Supply chain management method and device for cross-border e-commerce

Abstract

The invention discloses a supply chain management method and device for cross-border e-commerce, and relates to the technical field of logistics management. According to historical transaction data, determining statistics of each feature and target commodity sales volume in each time period, and sorting the statistics from high to low to obtain a feature correlation sorting set; and obtaining a first preset number of features in the feature correlation sorting set to obtain a second feature set, performing effective feature screening on the second feature set through sequential forward selection, performing clustering, performing sales prediction to obtain predicted sales, and performing supply chain management according to the predicted sales of all retailers. Through deep analysis of different features and a sequential forward selection method, features most relevant to sales volume prediction can be screened out, unnecessary noise and errors are reduced so that a prediction result is more accurate, then resource allocation is performed according to the prediction result, the phenomena of stock overstock and stockout caused by uneven resource allocation are avoided, and the resource allocation efficiency is improved. And the efficiency and the operation effect of the supply chain are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of logistics management, and specifically relates to a supply chain management method and device for cross-border e-commerce. Background Art

[0002] Cross-border e-commerce, or cross-border e-commerce in full, refers to the sale of goods or services by trading entities in different countries or regions through the Internet platform. This type of business activity involves multiple links such as cross-border transactions, logistics, payment, and customer service, breaking the geographical restrictions of traditional trade, allowing companies or individuals to directly face the global market and conduct more extensive and convenient business activities.

[0003] Traditional supply chain management usually relies on fixed processes and forecasting models for a single market. However, with the rise of cross-border e-commerce, which involves suppliers, retailers, consumers, and logistics networks in multiple countries and regions, traditional supply chain management methods can no longer effectively cope with globalization and diversification. Existing supply chain management is mainly optimized based on the experience and methods of existing markets, often meeting demand through static forecasts and fixed inventory management strategies. However, this model ignores the impact of factors such as demand differences and market fluctuations in cross-border e-commerce around the world, resulting in uneven resource allocation, inventory backlogs or out-of-stock phenomena, which seriously affects the efficiency and operational effectiveness of the supply chain. Summary of the invention

[0004] The purpose of the present invention is to solve the problems of uneven resource allocation, inventory backlog or out-of-stock, which seriously affect the efficiency and operation effect of the supply chain, and to propose a supply chain management method and device for cross-border e-commerce.

[0005] In a first aspect of the present invention, a supply chain management method for cross-border e-commerce is first proposed, the method comprising: Acquire historical transaction data of a target commodity from a target retailer, and determine the characteristics of each time period according to the historical transaction data to obtain a first feature set; the target retailer is any retailer that sells the target commodity; According to the first feature set, calculate the statistics of each feature and the sales volume of the target product in each time period, and sort the statistics from high to low to obtain a feature-related sorting set; Acquire a preset number of features in the feature-related sorted set to obtain a second feature set, and perform effective feature screening on the second feature set through sequential forward selection to obtain a target feature set; After clustering the features in the target feature set, sales forecasting is performed to obtain predicted sales, resource allocation for each retailer is obtained based on the predicted sales of all retailers, and supply chain management is performed based on the resource allocation for each retailer.

[0006] Optionally, calculating the statistics of each feature and the sales volume of the target product in each time period includes: Calculate the Pearson correlation coefficient and variance analysis between each feature and the target product sales volume in the same time period to obtain a first influence coefficient set and a second influence coefficient set; Normalizing the first influence coefficient set and the second influence coefficient set respectively to obtain a first normalized coefficient set and a second normalized coefficient set; For each feature, the first normalization coefficient and the second normalization coefficient corresponding to the feature are averaged to obtain the statistic corresponding to the feature.

[0007] Optionally, clustering the features in the target feature set and performing sales forecasting to obtain the forecasted sales includes: Clustering the target feature set according to each type of feature, obtaining a ranking of features of the same type in a time series to obtain a first feature ranking cluster; For each first feature sorting cluster, the first feature sorting cluster is input into a preset neural network for forward propagation to obtain a simulated sales volume corresponding to the feature in each time period in the time series; For each time period, the error coefficient corresponding to the feature in the time period is obtained according to the simulated sales volume and the actual sales volume in the time period, and the error coefficient in the time series is obtained to obtain the error change curve corresponding to the feature; The predicted sales volume is determined based on the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period.

[0008] Optionally, determining the predicted sales volume according to the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period includes: For each feature corresponding to the current time period, determine the change trend of each feature in the corresponding error change curve according to the sales volume corresponding to the current time period, and determine the error change value corresponding to the next time period of the current time period according to the change trend; For each feature, the feature is adjusted according to the error change value corresponding to the feature to obtain an adjusted feature; Substitute all adjusted features into the preset neural network and obtain the predicted sales volume through forward propagation.

[0009] Optionally, the resource allocation for each retailer based on the predicted sales of all retailers includes: For each retailer, obtain the retailer's current inventory and predicted sales, and obtain the retailer's required resources based on the current inventory and predicted sales; The resource allocation of each retailer is obtained by allocating resources on demand to each retailer according to the required resource amounts of all retailers and the total resource pool.

[0010] In a second aspect of the present invention, a supply chain management device for cross-border e-commerce is provided, comprising: A first feature set determination module is used to obtain historical transaction data of a target commodity from a target retailer, and determine the features of each time period according to the historical transaction data to obtain a first feature set; the target retailer is any retailer that sells the target commodity; A feature sorting module, used to calculate the statistics of each feature and the sales volume of the target product in each time period according to the first feature set, and sort the statistics from high to low to obtain a feature related sorting set; An effective feature screening module, used for obtaining a preset number of features in the feature-related sorting set to obtain a second feature set, and performing effective feature screening on the second feature set by sequential forward selection to obtain a target feature set; The resource allocation module is used to cluster the features in the target feature set and then perform sales forecasting to obtain predicted sales, obtain resource allocation for each retailer based on the predicted sales of all retailers, and perform supply chain management based on the resource allocation for each retailer.

[0011] Optionally, the feature sorting module includes: An influence coefficient determination module, used to calculate the Pearson correlation coefficient and variance analysis between each feature and the target product sales volume in the same time period to obtain a first influence coefficient set and a second influence coefficient set; A coefficient normalization module, used to normalize the first influence coefficient set and the second influence coefficient set respectively to obtain a first normalized coefficient set and a second normalized coefficient set; The statistic determination module is used to average the first normalization coefficient and the second normalization coefficient corresponding to each feature to obtain the statistic corresponding to the feature.

[0012] Optionally, the resource allocation module includes: A first feature sorting and clustering module is used to cluster the target feature set according to each type of feature, and obtain a first feature sorting and clustering by sorting features of the same type in a time series; A simulated sales volume determination module is used to input each first feature sorting cluster into a preset neural network for forward propagation to obtain a simulated sales volume corresponding to the feature in each time period in the time series; The error change curve determination module is used to obtain the error coefficient corresponding to the feature in each time period according to the simulated sales volume and the actual sales volume in the time period, and obtain the error coefficient on the time series to obtain the error change curve corresponding to the feature; The predicted sales volume determination module is used to determine the predicted sales volume based on the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period.

[0013] Optionally, the predicted sales volume determination module includes: The error change value determination module is used to determine the change trend of each feature in the corresponding error change curve according to the sales volume corresponding to the current time period for each feature corresponding to the current time period, and determine the error change value corresponding to the next time period of the current time period according to the change trend; An adjustment feature determination module is used to adjust each feature according to the error change value corresponding to the feature to obtain an adjustment feature; The predicted sales volume generation module is used to substitute all the adjusted features into the preset neural network and obtain the predicted sales volume through forward propagation.

[0014] Optionally, the resource allocation module further includes: A required resource quantity determination module is used to obtain the current inventory quantity and predicted sales volume of each retailer, and obtain the required resource quantity of the retailer according to the current inventory quantity and predicted sales volume; The on-demand allocation module is used to perform on-demand resource allocation for each retailer according to the required resource amounts of all retailers and the total resource pool to obtain resource allocation for each retailer.

[0015] Beneficial effects of the present invention: The present invention proposes a supply chain management method for cross-border e-commerce, which obtains historical transaction data of target goods in target retailers, determines the characteristics of each time period according to the historical transaction data to obtain a first feature set; the target retailer is any retailer selling the target goods; according to the first feature set, calculates the statistics of each feature and the sales volume of the target goods in each time period, and sorts the statistics from high to low to obtain a feature-related sorting set; obtains the first preset number of features in the feature-related sorting set to obtain a second feature set, and screens the second feature set for effective features through sequential forward selection to obtain a target feature set; clusters the features in the target feature set and then performs sales forecasting to obtain predicted sales, obtains the resource allocation of each retailer according to the predicted sales of all retailers, and performs supply chain management according to the resource allocation of each retailer. Through in-depth analysis of different features and a sequential forward selection method, the most relevant features for sales forecasting can be screened out, unnecessary noise and errors can be reduced to make the forecast results more accurate, and then resources are allocated according to the forecast results, avoiding the phenomenon of inventory backlog and out-of-stock caused by uneven resource allocation, and improving the efficiency and operation effect of the supply chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the accompanying drawings.

[0017] Figure 1 A flow chart of a supply chain management method for cross-border e-commerce is provided for an embodiment of the present invention; Figure 2 A schematic structural diagram of a supply chain management device for cross-border e-commerce is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the present invention, the description of "first", "second", etc. is only used for descriptive purposes, and cannot be understood as indicating or implying its relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0019] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0020] The embodiment of the present invention provides a supply chain management method for cross-border e-commerce. Figure 1 , Figure 1 A flow chart of a supply chain management method for cross-border e-commerce provided by an embodiment of the present invention. The method comprises the following steps: S101, obtaining historical transaction data of a target commodity from a target retailer, and determining the features of each time period according to the historical transaction data to obtain a first feature set; S102, based on the first feature set, calculating the statistics of each feature and the target product sales volume in each time period, and sorting the statistics from high to low to obtain a feature-related sorting set; S103, obtaining a preset number of features in the feature-related sorting set to obtain a second feature set, and performing effective feature screening on the second feature set by sequential forward selection to obtain a target feature set; S104, clustering the features in the target feature set and performing sales forecasting to obtain predicted sales, obtaining resource allocation for each retailer based on the predicted sales of all retailers, and performing supply chain management based on the resource allocation for each retailer.

[0021] The target retailer is any retailer that sells the target product.

[0022] A supply chain management method for cross-border e-commerce provided based on an embodiment of the present invention can screen out the features most relevant to sales forecasting through in-depth analysis of different features and a sequential forward selection method, thereby reducing unnecessary noise and errors to make the forecast results more accurate, and then allocating resources based on the forecast results, thereby avoiding inventory backlogs and out-of-stock phenomena caused by uneven resource allocation, and improving the efficiency and operational results of the supply chain.

[0023] In one implementation, by obtaining the historical transaction data of the target retailer and calculating the correlation between each feature and sales, features closely related to sales can be extracted. The data-driven approach is more adaptable to market changes than traditional static forecasting models, providing more accurate sales forecasts and reducing forecast errors.

[0024] In one implementation, the most effective features are selected through a sequential forward selection method, and then these features are clustered to further optimize the accuracy of sales forecasts, so that relevant features can be dynamically adjusted according to market changes, thereby continuously improving the adaptability and accuracy of the forecasting model.

[0025] In one implementation, the target product is the product to be managed, and the historical transaction data is all transaction data of the target retailer on the target product in the past five years. The historical transaction data are grouped according to time periods, and the time periods are determined by technical personnel. Assuming the time period is 7 days, all transaction data within 5 years are divided into time periods of 7 days and according to time series. The divided time periods must correspond to seasons and whether there is a promotion. The characteristics of each time period are determined based on the historical transaction data to obtain a first feature set, specifically obtaining the sales quantity, price, inventory level, promotional activity information, sales channel information, season, and sales volume trend of the previous time period compared to this period, etc. corresponding to each time period as the first feature.

[0026] In one implementation, a preset number of features are determined by a technician; each retailer here refers to a retailer that sells the target product; the second feature set is screened for effective features through sequential forward selection to obtain a target feature set. Specifically, an empty feature set is first established as a valid feature set, the feature ranked first in the second feature set is placed in the valid feature set, and then the other features in the second feature set are recorded as candidate features to obtain a candidate feature set. For each feature in the candidate feature set, the mean square error is calculated after the feature is added to the valid feature set, and the feature with the smallest mean square error is obtained and recorded as a valid feature and added to the valid feature set, and the feature is removed from the candidate feature set, and then valid features are repeatedly obtained from the candidate feature set until a preset termination condition is met to terminate the extraction of effective features; the preset termination condition may be that the calculated mean square error exceeds a preset error value, the feature selection in the second feature set is completed, and the number of cycles reaches a preset value. The specific cycle condition is determined by a technician.

[0027] In one implementation method, resources are allocated to each retailer based on sales forecasts, so that the demands of different regions and markets can be supported by more accurate resources. The inventory, transportation and production plans in the supply chain can better meet market demand, avoiding waste of resources or insufficient supply.

[0028] In one implementation, since sales forecasts are more accurate, the supply chain can reasonably adjust inventory levels based on predicted demand, avoiding inventory backlogs and product out-of-stock issues. Excess inventory and out-of-stock issues not only increase operating costs but may also affect customers' shopping experience. This solution can effectively reduce these problems.

[0029] In one embodiment, calculating the statistics of each feature and the sales volume of the target product in each time period includes: Calculate the Pearson correlation coefficient and variance analysis between each feature and the target product sales in the same time period to obtain the first influence coefficient set and the second influence coefficient set; Normalizing the first influence coefficient set and the second influence coefficient set respectively to obtain a first normalized coefficient set and a second normalized coefficient set; For each feature, the first normalization coefficient and the second normalization coefficient corresponding to the feature are averaged to obtain the statistic corresponding to the feature.

[0030] In one implementation, the linear relationship between the feature and the sales volume can be quantified by calculating the Pearson correlation coefficient between each feature and the sales volume of the target product. The value of the Pearson correlation coefficient can fluctuate between -1 and +1, with a value close to +1 indicating a strong positive correlation between the feature and the sales volume, a value close to -1 indicating a strong negative correlation, and a value close to 0 indicating almost no linear relationship.

[0031] In one implementation, the relationship between each feature and sales volume is quantified by the first influence coefficient set (based on the Pearson correlation coefficient) and the second influence coefficient set (based on variance analysis), respectively. The features can be evaluated from different perspectives. The Pearson correlation coefficient focuses on the linear relationship between the feature and sales volume, and can capture direct correlations. Variance analysis focuses on the differences between categories, helping to determine the impact of different categories (such as whether there are promotional activities, different price ranges, etc.) on sales volume. These two methods can be used to comprehensively evaluate the influence of features from multiple dimensions. Some features may have a greater impact on the Pearson correlation coefficient, but not significantly in the variance analysis, and vice versa. Combining the results of these two methods can provide a more comprehensive understanding of the relationship between features and sales volume.

[0032] In one implementation, normalization of the first influence coefficient set and the second influence coefficient set is performed using the Min-Max normalization formula. By integrating the analysis results of multiple dimensions and performing normalization, the feature statistics finally obtained can more accurately reflect the influence of each feature on sales, providing a more accurate input feature for the sales forecasting model, thereby improving the accuracy of the forecast.

[0033] In one embodiment, clustering the features in the target feature set and then performing sales forecasting to obtain the forecasted sales includes: Cluster the target feature set according to each type of feature, and obtain the ranking of the same type of features in the time series to obtain the first feature ranking cluster; For each first feature sorting cluster, the first feature sorting cluster is input into a preset neural network for forward propagation to obtain a simulated sales volume corresponding to the feature in each time period in the time series; For each time period, the error coefficient corresponding to the feature in the time period is obtained according to the simulated sales volume and the actual sales volume in the time period, and the error coefficient in the time series is obtained to obtain the error change curve corresponding to the feature; The predicted sales volume is determined based on the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period.

[0034] In one implementation, the target feature set is clustered according to each type of feature, which can group the same features and extract different types of feature patterns. This helps to identify related feature performance in the time series, so that the model can make more accurate sales forecasts based on the same features; the features are modeled through neural networks and forward propagation is performed to obtain simulated sales, which can capture more complex nonlinear relationships.

[0035] In one implementation, the actual sales volume is the sales volume recorded at the time corresponding to the time period in the historical data; and the error coefficient is the value obtained by subtracting the simulated sales volume from the actual sales volume and dividing it by the actual sales volume.

[0036] In one implementation, the preset neural network consists of an input layer, two hidden layers and an output layer. The input layer has N input layer nodes, where N is the same as the number of features in the target feature set. The first hidden layer has 15 neurons, the second hidden layer has 10 neurons, and the sigmoid logarithm is assigned to the two hidden layers. The number of hidden layer nodes of each network will be automatically selected according to the optimization process through the CapSA optimization algorithm; the output layer contains one neuron, and the preset neural network is trained through historical data to obtain the corresponding output predicted sales under different feature combinations.

[0037] In one implementation, by comparing simulated sales with actual sales, calculating the error coefficient, and observing the error change curve in the time series, it is helpful to monitor the prediction performance of the model in different time periods, discover the changing trend of model errors and make adaptive adjustments; error analysis enables the model to be continuously improved, reduce error accumulation, and provide a powerful feedback mechanism for the self-correction of the model. By observing the changing trend of errors, the prediction can be dynamically adjusted in different time periods to avoid the model from deviating from the actual situation for a long time.

[0038] In one implementation, the error change curve and the performance of the current features can be combined to more accurately predict future sales. The dynamic adjustment of feature errors enables the model to correct errors in a timely manner when facing real-time changes, providing more accurate sales forecasts. By comprehensively considering the error change curves of all features, current features and sales data, an accurate predicted sales volume can be obtained. This method can combine multiple features and error correction mechanisms to improve the stability and accuracy of the forecast.

[0039] In one embodiment, determining the predicted sales volume according to the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period includes: For each feature corresponding to the current time period, determine the change trend of each feature in the corresponding error change curve according to the sales volume corresponding to the current time period, and determine the error change value corresponding to the next time period of the current time period according to the change trend; For each feature, the feature is adjusted according to the error change value corresponding to the feature to obtain an adjusted feature; Substitute all adjusted features into the preset neural network and obtain the predicted sales through forward propagation.

[0040] In one implementation, the error change trend in the previous time period is analyzed. The error change curve shows how the model prediction error changes over time. By comparing the previous predicted sales volume with the actual sales volume, the direction and magnitude of the error change can be identified.

[0041] In one implementation, for each feature, the sales volume in the current time period is first predicted through a preset neural network, and the error is obtained based on the difference between the actual sales volume and the predicted sales volume. This error is the error corresponding to the feature in the current time period. By comparing the historical error (the error in the previous time period) and the current error, the error change rate is obtained by subtracting the error in the previous time period from the current error and dividing it by the error in the previous time period. The corresponding area in the error change curve is searched according to the error change rate, and then the error change rate of the corresponding area in the next time period is calculated and recorded as the error change value.

[0042] In one implementation, the feature is adjusted according to the error change value corresponding to the feature to obtain an adjusted feature. If the error change value increases, the feature is multiplied by 1 minus the error change value; if the error change value decreases, the feature is multiplied by 1 plus the error change value.

[0043] In one embodiment, obtaining resource allocation for each retailer based on the predicted sales of all retailers includes: For each retailer, obtain the retailer's current inventory and predicted sales, and obtain the retailer's required resources based on the current inventory and predicted sales; The resource allocation of each retailer is obtained by allocating resources on demand to each retailer according to the required resource amounts of all retailers and the total resource pool.

[0044] In one implementation, by allocating resources based on each retailer's actual needs and predicted sales, it is possible to avoid waste of resources and ensure that each retailer obtains the required resources, thereby avoiding resource surplus or shortage; the required resource amount for the retailer is obtained by subtracting the current inventory from the predicted sales.

[0045] In one implementation, dynamic adjustments are made based on actual inventory and predicted sales, which can cope with rapidly changing demands in markets such as cross-border e-commerce and improve the flexibility and adaptability of the overall supply chain; the total required resources of all retailers are calculated to obtain the total required resources, and for each retailer, the ratio of the retailer's corresponding required resources to the total required resources is calculated to obtain the retailer's demand ratio, and the retailer's resource volume is obtained by multiplying the retailer's demand ratio by the existing total resources. If the existing total resources are greater than the total required resources, resources are allocated only according to each retailer's demand; if the existing total resources are less than the total required resources, resources are proportionally reduced according to each retailer's resource allocation ratio.

[0046] Based on the same inventive concept, the present invention also provides a supply chain management device for cross-border e-commerce. Figure 2 , Figure 2 A schematic diagram of a supply chain management device for cross-border e-commerce provided by an embodiment of the present invention includes: A first feature set determination module is used to obtain historical transaction data of target commodities in a target retailer, and determine the features of each time period according to the historical transaction data to obtain a first feature set; the target retailer is any retailer that sells the target commodity; A feature sorting module, used to calculate the statistics of each feature and the target product sales volume in each time period according to the first feature set, and sort the statistics from high to low to obtain a feature-related sorting set; An effective feature screening module, used for obtaining a preset number of features in the feature related sorting set to obtain a second feature set, and performing effective feature screening on the second feature set by sequential forward selection to obtain a target feature set; The resource allocation module is used to cluster the features in the target feature set and then perform sales forecasting to obtain predicted sales, obtain resource allocation for each retailer based on the predicted sales of all retailers, and perform supply chain management based on the resource allocation of each retailer.

[0047] A supply chain management device for cross-border e-commerce provided based on an embodiment of the present invention can screen out the features most relevant to sales forecasting through in-depth analysis of different features and a sequential forward selection method, thereby reducing unnecessary noise and errors to make the forecast results more accurate, and then allocating resources based on the forecast results, thereby avoiding inventory backlogs and out-of-stock phenomena caused by uneven resource allocation, and improving the efficiency and operational results of the supply chain.

[0048] In one embodiment, the feature ranking module includes: An influence coefficient determination module is used to calculate the Pearson correlation coefficient and variance analysis between each feature and the target product sales volume in the same time period to obtain a first influence coefficient set and a second influence coefficient set; A coefficient normalization module, used for normalizing the first influence coefficient set and the second influence coefficient set to obtain a first normalized coefficient set and a second normalized coefficient set respectively; The statistic determination module is used to average the first normalization coefficient and the second normalization coefficient corresponding to each feature to obtain the statistic corresponding to the feature.

[0049] In one embodiment, the resource allocation module includes: The first feature sorting and clustering module is used to cluster the target feature set according to each type of feature, and obtain the sorting of the same type of features in the time series to obtain the first feature sorting and clustering; A simulated sales volume determination module is used to input each first feature sorting cluster into a preset neural network for forward propagation to obtain a simulated sales volume corresponding to the feature in each time period in the time series; The error change curve determination module is used to obtain the error coefficient corresponding to the feature in each time period according to the simulated sales volume and the actual sales volume in the time period, and obtain the error coefficient on the time series to obtain the error change curve corresponding to the feature; The predicted sales volume determination module is used to determine the predicted sales volume based on the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period.

[0050] In one embodiment, the predicted sales volume determination module includes: The error change value determination module is used to determine the change trend of each feature in the corresponding error change curve according to the sales volume corresponding to the current time period for each feature corresponding to the current time period, and determine the error change value corresponding to the next time period of the current time period according to the change trend; An adjustment feature determination module is used to adjust each feature according to the error change value corresponding to the feature to obtain an adjustment feature; The predicted sales volume generation module is used to substitute all adjustment features into the preset neural network and obtain the predicted sales volume through forward propagation.

[0051] In one embodiment, the resource allocation module further includes: A required resource quantity determination module is used to obtain the current inventory quantity and predicted sales volume of each retailer, and obtain the required resource quantity of the retailer according to the current inventory quantity and predicted sales volume; The on-demand allocation module is used to perform on-demand resource allocation for each retailer according to the required resource amounts of all retailers and the total resource pool to obtain resource allocation for each retailer.

[0052] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A supply chain management method for cross-border e-commerce, characterized in that: The method comprises: Acquire historical transaction data of a target commodity from a target retailer, and determine the characteristics of each time period according to the historical transaction data to obtain a first feature set; the target retailer is any retailer that sells the target commodity; According to the first feature set, calculate the statistics of each feature and the sales volume of the target product in each time period, and sort the statistics from high to low to obtain a feature-related sorting set; Acquire a preset number of features in the feature-related sorted set to obtain a second feature set, and perform effective feature screening on the second feature set through sequential forward selection to obtain a target feature set; After clustering the features in the target feature set, sales forecasting is performed to obtain predicted sales, resource allocation for each retailer is obtained based on the predicted sales of all retailers, and supply chain management is performed based on the resource allocation for each retailer.

2. A supply chain management method for cross-border e-commerce according to claim 1, characterized in that: The statistics of calculating each feature and the sales volume of the target product in each time period include: Calculate the Pearson correlation coefficient and variance analysis between each feature and the target product sales volume in the same time period to obtain a first influence coefficient set and a second influence coefficient set; Normalizing the first influence coefficient set and the second influence coefficient set respectively to obtain a first normalized coefficient set and a second normalized coefficient set; For each feature, the first normalization coefficient and the second normalization coefficient corresponding to the feature are averaged to obtain the statistic corresponding to the feature.

3. A supply chain management method for cross-border e-commerce according to claim 1, characterized in that: After clustering the features in the target feature set, sales forecasting is performed to obtain the predicted sales volume, including: Clustering the target feature set according to each type of feature, obtaining a ranking of features of the same type in a time series to obtain a first feature ranking cluster; For each first feature sorting cluster, the first feature sorting cluster is input into a preset neural network for forward propagation to obtain a simulated sales volume corresponding to the feature in each time period in the time series; For each time period, the error coefficient corresponding to the feature in the time period is obtained according to the simulated sales volume and the actual sales volume in the time period, and the error coefficient in the time series is obtained to obtain the error change curve corresponding to the feature; The predicted sales volume is determined based on the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period.

4. A supply chain management method for cross-border e-commerce according to claim 3, characterized in that: Determining the predicted sales volume based on the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period includes: For each feature corresponding to the current time period, determine the change trend of each feature in the corresponding error change curve according to the sales volume corresponding to the current time period, and determine the error change value corresponding to the next time period of the current time period according to the change trend; For each feature, the feature is adjusted according to the error change value corresponding to the feature to obtain an adjusted feature; Substitute all adjusted features into the preset neural network and obtain the predicted sales volume through forward propagation.

5. The supply chain management method for cross-border e-commerce according to claim 1, characterized in that: The resource allocation for each retailer based on the forecasted sales of all retailers includes: For each retailer, obtain the current inventory and predicted sales of the retailer, and obtain the required resources of the retailer based on the current inventory and predicted sales; The resource allocation of each retailer is obtained by allocating resources on demand to each retailer according to the required resource amounts of all retailers and the total resource pool.

6. A supply chain management device for cross-border e-commerce, characterized in that: The device comprises: A first feature set determination module is used to obtain historical transaction data of a target commodity from a target retailer, and determine the features of each time period according to the historical transaction data to obtain a first feature set; the target retailer is any retailer that sells the target commodity; A feature sorting module, used to calculate the statistics of each feature and the sales volume of the target product in each time period according to the first feature set, and sort the statistics from high to low to obtain a feature related sorting set; An effective feature screening module, used for obtaining a preset number of features in the feature-related sorting set to obtain a second feature set, and performing effective feature screening on the second feature set by sequential forward selection to obtain a target feature set; The resource allocation module is used to cluster the features in the target feature set and then perform sales forecasting to obtain predicted sales, obtain resource allocation for each retailer based on the predicted sales of all retailers, and perform supply chain management based on the resource allocation for each retailer.

7. A supply chain management device for cross-border e-commerce according to claim 6, characterized in that: The feature sorting module comprises: An influence coefficient determination module, used to calculate the Pearson correlation coefficient and variance analysis between each feature and the target product sales volume in the same time period to obtain a first influence coefficient set and a second influence coefficient set; A coefficient normalization module, used to normalize the first influence coefficient set and the second influence coefficient set respectively to obtain a first normalized coefficient set and a second normalized coefficient set; The statistic determination module is used to average the first normalization coefficient and the second normalization coefficient corresponding to each feature to obtain the statistic corresponding to the feature.

8. A supply chain management device for cross-border e-commerce according to claim 6, characterized in that: The resource allocation module comprises: A first feature sorting and clustering module is used to cluster the target feature set according to each type of feature, and obtain a first feature sorting and clustering by sorting features of the same type in a time series; A simulated sales volume determination module is used to input each first feature sorting cluster into a preset neural network for forward propagation to obtain a simulated sales volume corresponding to the feature in each time period in the time series; The error change curve determination module is used to obtain the error coefficient corresponding to the feature in each time period according to the simulated sales volume and the actual sales volume in the time period, and obtain the error coefficient on the time series to obtain the error change curve corresponding to the feature; The predicted sales volume determination module is used to determine the predicted sales volume based on the error change curves corresponding to all features, each feature corresponding to the current time period, and the sales volume corresponding to the current time period.

9. A supply chain management device for cross-border e-commerce according to claim 8, characterized in that: The predicted sales volume determination module includes: The error change value determination module is used to determine the change trend of each feature in the corresponding error change curve according to the sales volume corresponding to the current time period for each feature corresponding to the current time period, and determine the error change value corresponding to the next time period of the current time period according to the change trend; An adjustment feature determination module is used to adjust each feature according to the error change value corresponding to the feature to obtain an adjustment feature; The predicted sales volume generation module is used to substitute all the adjusted features into the preset neural network and obtain the predicted sales volume through forward propagation.

10. A supply chain management device for cross-border e-commerce according to claim 6, characterized in that: The resource allocation module also includes: A required resource quantity determination module is used to obtain the current inventory quantity and predicted sales volume of each retailer, and obtain the required resource quantity of the retailer according to the current inventory quantity and predicted sales volume; The on-demand allocation module is used to perform on-demand resource allocation for each retailer according to the required resource amounts of all retailers and the total resource pool to obtain resource allocation for each retailer.

Images