based on the planned total replenishment quantity of all goods to be supplied to all stores in the current supply cycle, the estimated total replenishment quantity of any product to be supplied to all stores is obtained; the product of the estimated total replenishment quantity of any product to be supplied to all stores and the replenishment ratio of its product cluster and the current store cluster is used as the constrained total supply quantity.

[0066] The planned total replenishment quantity mentioned above is the maximum total replenishment quantity of all goods that can be provided within the current supply cycle, and it is known.

[0067] Furthermore, based on the total supply constraint, an initial population is randomly generated, and the initial population is locally optimized to obtain the optimal replenishment scheme for replenishing any product to different stores, and to obtain the set of optimal replenishment schemes for replenishing all products to the store cluster.

[0068] In this embodiment, based on the optimized global allocation matrix, the global optimization problem is decomposed into n parallel local optimization sub-problems for a single store cluster. That is, the allocation scheme of the optimized global allocation matrix is ​​refined into executable replenishment orders for specific products to specific stores.

[0069] Specifically, based on the optimized global allocation matrix and the planned total replenishment quantity of all goods to be supplied to all stores in the current supply cycle, a genetic algorithm is used to perform local optimization for each store cluster to obtain the optimal replenishment scheme set for each store cluster, including: First, initialize the population.

[0070] In this embodiment, after obtaining the total supply quantity of any product to replenish any store cluster, the initial allocation quantity of any product to replenish different stores in any store cluster is randomly generated.

[0071] Secondly, the fitness function values ​​for each individual local optimization subproblem are constructed.

[0072] Wherein, fitness function for: ; in, For the degree of satisfaction of needs, This is the inventory deviation item.

[0073] The requirement satisfaction item is: ; in, Let k be the maximum inventory level of product k in store l. The estimated replenishment quantity for product k to store l. This is the correlation indicator between product cluster i (the cluster to which product k belongs) and store cluster j (the cluster to which store l belongs).

[0074] The inventory deviation item represents the degree of deviation between the replenished inventory and the target inventory. Specifically, the calculation formula is as follows: ; This represents the current inventory level.

[0075] The value is in the range [0, 1], and the smaller the value, the more reasonable the inventory allocation.

[0076] In another embodiment, the fitness function value also takes into account the transportation load term, that is, the balance of transportation resource utilization is incorporated into the fitness function value.

[0077] Specifically, the corresponding fitness function for: ; in, This refers to the transportation load item when replenishing product k to store l.

[0078] In the above formula, , , These are hyperparameters, representing the degree of importance placed on demand fulfillment, inventory deviation, and transportation load, respectively. As a preferred option, α1=0.6, β1=0.2, and γ1=0.2 can be set to prioritize sales opportunities.

[0079] Among them, the transportation load item for: ;in, The ratio of the estimated replenishment quantity of product k to store l to the maximum inventory of product k in store l.

[0080] Among them, due to An efficiency index is a value between (0, 1] (the closer to 1, the higher the efficiency). This can be considered a unit transportation cost factor. The lower the transportation efficiency, the larger this factor, which in turn leads to a higher value for the transportation load factor penalty.

[0081] It should be noted that the fitness function for introducing the transportation load term mentioned above must correspond to the fitness function for introducing transportation efficiency in the initial global optimization phase.

[0082] Then, multiple iterations are performed until the iteration stops, to obtain the optimal replenishment quantity plan for each product in each store cluster, and thus obtain the set of optimal replenishment plans for each store cluster.

[0083] In this embodiment, after the independent optimization process for each store cluster terminates, an optimal replenishment plan set is output. This optimal replenishment plan set defines the set of specific replenishment quantities for each product k from each store cluster to each store l.

[0084] It should be noted that the above iterative optimization from global to local can be regarded as a complete iterative cycle, and the total number of iterations can be limited to a preset maximum number of iterations, Nmax (e.g., 50 times). At the end of each iterative cycle, the system judges the above two criteria. When either criterion is satisfied, the system terminates the iteration process.

[0085] The solution of this invention optimizes the replenishment plan between goods and stores by first optimizing the global plan and then optimizing the local plan, which can obtain an accurate final replenishment decision result.

[0086] This invention also provides a store replenishment decision-making system based on data analysis. For example... Figure 2As shown, the system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the data analysis-based store replenishment decision-making method according to the present invention.

[0087] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0088] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), etc. While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A data-driven store replenishment decision-making method, characterized in that, include: To acquire multi-dimensional data on each product at each time point within multiple historical supply cycles in each store, in order to construct a data matrix of products and stores; A collaborative clustering algorithm is used to cluster the data matrix, and the clustered product clusters, store clusters, and correlation indicators between product clusters and store clusters are obtained. The correlation indicator is the ratio of the mean of the overlapping elements of any product cluster and any store cluster in the data matrix to the mean of all elements in the data matrix. A global optimization algorithm is used to globally optimize the pre-constructed initial global allocation matrix to obtain an optimized global allocation matrix. The global allocation matrix includes the replenishment ratio of each product cluster to each store cluster. The fitness function during global optimization is positively correlated with the correlation index and the replenishment ratio at each iteration. The total supply of any product is determined by multiplying the estimated total replenishment quantity of all stores by the replenishment ratio of its product cluster to the corresponding store cluster. Based on the total supply constraint, an initial population is randomly generated. A local optimization algorithm is used to locally optimize the initial population to obtain the optimal replenishment scheme for any product to replenish different stores, and to obtain the set of optimal replenishment schemes for all products to replenish the store cluster.

2. The store replenishment decision-making method based on data analysis according to claim 1, characterized in that, The data for each product in each store includes total sales, average supply, and average days in stock; each element of the data matrix is ​​a weighted sum of the standardized data for the product and the store.

3. The store replenishment decision-making method based on data analysis according to claim 1, characterized in that, The fitness function is also positively correlated with the historical importance weight, which is the total sales volume of any product cluster within a set historical time period divided by the total sales volume of all products within the set historical time period.

4. The store replenishment decision-making method based on data analysis according to claim 3, characterized in that, The fitness function is a maximization objective function, which includes the sum of the products of each replenishment ratio in the global allocation matrix, the historical importance weight of the corresponding product cluster, and the correlation indicators between the corresponding product cluster and the store cluster.

5. The store replenishment decision-making method based on data analysis according to claim 4, characterized in that, The maximization objective function is also positively correlated with the transportation efficiency index between the corresponding commodity cluster and store cluster. The transportation efficiency index is the reciprocal of the ratio of the average transportation distance from the central warehouse of the commodity cluster to each store in the store cluster to the maximum transportation distance from all warehouses to all stores plus 1.

6. The store replenishment decision-making method based on data analysis according to claim 5, characterized in that, The global optimization process includes: Randomly obtain an initial global allocation matrix and concatenate all elements of the initial global allocation matrix to form a one-dimensional real-number encoded chromosome, generating an initial population containing multiple chromosomes; Construct the fitness function; Based on the fitness function, selection, crossover, and mutation operations are performed until the termination condition is met, resulting in an optimized global allocation matrix.

7. The store replenishment decision-making method based on data analysis according to claim 1, characterized in that, The estimated total replenishment quantity is obtained from the planned total replenishment quantity of all goods that should be supplied to all stores in the current supply cycle.

8. The store replenishment decision-making method based on data analysis according to claim 1, characterized in that, The fitness function value of the local optimization algorithm is positively correlated with the product of the correlation index of the corresponding product cluster and store cluster and the replenishment quantity, and negatively correlated with the absolute value of the difference between the replenished inventory and the maximum inventory.

9. The store replenishment decision-making method based on data analysis according to claim 8, characterized in that, Both the global optimization algorithm and the local optimization algorithm are genetic algorithms.

10. A store replenishment decision-making system based on data analysis, characterized in that: include: processor; A memory storing computer instructions for data-driven store replenishment decisions, which, when executed by the processor, cause the system to perform the data-driven store replenishment decision method according to any one of claims 1-9.

Images