Supply chain inventory management optimization method based on machine learning

Abstract

The invention discloses a supply chain inventory management optimization method based on machine learning, and relates to the technical field of inventory management, and the method comprises the steps: carrying out the node disassembly of a target supply chain, obtaining N supply chain nodes, carrying out the associated data collection of the supply chain nodes, and obtaining N node product data sets; performing association index extraction and correlation screening to obtain N node key analysis index sets, and then performing multi-dimensional feature analysis on the N node product data sets to obtain N node supply chain feature sets; constructing a machine learning algorithm list to perform matching analysis and integrated prediction on the supply chain feature sets of the N nodes to obtain product inventory demand prediction information; and constructing an inventory strategy optimization space, carrying out inventory adjustment analysis on the product inventory demand prediction information, and outputting inventory optimization regulation and control parameters to carry out supply chain inventory management. The technical problem of inaccurate inventory management in the prior art is solved, and the technical effect of improving the accuracy of inventory management and the efficiency of a supply chain is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of inventory management, and in particular to a supply chain inventory management optimization method based on machine learning. Background Art

[0002] With the rapid development of globalization and e-commerce, supply chain management has become increasingly important in all walks of life, especially in industries such as retail, manufacturing and logistics. The efficiency of inventory management directly affects the operating costs and market competitiveness of enterprises. However, with the expansion of the supply chain scale and the continuous changes in market demand, traditional inventory management methods are facing increasing challenges. Existing inventory management systems usually rely on historical data and experience, lack the ability to dynamically adjust to real-time demand changes and various links in the supply chain, resulting in inaccurate inventory forecasts, frequent overstocking or out-of-stock problems, and cannot meet the needs of modern enterprises for efficient, accurate and flexible inventory management. Summary of the invention

[0003] This application provides a supply chain inventory management optimization method based on machine learning, which is used to solve the technical problem of inaccurate inventory management in the prior art.

[0004] In view of the above problems, this application provides a supply chain inventory management optimization method based on machine learning.

[0005] The present application provides a supply chain inventory management optimization method based on machine learning, the method comprising: Decompose the target supply chain into nodes to obtain N supply chain nodes, and collect related data for the N supply chain nodes to obtain N node product data sets; extract related indicators and perform correlation screening on the N supply chain nodes to obtain N node key analysis indicator sets; perform multidimensional feature analysis on the N node product data sets based on the N node key analysis indicator sets to obtain N node supply chain feature sets; construct a machine learning algorithm list, and perform matching analysis and integrated prediction on the N node supply chain feature sets based on the machine learning algorithm list to obtain product inventory demand forecast information; construct an inventory strategy optimization space based on the N supply chain nodes, use the inventory strategy optimization space to perform inventory adjustment analysis on the product inventory demand forecast information, output inventory optimization control parameters, and perform supply chain inventory management based on the inventory optimization control parameters.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The present application disassembles the target supply chain nodes to obtain N supply chain nodes, and collects related data for the N supply chain nodes to obtain N node product data sets; extracts related indicators and screens correlation for the N supply chain nodes respectively to obtain N node key analysis indicator sets; performs multidimensional feature analysis on the N node product data sets based on the N node key analysis indicator sets to obtain N node supply chain feature sets; constructs a machine learning algorithm list, performs matching analysis and integrated prediction on the N node supply chain feature sets based on the machine learning algorithm list, and obtains product inventory demand forecast information; constructs an inventory strategy optimization space according to the N supply chain nodes, uses the inventory strategy optimization space to perform inventory adjustment analysis on the product inventory demand forecast information, outputs inventory optimization control parameters, and performs supply chain inventory management based on the inventory optimization control parameters. The present invention solves the technical problem of inaccurate inventory management in the prior art, extracts key analysis indicators and feature sets of each node by disassembling and collecting data for the target supply chain nodes, combines related indicator extraction, correlation screening, feature analysis and other technologies, uses the machine learning algorithm list to perform matching analysis and integrated prediction, and obtains product inventory demand forecast information. Based on the prediction results, the inventory strategy optimization space is constructed, inventory adjustment analysis is performed and optimization control parameters are output, so as to realize inventory optimization management and achieve the technical effect of improving the accuracy of inventory management and supply chain efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic diagram of a supply chain inventory management optimization method based on machine learning provided in an embodiment of the present application; Figure 2 A schematic diagram of a process for obtaining a set of key analysis indicators of N nodes in a supply chain inventory management optimization method based on machine learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0009] This application provides a supply chain inventory management optimization method based on machine learning, which is used to solve the technical problem of inaccurate inventory management in the existing technology. By disassembling and collecting data from the target supply chain nodes, combined with related indicator extraction, correlation screening, feature analysis and other technologies, the key analysis indicators and feature sets of each node are extracted, and matching analysis and integrated prediction are performed using a list of machine learning algorithms to obtain product inventory demand forecast information. Based on the prediction results, an inventory strategy optimization space is constructed, inventory adjustment analysis is performed, and optimization control parameters are output, thereby realizing inventory optimization management and achieving the technical effect of improving the accuracy of inventory management and supply chain efficiency.

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0011] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0012] Examples, such as Figure 1 As shown, the present application provides a supply chain inventory management optimization method based on machine learning, the method comprising: Step S100: Decompose the target supply chain into nodes to obtain N supply chain nodes, and collect related data for the N supply chain nodes to obtain N node product data sets.

[0013] In the embodiment of the present application, the target supply chain is first disassembled into nodes. This process is carried out by dividing the functional units, and the specific steps include identifying the main links of the supply chain, such as raw material procurement, production and manufacturing, warehouse management, logistics and transportation, and terminal sales. In the disassembly process, each link is independently defined as a supply chain node. These nodes have dependencies or data interactions with each other, forming the basic unit of the entire supply chain network. Through this disassembly process, the global structure of the supply chain can be obtained, and then the N supply chain nodes in the supply chain can be determined.

[0014] Then, the associated data of the N disassembled supply chain nodes are collected. This process directly obtains the original data related to the node through the existing sensor equipment and information management system. For example, at the warehouse management node, information such as inventory quantity, warehouse entry records, and warehouse exit records are directly collected; at the manufacturing node, data such as raw material usage, production plan, and production completion status are collected; at the logistics distribution node, data such as transport vehicle status, cargo transportation records, and delivery status are collected. These data are recorded and uploaded in real time through sensors, automation equipment, or enterprise resource planning (ERP) systems.

[0015] Through the above steps, each node in the supply chain obtains its associated original data, forming N node product data sets.

[0016] Step S200: extracting related indicators and screening correlations for the N supply chain nodes respectively to obtain a set of key analysis indicators for N nodes.

[0017] In an embodiment of the present application, historical product data of N supply chain nodes are first mined to obtain supply chain data sets of each node; then, related indicators closely related to node operation are extracted from these data sets to form a supply chain related indicator set; then, similar or highly correlated indicators are classified into related indicator clusters through cluster analysis; finally, these related indicator clusters are screened for correlation to select key indicators that have a significant impact on inventory optimization, and finally a key analysis indicator set of N nodes is formed.

[0018] Further, such as Figure 2 As shown, in the method provided in the embodiment of the application, the obtaining of the N node key analysis indicator sets also includes: Based on the N supply chain nodes, historical product data mining is performed to obtain N node supply chain data sets; related indicators are extracted from the N node supply chain data sets to obtain N node supply chain related indicator sets; cluster analysis is performed on the N node supply chain related indicator sets to generate N node related indicator clusters; correlation screening is performed on the N node related indicator clusters to obtain N node key analysis indicator sets.

[0019] In an embodiment of the present application, historical product data mining is first performed based on N supply chain nodes. This process obtains the historical behavior data of each node by comprehensively mining the historical product data of N supply chain nodes. The data not only includes inventory information, but also involves multi-dimensional data such as sales records, order requirements, production plans, and transportation status. These data come from multiple business systems, such as ERP systems, warehouse management systems, and transportation management systems. Using database query and ETL tools, data is extracted and integrated from various data sources through SQL queries, converted into a unified format, and a supply chain data set for each node is generated. These historical data sets contain the operation records and behavior patterns of the nodes, providing comprehensive basic information for subsequent analysis. Through this step, N node supply chain data sets are obtained.

[0020] Next, the associated indicators of the N node supply chain data sets are extracted respectively, and the correlation coefficient between each indicator and the target variable (such as inventory demand, production plan, sales demand, etc.) is calculated through the correlation analysis method. The linear relationship between each indicator and the target variable is evaluated through Pearson correlation analysis. For example, in the warehousing node, it is necessary to screen out indicators that are highly correlated with inventory demand, such as inventory turnover rate, inventory storage density, etc. The value range of the Pearson correlation coefficient is from -1 to 1. The closer to 1, the stronger the correlation, and the closer to 0, the less relevant. In this process, the indicators with a strong relationship with the target variable are screened out. For example, if the inventory turnover rate has a high correlation with the change in inventory demand (such as the Pearson coefficient is 0.85), the inventory turnover rate is an important indicator; if the inventory storage density has a low correlation with the change in inventory demand (such as the Pearson coefficient is 0.2), the indicator can be excluded. In this way, the supply chain associated indicator set of each node is finally screened out, and the supply chain associated indicator set of N nodes is obtained.

[0021] After that, the supply chain related indicator set of each node is clustered. This step uses the K-means clustering algorithm to group the indicators. Through K-means clustering, indicators with high similarity are grouped into the same group. For example, in a warehousing node, inventory turnover rate and inventory storage density may be grouped into the same cluster because they are both related to inventory liquidity and space utilization efficiency. K-means clustering iteratively optimizes the cluster center to maximize the similarity of indicators in the same cluster and the difference between different clusters. Through this method, the supply chain related indicator set of each node is divided into multiple clusters, generating clusters of related indicators of N nodes.

[0022] Finally, for the N node-related indicator clusters, the Pearson correlation analysis is used to calculate the correlation coefficient between the indicators in each cluster and the target variable (such as inventory demand, production plan, etc.). The higher the correlation coefficient, the greater the impact of the indicator on the target variable, so it will be retained. A correlation threshold (such as 0.7) is set to retain only those indicators that have a significant correlation with the target variable, and those indicators that are redundant or have a weak relationship with the target variable are eliminated. Through this screening process, the key analysis indicator set of N nodes is finally obtained.

[0023] Furthermore, in the method provided in the embodiment of the application, the step of obtaining the N node key analysis indicator sets further includes: According to the node impact evaluation index set, the inventory impact degree of the N supply chain nodes is analyzed respectively to obtain the inventory impact of the N nodes; according to the inventory impact of the N nodes, the threshold value of selecting the N node indicator clusters is determined; the correlation analysis is performed on each associated indicator in the N node associated indicator clusters respectively to obtain the N node cluster indicator correlation coefficient set; based on the N node indicator cluster selection threshold value, the indicator correlation screening is performed on the N node cluster indicator correlation coefficient set respectively to obtain the N node key analysis indicator set.

[0024] In the embodiment of the present application, the node impact evaluation index set of each supply chain node is first analyzed for the degree of inventory impact to quantify the contribution of each index to the inventory management target. The node impact evaluation index set contains indicators closely related to inventory management, such as inventory turnover rate, inventory storage density, and in-and-out frequency of storage nodes; production efficiency, production completion rate, and raw material consumption rate of production nodes; transportation timeliness, transportation delay rate, and delivery accuracy of transportation nodes. In the analysis, these indicators are first normalized to eliminate the impact of different indicators due to different dimensions and data ranges. Through normalization, the values ​​of all indicators are unified to between 0 and 1 to make them comparable. Then, technical experts assign weights to each indicator according to demand (for example, inventory turnover rate has a higher weight and inventory storage density has a lower weight), and perform weighted calculations on the normalized indicator values ​​to quantify their impact on inventory management targets (such as inventory demand and inventory turnover rate). Finally, the weighted contribution of each indicator is summarized to calculate the inventory impact of each node, reflecting the overall contribution of the node to inventory management. Through this process, the inventory impact of N nodes is obtained.

[0025] Next, based on the calculated inventory impact of the N nodes, the percentile setting method is used to determine the threshold for selecting the indicator cluster of each node. Specifically, the inventory impact of each node is sorted, its distribution characteristics are analyzed, and a threshold is set (such as selecting the indicator clusters with the top 30% inventory impact or clusters with an impact greater than 0.6). This threshold is used to screen out the indicator clusters that have the greatest impact on inventory management goals. In this way, the indicator clusters with inventory impacts higher than the threshold are marked as the focus of subsequent analysis. Through this process, the threshold for selecting the indicator clusters of N nodes is determined.

[0026] Then, correlation analysis is performed on each indicator in the associated indicator cluster of each node to evaluate the relationship between these indicators and the target variables (such as inventory demand, production plan, inventory turnover rate, etc.). Pearson correlation analysis is used to calculate the correlation coefficient, and the value range of the correlation coefficient is [-1, 1]. The closer to 1, the stronger the positive correlation, and the closer to -1, the stronger the negative correlation. Through this analysis, the correlation coefficient set of N node cluster indicators is obtained.

[0027] Finally, based on the N node indicator cluster selection thresholds, the indicator correlation coefficient sets of the N node cluster indicators are screened for indicator correlation, and indicators with low correlation are eliminated, and finally the key analysis indicator set of each node is obtained. Specifically, combined with the indicator cluster selection threshold of each node, a correlation screening standard (such as a correlation coefficient greater than 0.7) is set to select those indicators that are highly correlated with the target variable and eliminate redundant indicators with low correlation. For example, the inventory turnover rate with a correlation coefficient of 0.85 is retained, and the inventory storage density with a correlation coefficient of 0.3 is eliminated. Through this screening process, the key analysis indicator set of N nodes is finally obtained.

[0028] Step S300: performing multi-dimensional feature analysis on the N node product data sets based on the N node key analysis indicator sets to obtain N node supply chain feature sets.

[0029] In the embodiment of the present application, firstly, in order to eliminate the dimensional differences between different indicators and ensure that each indicator has equal importance in the analysis, all data are processed using Z-score standardization. Through standardization, the mean of each indicator is adjusted to 0 and the standard deviation is adjusted to 1, ensuring that all indicators are within the same scale range, thus eliminating the impact of the dimension on subsequent analysis.

[0030] Then, principal component analysis is used to extract features from the standardized data. PCA generates several principal components by linearly combining multiple indicators in the original data. These principal components can retain most of the variance in the data. First, the correlation between different indicators is quantified by calculating the covariance matrix; then the eigenvalues ​​and eigenvectors of the covariance matrix are calculated. The eigenvalues ​​reflect the contribution of each principal component to the data variance, and the eigenvectors represent the direction of the principal components; then, according to the size of the eigenvalues, the first few principal components are selected, which can explain most of the variability in the original data. Usually, the principal components with a cumulative explained variance of 70%-90% are selected to ensure that most of the information is retained. Finally, the standardized data is projected onto the selected principal components to generate the reduced-dimensional data. The feature set of each node is composed of these principal components, which represent the core characteristics of the node, such as inventory efficiency, production capacity, transportation time, etc. Through this process, the extracted principal components directly constitute the supply chain feature set of N nodes. These feature sets contain the core characteristics of each node in terms of inventory management, production efficiency, transportation time, etc., and can highly summarize the operating characteristics of the node.

[0031] Step S400: Construct a machine learning algorithm list, and perform matching analysis and integrated prediction on the N node supply chain feature sets based on the machine learning algorithm list to obtain product inventory demand forecast information.

[0032] In an embodiment of the present application, a list of machine learning algorithms is first constructed, which includes a variety of algorithms suitable for different node data characteristics, such as linear regression, decision tree, random forest, support vector machine, etc. Next, data characteristic analysis is performed for the supply chain feature set of each node to obtain the data characteristic information of each node. Based on these characteristic information, the most appropriate algorithm is selected from the list of machine learning algorithms for matching selection, and the matching machine learning algorithm for each node is determined. Then, these matching algorithms are used to perform inventory forecasting training on the supply chain data set of each node to obtain the inventory forecasting algorithm model of each node. Finally, through these trained algorithm models, the supply chain feature set of each node is subjected to inventory integrated forecasting, the forecast results of each node are summarized, and the product inventory demand forecasting information is finally obtained.

[0033] Furthermore, in the method provided in the embodiment of the application, the obtaining of product inventory demand forecast information further includes: Perform data characteristic analysis on the N node supply chain feature sets respectively to obtain data characteristic information of N nodes; perform matching selection according to the N node data characteristic information and the machine learning algorithm list to obtain N node matching machine learning algorithms; use the N node matching machine learning algorithms to perform inventory forecasting training on the N node supply chain data set to obtain an N node inventory forecasting algorithm model; perform inventory integrated forecasting on the N node supply chain feature set based on the N node inventory forecasting algorithm model to obtain product inventory demand forecasting information.

[0034] In an embodiment of the present application, data characteristic analysis is first performed on the supply chain feature sets of N nodes respectively, and the descriptive statistical analysis method is used to calculate the mean, standard deviation, minimum value, maximum value and other indicators of the data set of each node to understand the central tendency and dispersion of the data. In addition, the Pearson correlation analysis method is used to calculate the correlation coefficients between each feature and between the feature and the inventory demand target variable, so as to identify the key features that have a greater impact on inventory demand. For example, if the correlation coefficient between the inventory turnover rate and inventory demand of a node is high, it indicates that the inventory turnover rate is an important feature of the node. Through data characteristic analysis, the data characteristic information of N nodes is obtained.

[0035] Next, based on the data feature information of each node and the pre-built list of machine learning algorithms, the best matching algorithm is selected for each node. The constructed algorithm list includes linear regression, random forest, support vector machine, and gradient boosting machine. A rule-based matching method is used when selecting algorithms. For nodes with linear characteristics, a linear regression algorithm is selected; for nodes with complex nonlinear relationships, a random forest or GBM algorithm is selected; for high-dimensional data nodes, a support vector machine algorithm is selected. This rule-based algorithm matching ensures that the selected algorithm is highly consistent with the data characteristics. Through this process, N node matching machine learning algorithms are obtained.

[0036] Then, the selected matching machine learning algorithm is used to train inventory forecasting on the supply chain data set of each node. In this process, historical sample data of each node is obtained from the historical database. The input data includes various indicators in the supply chain feature set, such as inventory turnover rate, production efficiency, transportation time, etc. The output data is the inventory demand of each node. After preprocessing, these data are divided according to a certain ratio (such as 80% training set and 20% test set) for training and evaluating machine learning models. Through training, the algorithm learns the relationship between the input features and the target variable (inventory demand) and generates N node inventory forecasting algorithm models.

[0037] Finally, based on the inventory forecasting algorithm model of N nodes, the inventory integrated forecasting of the supply chain feature set of N nodes is carried out. The inventory forecasting algorithm model of N nodes is integrated through the model integration strategy to generate a meta-forecasting integrated strategy model. Then the prediction performance of the model is evaluated and the performance parameters are obtained. Based on these parameters, the model optimizer is designed, and the integrated model is iteratively tuned to finally obtain the inventory forecasting integrated strategy model. The optimized model performs inventory forecasting based on the supply chain feature set of N nodes and finally generates product inventory demand forecasting information.

[0038] Furthermore, in the method provided in the embodiment of the application, the obtaining of product inventory demand forecast information further includes: The model integration strategy is adopted to integrate and fuse the inventory forecasting algorithm models of the N nodes to generate a meta-forecasting integrated strategy model; the meta-forecasting integrated strategy model is evaluated and verified to obtain model forecasting performance parameters, and a model optimizer is designed based on the model forecasting performance parameters; the model optimizer is used to iteratively tune the meta-forecasting integrated strategy model to obtain an inventory forecasting integrated strategy model; inventory forecasting is performed on the N-node supply chain feature set based on the inventory forecasting integrated strategy model to obtain the product inventory demand forecast information.

[0039] In an embodiment of the present application, a model integration strategy (such as weighted average method) is first used to fuse N node inventory forecasting algorithm models to generate a meta-forecasting integrated strategy model. This process weights the prediction results of each node by weighted averaging. Specifically, the weight is assigned according to the mean square error of each model, which is a common indicator for measuring model prediction error. A lower MSE means that the model performs better on the training data and can more accurately predict inventory demand. Therefore, a model with a lower MSE should be given a larger weight, and conversely, a model with a higher MSE should be given a smaller weight. Specifically, for each inventory forecasting algorithm model, by predicting on the training set, the mean square error between its predicted value and the actual value is calculated, and then the MSE value of each model is taken inversely (i.e., using the inverse method) to obtain the weight of each model. In order to ensure that the sum of the weights is 1, all weights are normalized. Finally, the meta-forecasting integrated strategy model is generated by weighted averaging the prediction results of each model according to its weight.

[0040] Next, the K-fold cross-validation method is used to evaluate and verify the meta-forecasting integration strategy model. Cross-validation divides the historical inventory demand data set and the supply chain feature data set into K subsets (usually K=5), uses K-1 subsets for training each time, and the remaining subset is used for testing. After each round of cross-validation, the performance of the model on the test set is quantified by calculating evaluation indicators such as mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R²) to ensure that it has good generalization ability and prediction accuracy. Through cross-validation, the model prediction performance parameters such as MSE, MAE, and R² are obtained. These performance parameters provide a basis for model optimization and help identify models with better performance.

[0041] Based on the obtained model prediction performance parameters, a model optimizer was designed and introduced. The optimizer uses the grid search method to tune the hyperparameters of the meta-forecasting ensemble strategy model. The grid search method selects the optimal hyperparameter configuration by exhaustively enumerating all possible hyperparameter combinations and evaluating the performance of each combination on the validation set. Common hyperparameters include the maximum depth of the decision tree, the number of trees in the random forest, the learning rate of XGBoost, etc. In this process, the optimizer adjusts these hyperparameters to ensure that the model can better adapt to the supply chain characteristics of different nodes, thereby improving the accuracy of inventory prediction. After multiple rounds of optimization and hyperparameter adjustment, the inventory prediction ensemble strategy model is finally obtained.

[0042] Finally, based on the optimized inventory forecasting integrated strategy model, inventory forecasting is performed on the N node supply chain feature sets. The supply chain feature set of each node contains multiple factors related to inventory demand, such as inventory turnover, production capacity, sales demand, transportation time, etc. These features are input into the optimized integrated strategy model, and the model outputs the inventory demand forecast value of each node. Finally, the forecast results of all nodes are summarized to obtain the overall product inventory demand forecast information.

[0043] Step S500: construct an inventory strategy optimization space based on the N supply chain nodes, use the inventory strategy optimization space to perform inventory adjustment analysis on the product inventory demand forecast information, output inventory optimization control parameters, and perform supply chain inventory management based on the inventory optimization control parameters.

[0044] In an embodiment of the present application, an inventory strategy optimization space is first constructed based on N supply chain nodes. The space contains multiple possible combinations of different inventory management strategies, aiming to provide a multi-dimensional optimization framework for inventory management. Then, the optimization space is used to perform inventory adjustment analysis in combination with product inventory demand forecast information, analyze the difference between inventory demand and strategy, and generate inventory optimization control parameters. These control parameters include optimal inventory levels, replenishment points, replenishment quantities, etc., which can guide the dynamic adjustment of inventory. Finally, based on the inventory optimization control parameters, supply chain inventory management is performed to achieve refined control of inventory, thereby ensuring that inventory meets demand, reduces inventory costs, and improves the overall efficiency of the supply chain.

[0045] Furthermore, in the method provided in the embodiment of the application, the building of the inventory strategy optimization space further includes: According to the N supply chain nodes, N node inventory policy databases are obtained; an inventory control effect evaluation function is constructed, and the effect of the N node inventory policy databases is evaluated based on the inventory control effect evaluation function to obtain the N node inventory policy effect fitness; based on the N node inventory policy effect fitness, the N node inventory policy databases are mutated and cross-expanded to obtain N node inventory policy update databases; the N node inventory policy update databases are arranged and integrated according to the order of supply chain nodes to construct the inventory policy optimization space.

[0046] In the embodiment of the present application, N node inventory strategy databases are first constructed based on N supply chain nodes. The database records the inventory management strategy of each node, including inventory level, replenishment frequency, replenishment quantity, etc. These inventory strategies are formulated by technical experts based on historical data, demand forecasts and supply chain operation characteristics, with the aim of providing diversified management solutions for subsequent inventory optimization. Through this process, N node inventory strategy databases are obtained.

[0047] Next, we construct an inventory control effect evaluation function, which is used to evaluate the effect of each inventory strategy and quantify the performance of the strategy in reducing inventory costs, improving inventory turnover, and reducing out-of-stock rates. The evaluation function is implemented by weighted summation of multiple objectives (such as inventory cost, out-of-stock rate, inventory turnover rate, etc.), where inventory cost, out-of-stock rate, and inventory turnover rate have the same weight. Then, the inventory strategy of each node is evaluated by applying the evaluation function to obtain the fitness of the inventory strategy effects of N nodes.

[0048] Then, based on the fitness of the inventory strategy effects of N nodes, the N node inventory strategy database is mutated and cross-expanded. This process draws on the ideas of genetic algorithms, generates new inventory strategies through mutation and crossover operations, and further improves the diversity and optimization effect of inventory strategies. Specifically, the mutation operation explores new inventory strategy combinations by randomly adjusting certain strategy parameters (such as replenishment frequency). The crossover operation selects two inventory strategies with better performance (i.e., strategies with higher fitness), exchanges some of their parameters, and generates new inventory strategy combinations. These new strategies can combine the advantages of the two strategies and explore more flexible inventory management solutions. Through multiple rounds of mutation and cross-expansion, a large number of new inventory strategies are generated, and the strategies with excellent performance are screened out through fitness evaluation, and finally an N-node inventory strategy update database is obtained.

[0049] Finally, the N node inventory strategy update databases are arranged and integrated according to the order of supply chain nodes to ensure that the inventory strategies between different nodes can effectively coordinate in the optimization space. Through integration, the inventory strategy optimization space is obtained.

[0050] Furthermore, in the method provided in the embodiment of the application, the constructing of the inventory strategy optimization space further includes: The N node inventory policy update databases are arranged and integrated according to the order of supply chain nodes to obtain an inventory policy global database; the influencing factors of the N supply chain nodes are extracted in turn according to the order of supply chain nodes to obtain N node influencing factor sets; based on the N node influencing factor sets, the N supply chain nodes are subjected to cascade impact fitting analysis to construct a supply chain node correlation impact model; based on the supply chain node correlation impact model, the inventory policy global database is optimized and revised to construct the inventory policy optimization space.

[0051] In the embodiment of the present application, first, according to N supply chain nodes, the N node inventory strategy update database is arranged and integrated according to the order of supply chain nodes. This step adopts data integration and sorting methods to ensure that the inventory strategy can be arranged according to the order of each node in the supply chain, and the logic and coordination between the strategies are maintained. By integrating the inventory strategies of each node according to the order of nodes in the supply chain, a global inventory strategy database is obtained, which contains the inventory management strategies of all nodes and provides a comprehensive data framework for subsequent optimization.

[0052] Next, in accordance with the order of supply chain nodes, the influencing factors of each N supply chain node are extracted in turn. In this process, feature extraction and data mining methods are used to identify the key influencing factors of each node in inventory management. These influencing factors may include demand fluctuations, production capacity, inventory turnover, transportation time, etc., all of which are closely related to inventory management decisions. By using data mining techniques, such as time series analysis to identify the trend of demand changes and correlation analysis to identify the key factors affecting inventory strategies, the influencing factors of each node are finally extracted. These influencing factors constitute the N node influencing factor set.

[0053] Based on the extracted set of N node influencing factors, a cascade impact fitting analysis is performed to identify the interactions between nodes and how they jointly affect inventory strategies. In this process, regression analysis is used to analyze which factors affect the inventory strategy of each node, and these effects are quantified through regression equations. At the same time, considering the dependencies between the nodes, regression analysis is used to clarify how the inventory strategy of one node is fed back to the inventory strategies of other nodes through factors such as demand fluctuations and production capacity. Through this analysis, a supply chain node association impact model is derived, which reveals the interdependencies between the nodes and quantifies the impact of these relationships on inventory management.

[0054] Finally, based on the supply chain node association impact model, the global database of inventory strategies is optimized and revised, and the inventory strategy optimization space is constructed. In this stage, the constrained optimization method is used to analyze how the inventory strategies of each node are affected by other nodes, and the inventory strategies of each node are adjusted to ensure the coordination between them. The goal of the optimization is to minimize inventory costs, improve inventory turnover, and reduce out-of-stock rates by adjusting parameters such as inventory levels and replenishment frequency. In the optimization process, the resource limitations and constraints of each node are considered, such as the minimum inventory quantity and replenishment cycle. Through this optimization process, the inventory strategy optimization space is obtained, which contains a variety of optimized inventory strategies and provides a variety of possible decision-making schemes for inventory management in the supply chain.

[0055] Furthermore, in the method provided in the embodiment of the application, the outputting of inventory optimization control parameters further includes: The product inventory demand forecast information is used as a constraint parameter, and a parameter search is performed in the inventory strategy optimization space to obtain a target inventory strategy optimization memory library; a parameter particle swarm is initialized according to the target inventory strategy optimization memory library; and the inventory control effect evaluation function is used to iteratively update and optimize in the parameter particle swarm until a preset termination condition is reached, and the inventory optimization control parameters are output.

[0056] In an embodiment of the present application, the product inventory demand forecast information is first used as a constraint parameter to perform a parameter search in the inventory strategy optimization space. The purpose of this process is to optimize the inventory management strategy through forecast information to obtain the best inventory strategy that adapts to future needs. At this stage, an optimization algorithm (such as particle swarm optimization or genetic algorithm) is used to search for the best inventory management strategy in the inventory strategy optimization space. By using the product inventory demand forecast information as a constraint condition and limiting the search scope, it is ensured that the obtained strategy can not only meet future inventory needs, but also achieve the optimal inventory management goals (such as reducing inventory costs, improving inventory turnover, etc.). After this search process, the target inventory strategy optimization memory library is obtained, which contains a variety of optimized inventory management strategies.

[0057] Next, based on the target inventory policy optimization memory, the parameter particle swarm is initialized to prepare for subsequent optimization iterations. Particle swarm optimization is an optimization algorithm that simulates the foraging behavior of bird flocks in nature. Each particle represents a potential inventory management strategy, and the "position" of the particle represents the specific inventory strategy parameters, such as inventory level, replenishment quantity, replenishment frequency, etc. In this process, the position of each particle in the particle swarm is initialized according to the strategy extracted from the target inventory policy optimization memory, and the starting solution of the particle is set according to these initial strategies. By simulating natural selection and collaboration between individuals, the particle swarm can efficiently search the solution space and find the optimal inventory management strategy.

[0058] Then, the inventory control effect evaluation function is used to iteratively update and optimize within the parameter particle swarm. This function evaluates the inventory strategy effect of each particle based on the effect of the inventory strategy, such as inventory cost, inventory turnover rate, out-of-stock rate and other indicators. Each particle will be adjusted according to its current effect so that it can get closer to the global optimal solution in the next iteration. Through multiple iterations, the particle swarm gradually updates the position of each particle and gradually optimizes the inventory strategy until the preset termination condition (such as the maximum number of iterations) is met.

[0059] Finally, when the particle swarm iteration is completed, the inventory optimization control parameters are output. These parameters include key inventory management strategies such as optimized inventory level, replenishment quantity, replenishment cycle, etc.

[0060] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects: The present application disassembles the target supply chain nodes to obtain N supply chain nodes, and collects related data for the N supply chain nodes to obtain N node product data sets; extracts related indicators and screens correlation for the N supply chain nodes respectively to obtain N node key analysis indicator sets; performs multidimensional feature analysis on the N node product data sets based on the N node key analysis indicator sets to obtain N node supply chain feature sets; constructs a machine learning algorithm list, performs matching analysis and integrated prediction on the N node supply chain feature sets based on the machine learning algorithm list, and obtains product inventory demand forecast information; constructs an inventory strategy optimization space according to the N supply chain nodes, uses the inventory strategy optimization space to perform inventory adjustment analysis on the product inventory demand forecast information, outputs inventory optimization control parameters, and performs supply chain inventory management based on the inventory optimization control parameters. The present invention solves the technical problem of inaccurate inventory management in the prior art, extracts key analysis indicators and feature sets of each node by disassembling and collecting data for the target supply chain nodes, combines related indicator extraction, correlation screening, feature analysis and other technologies, uses the machine learning algorithm list to perform matching analysis and integrated prediction, and obtains product inventory demand forecast information. Based on the prediction results, the inventory strategy optimization space is constructed, inventory adjustment analysis is performed and optimization control parameters are output, so as to realize inventory optimization management and achieve the technical effect of improving the accuracy of inventory management and supply chain efficiency.

[0061] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0062] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0063] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A supply chain inventory management optimization method based on machine learning, characterized in that: The method comprises: Decompose the target supply chain into nodes to obtain N supply chain nodes, and collect related data for the N supply chain nodes to obtain N node product data sets; Extracting related indicators and screening correlations for the N supply chain nodes respectively to obtain a set of key analysis indicators for the N nodes; Performing multi-dimensional feature analysis on the N node product data sets based on the N node key analysis indicator sets to obtain N node supply chain feature sets; Construct a machine learning algorithm list, and perform matching analysis and integrated prediction on the N node supply chain feature sets based on the machine learning algorithm list to obtain product inventory demand forecast information; According to the N supply chain nodes, an inventory strategy optimization space is constructed, and the inventory adjustment analysis of the product inventory demand forecast information is performed using the inventory strategy optimization space to output inventory optimization control parameters, and supply chain inventory management is performed based on the inventory optimization control parameters.

2. The supply chain inventory management optimization method based on machine learning as claimed in claim 1, characterized in that: The obtaining of the N node key analysis indicator sets includes: Perform historical product data mining based on the N supply chain nodes to obtain N node supply chain data sets; Extracting related indicators from the N node supply chain data sets respectively to obtain N node supply chain related indicator sets; Performing cluster analysis on the N node supply chain related indicator sets respectively to generate N node related indicator clusters; The N node-related indicator clusters are respectively screened for correlation to obtain N node key analysis indicator sets.

3. The supply chain inventory management optimization method based on machine learning as claimed in claim 2, characterized in that: The obtaining of the N node key analysis indicator sets includes: According to the node impact evaluation indicator set, the inventory impact degree of the N supply chain nodes is analyzed respectively to obtain the inventory impact degree of the N nodes; Determine the selection threshold of the N node indicator clusters according to the inventory influence of the N nodes; Performing correlation analysis on each correlation indicator in the N node correlation indicator clusters respectively to obtain a set of correlation coefficients of N node cluster indicators; Based on the N node indicator cluster selection thresholds, the N node cluster indicator correlation coefficient sets are respectively screened for indicator correlation to obtain the N node key analysis indicator sets.

4. The supply chain inventory management optimization method based on machine learning as claimed in claim 2, characterized in that: The obtaining of product inventory demand forecast information includes: Performing data characteristic analysis on the N node supply chain feature sets respectively to obtain data characteristic information of the N nodes; Matching and selecting according to the N node data characteristic information and the machine learning algorithm list to obtain N node matching machine learning algorithms; Using the N-node matching machine learning algorithm to perform inventory forecasting training on the N-node supply chain data set to obtain an N-node inventory forecasting algorithm model; Based on the N-node inventory forecasting algorithm model, an inventory integrated forecast is performed on the N-node supply chain feature set to obtain product inventory demand forecast information.

5. The supply chain inventory management optimization method based on machine learning as claimed in claim 4, characterized in that: The obtaining of product inventory demand forecast information includes: The N node inventory forecasting algorithm models are integrated and merged by adopting a model integration strategy to generate a meta-forecasting integration strategy model; Evaluating and verifying the meta-prediction integrated strategy model to obtain model prediction performance parameters, and designing a model optimizer based on the model prediction performance parameters; Iteratively tuning the meta-forecasting integrated strategy model using the model optimizer to obtain an inventory forecasting integrated strategy model; Based on the inventory forecasting integrated strategy model, inventory forecasting is performed on the N node supply chain feature sets to obtain the product inventory demand forecasting information.

6. The supply chain inventory management optimization method based on machine learning as claimed in claim 1, characterized in that: The construction of inventory strategy optimization space includes: According to the N supply chain nodes, obtain N node inventory strategy databases; Constructing an inventory control effect evaluation function, and performing effect evaluation on the N-node inventory strategy database based on the inventory control effect evaluation function to obtain the N-node inventory strategy effect fitness; Based on the fitness of the N node inventory policy effects, the N node inventory policy database is mutated and cross-expanded to obtain an N node inventory policy update database; The N node inventory strategy update databases are arranged and integrated according to the order of supply chain nodes to construct the inventory strategy optimization space.

7. The supply chain inventory management optimization method based on machine learning as claimed in claim 6, characterized in that: The constructing of the inventory strategy optimization space includes: Arrange and integrate the N node inventory policy update databases according to the order of supply chain nodes to obtain an inventory policy global database; Extracting influencing factors from the N supply chain nodes in order of supply chain nodes to obtain a set of influencing factors of N nodes; Based on the N node influencing factor sets, a cascade impact fitting analysis is performed on the N supply chain nodes to construct a supply chain node correlation impact model; The inventory strategy global database is optimized and modified based on the supply chain node association impact model to construct the inventory strategy optimization space.

8. The supply chain inventory management optimization method based on machine learning as claimed in claim 7, characterized in that: The output inventory optimization control parameters include: Using the product inventory demand forecast information as a constraint parameter, performing parameter search in the inventory strategy optimization space to obtain a target inventory strategy optimization memory library; Optimize the memory base according to the target inventory strategy and initialize the parameter particle swarm; The inventory control effect evaluation function is used to iteratively update and optimize the parameter particle swarm until a preset termination condition is reached, and the inventory optimization control parameters are output.

Images