Method, system and medium for collaborative optimization of format supply chain based on demand traction

Abstract

This application provides a demand-driven method, system, and medium for collaborative optimization of the supply chain across different business formats. The method includes: acquiring historical demand data for goods and generating a business format collaboration feature vector that integrates the collaborative relationships between different locations; training a time-series prediction model based on the business format collaboration feature vector, outputting the predicted demand for the next few days and the prediction confidence interval; performing dynamic inventory simulation by combining current inventory with future demand forecasts, calculating the stockout risk score and slow-moving inventory risk score for each SKU-store combination; prioritizing based on the risk scores, preferentially utilizing internal regional inventory to generate transfer candidate solutions, and using an optimization model to generate procurement plans for items that cannot be satisfied through transfer. This invention can integrate demand collaboration information between multiple stores, achieving intelligent diversion of dynamic risk warnings and replenishment decisions, effectively reducing the overall supply chain cost.

Description

Technical Field

[0001] This application relates to the fields of supply chain management and intelligent optimization technology, and more specifically, to demand-driven business supply chain collaborative optimization methods, systems, and media. Background Technology

[0002] As entertainment venues such as arcades, children's playgrounds, and cultural and tourism scenic spots gradually expand to a larger scale, the same operating entity often operates multiple locations across regions. This multi-location operation model provides a data foundation and scheduling authority for centralized supply chain management, making it possible to share demand information, coordinate inventory, and conduct unified procurement across locations. However, existing technologies still face the following challenges when serving the supply chain management of such multi-location entertainment venues. First, the demand for entertainment venues is typically characterized by strong seasonality, high volatility, and significant susceptibility to external factors. Holidays, large-scale events, and other factors can lead to drastic fluctuations in customer traffic and product demand, making it difficult for traditional static forecasting methods based on historical averages to accurately capture these dynamic patterns. Second, there is a significant demand linkage effect between different locations under the same operating entity. For example, when a location reaches its capacity limit, customer flow will overflow to nearby similar locations, leading to abnormal demand growth in the latter. Furthermore, multiple locations within the same business district or scenic area are often driven by the same external factors, and their demand changes exhibit a high degree of temporal correlation. Existing supply chain optimization technologies have several shortcomings when applied to the operation and management of multi-venue entertainment businesses: First, traditional demand forecasting methods treat each venue as an independent unit, ignoring the spatial coordination between venues and failing to correct fortification errors at individual venues based on comprehensive operational data. Second, inventory risk warnings rely solely on static assessments of real-time inventory data, neglecting dynamic analysis based on future demand forecasts, leading to delayed or false warnings. Third, replenishment strategies lack quantitative analysis of future demand uncertainties, making it difficult to balance inventory holding costs and stockout losses, and hindering optimal replenishment allocation. Fourth, the lack of an intelligent decision-making system for inventory allocation and centralized procurement prevents the prioritization of idle inventory resources within the same operating entity, resulting in higher overall supply chain costs. Therefore, there is an urgent need for a collaborative supply chain optimization method that integrates multi-venue information, combines dynamic demand forecasting, considers risk quantification, and enables intelligent allocation and procurement. Summary of the Invention

[0003] The purpose of this application is to provide a demand-driven business supply chain collaborative optimization method, system and medium. It constructs a feature vector that integrates store collaborative relationships for demand forecasting, combines dynamic inventory simulation for risk assessment, and generates the optimal replenishment plan based on priority grading and mixed integer programming, so as to solve the problems of insufficient forecast accuracy, delayed risk warning and crude replenishment decision-making in the existing technology.

[0004] This application also provides a demand-driven method for optimizing supply chain collaboration, including the following steps: Acquire historical demand data for goods and generate business format collaboration feature vectors; A time-series prediction model is obtained by training the model based on the feature vector of business format collaboration, and the predicted demand for the next few days and the prediction confidence interval are output based on the time-series prediction model. For each SKU-store combination, calculate the out-of-stock risk score and the slow-moving risk score; Prioritize each SKU-store combination based on out-of-stock risk score and slow-moving risk score; For SKU-store combinations that have not generated transfer candidate plans among high-priority objects, or have generated transfer candidate plans but whose transfer volume is insufficient to cover all out-of-stock quantities, a procurement plan is generated using an optimization model based on the demand forecast for the next few days and the forecast confidence interval.

[0005] Optionally, in the demand-driven business supply chain collaborative optimization method described in this application, the step of acquiring historical demand data for goods and generating a business collaboration feature vector includes: Obtain historical demand data for products, including time, store identifier, SKU, sales quantity, and trigger scenario fields, and generate a demand metadata dataset; The demand metadata is aggregated at a preset time granularity to generate the original demand time series; Extract demand feature vectors from the original demand time series, including historical actual demand volume, time feature data and business scenario feature data within a preset time window; Using a preset length of rolling time window as the statistical unit, the system slides forward step by step. Within each time window, the Pearson correlation coefficient and demand overflow rate between the demand for the same SKU in each store in the region are calculated, generating a collaborative time-series feature vector that changes dynamically over time. The demand feature vector and the collaborative time-series feature vector are concatenated and merged to form the business format collaborative feature vector.

[0006] Optionally, in the demand-driven business supply chain collaborative optimization method described in this application, the step of training a time-series forecast model based on business collaboration feature vectors to obtain a time-series forecast model, and outputting the forecasted demand for multiple future days and the forecast confidence interval based on the time-series forecast model, includes: Using the business format collaboration feature vector as input and the actual demand within a preset future time period as label, a supervised learning training sample set is generated; A time series prediction model is trained using a pre-defined LSTM algorithm based on a supervised learning training sample set. Obtain the business collaboration feature vector for the current preset time period, input it into the time series prediction model, and output the predicted demand for the next few days and the prediction confidence interval.

[0007] Optionally, in the demand-driven business supply chain collaborative optimization method described in this application, the calculation of out-of-stock risk score and slow-moving risk score for each SKU-store combination includes: For each SKU-store combination, obtain the current inventory of the SKU, and calculate the predicted inventory for the next few days based on the current inventory and the demand forecast for the next few days. If the current inventory or the predicted inventory for any future day is less than the preset safety stock threshold, the difference between the safety stock threshold and the minimum predicted inventory is calculated and divided by the safety stock threshold. The stockout risk score is determined based on the preset range to which the obtained ratio belongs. If both the current inventory level and the projected inventory level for the next few days are greater than the maximum inventory threshold, then the difference between the current inventory level and the maximum inventory threshold is calculated and divided by the maximum inventory threshold. The unsold inventory risk score is determined based on the preset range to which the obtained ratio belongs.

[0008] Optionally, in the demand-driven business supply chain collaborative optimization method described in this application, the step of prioritizing each SKU-store combination based on out-of-stock risk score and slow-moving inventory risk score includes: If the stockout risk score of a certain SKU-store combination is greater than the first preset threshold, it is marked as a high-priority object, and it is determined whether the slow-moving risk score of other stores with the same SKU is greater than the second preset threshold. If yes, then generate a transfer candidate plan; otherwise, do not generate a transfer candidate plan. If the slow-moving risk score of a certain SKU-store combination is greater than the second preset threshold, it is marked as a medium priority object, and it is determined whether the stockout risk score of other stores with the same SKU is greater than the first preset threshold. If yes, then generate a transfer candidate plan; otherwise, do not generate a transfer candidate plan. If the stockout risk score of a certain SKU-store combination is less than or equal to the first preset threshold and the slow-moving risk score is less than or equal to the second preset threshold, it is marked as a low-priority object.

[0009] Optionally, in the demand-driven business supply chain collaborative optimization method described in this application, the generation of allocation candidate schemes includes: Stores with a slow-moving inventory risk score greater than the second preset threshold are marked as potential transferees; Stores with out-of-stock risk scores greater than the first preset threshold are marked as potential transferees; Calculate the shortage amount based on the multi-day demand forecast and confidence interval of the potential recipient; Calculate the amount that can be transferred out based on the current inventory of the potential transferor and its own multi-day demand forecast. The smaller value between the out-of-stock quantity and the available transfer quantity is taken as the suggested transfer quantity, and transfer candidate plans are generated, including: SKU, transferring-out store, transferring-in store, and suggested transfer quantity.

[0010] Optionally, in the demand-driven supply chain collaborative optimization method described in this application, the step of generating a procurement plan using an optimization model based on the multi-day demand forecast and prediction confidence interval for SKU-store combinations that have not generated transfer candidate plans for high-priority objects, or have generated transfer candidate plans but whose transfer volume is insufficient to cover all stockouts, includes: Obtain basic parameter data for each SKU in each store to be optimized, including cost parameters, constraint parameters, current inventory level, and transferred quantity; Using basic parameter data and the predicted demand and confidence interval for a future preset period as model inputs, a mixed integer programming model is constructed with the optimization objective of minimizing total cost, which includes procurement cost, warehousing and holding cost, and stockout loss cost. The constraints of the model include: minimum order quantity constraint, supplier delivery cycle constraint, warehouse capacity constraint, and maximum inventory turnover days constraint. The mixed-integer programming model is solved using a mixed-integer programming solver to obtain a suggested procurement plan, including suggested procurement quantity, suggested procurement time, and suggested procurement scheme.

[0011] Secondly, this application provides a demand-driven business supply chain collaborative optimization system, which includes: a memory and a processor. The memory stores a program for a demand-driven business supply chain collaborative optimization method. When the program for the demand-driven business supply chain collaborative optimization method is executed by the processor, it performs the following steps: Acquire historical demand data for goods and generate business format collaboration feature vectors; A time-series prediction model is obtained by training the model based on the feature vector of business format collaboration, and the predicted demand for the next few days and the prediction confidence interval are output based on the time-series prediction model. For each SKU-store combination, calculate the out-of-stock risk score and the slow-moving risk score; Prioritize each SKU-store combination based on out-of-stock risk score and slow-moving risk score; For SKU-store combinations that have not generated transfer candidate plans among high-priority objects, or have generated transfer candidate plans but whose transfer volume is insufficient to cover all out-of-stock quantities, a procurement plan is generated using an optimization model based on the demand forecast for the next few days and the forecast confidence interval.

[0012] Optionally, in the demand-driven business supply chain collaborative optimization system described in this application, the step of acquiring historical demand data for goods and generating a business collaboration feature vector includes: Obtain historical demand data for products, including time, store identifier, SKU, sales quantity, and trigger scenario fields, and generate a demand metadata dataset; The demand metadata is aggregated at a preset time granularity to generate the original demand time series; Extract demand feature vectors from the original demand time series, including historical actual demand volume, time feature data and business scenario feature data within a preset time window; Using a preset length of rolling time window as the statistical unit, the system slides forward step by step. Within each time window, the Pearson correlation coefficient and demand overflow rate between the demand for the same SKU in each store in the region are calculated, generating a collaborative time-series feature vector that changes dynamically over time. The demand feature vector and the collaborative time-series feature vector are concatenated and merged to form the business format collaborative feature vector.

[0013] Optionally, in the demand-driven business supply chain collaborative optimization system described in this application, the step of obtaining a time-series prediction model by training the model based on the business collaboration feature vector, and outputting the predicted demand for multiple future days and the prediction confidence interval based on the time-series prediction model, includes: Using the business format collaboration feature vector as input and the actual demand within a preset future time period as label, a supervised learning training sample set is generated; A time series prediction model is trained using a pre-defined LSTM algorithm based on a supervised learning training sample set. Obtain the business collaboration feature vector for the current preset time period, input it into the time series prediction model, and output the predicted demand for the next few days and the prediction confidence interval.

[0014] Thirdly, this application also provides a computer-readable storage medium storing a demand-driven business supply chain collaborative optimization method program, which, when executed by a processor, implements the steps of the demand-driven business supply chain collaborative optimization method as described in any of the preceding claims.

[0015] As can be seen from the above, the demand-driven business supply chain collaborative optimization method, system and medium provided in this application predicts demand by constructing feature vectors that integrate collaborative relationships between multiple business locations, conducts risk assessment by combining dynamic inventory simulation, and generates the optimal replenishment plan based on priority grading and mixed integer programming, so as to solve the problems of insufficient prediction accuracy, delayed risk warning and crude replenishment decision-making in the prior art.

[0016] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a demand-driven business supply chain collaborative optimization method provided in this application embodiment; Figure 2 The flowchart illustrates the generation of business format collaboration feature vectors using a demand-driven business format supply chain collaboration optimization method provided in this application embodiment. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Please refer to Figure 1 , Figure 1 This is a flowchart of a demand-driven business supply chain collaborative optimization method according to some embodiments of this application. This demand-driven business supply chain collaborative optimization method is used in terminal devices, such as computers and mobile phones. The demand-driven business supply chain collaborative optimization method includes the following steps: S11. Obtain historical demand data for goods and generate a business format collaboration feature vector; S12. Train the model based on the business format synergy feature vector to obtain the time series prediction model, and output the demand forecast and prediction confidence interval for the next few days based on the time series prediction model. S13. For each SKU-store combination, calculate the out-of-stock risk score and the slow-moving risk score; S14. Prioritize each SKU-store combination based on the stockout risk score and the slow-moving risk score. S15. For SKU-store combinations that have not generated transfer candidate plans among high-priority objects, or have generated transfer candidate plans but whose transfer volume is insufficient to cover all out-of-stock quantities, generate a procurement plan based on the multi-day demand forecast and the forecast confidence interval.

[0022] It should be noted that demand forecasting is performed by constructing feature vectors that integrate the collaborative relationships between multiple business locations, risk assessment is conducted by combining dynamic inventory simulation, and an optimal replenishment plan is generated based on priority grading and mixed integer programming. This addresses the problems of insufficient forecasting accuracy, delayed risk warning, and coarse replenishment decision-making in existing technologies. In this embodiment of the invention, SKU (Stock Keeping Unit) refers to the smallest independently manageable inventory unit of a product, such as a specific specification of souvenir, a model of consumable, or a package deal.

[0023] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating the generation of business model collaboration feature vectors using a demand-driven business model supply chain collaboration optimization method as described in some embodiments of this application.

[0024] According to an embodiment of the present invention, the step of acquiring historical demand data for goods and generating a business format collaboration feature vector includes: Obtain historical demand data for products, including time, store identifier, SKU, sales quantity, and trigger scenario fields, and generate a demand metadata dataset; The demand metadata is aggregated at a preset time granularity to generate the original demand time series; Extract demand feature vectors from the original demand time series, including historical actual demand volume, time feature data and business scenario feature data within a preset time window; Using a preset length of rolling time window as the statistical unit, the system slides forward step by step. Within each time window, the Pearson correlation coefficient and demand overflow rate between the demand for the same SKU in each store in the region are calculated, generating a collaborative time-series feature vector that changes dynamically over time. The demand feature vector and the collaborative time-series feature vector are concatenated and merged to form the business format collaborative feature vector.

[0025] It's important to note that traditional entertainment industry supply chain demand forecasting typically treats each store or park as an independent forecasting unit, building forecasting models solely based on historical sales data from individual points. However, in real-world entertainment scenarios, particularly in experiential and visitor-driven sectors like arcades, children's playgrounds, and scenic areas, significant demand linkages exist between stores or parks. For instance, during holidays or peak periods, when arcade A or children's playground A reaches its capacity limit, customers turned away will turn to nearby stores of the same brand or type, causing an abnormal surge in foot traffic and demand for related merchandise (such as game tokens, beverages, and toys) at store B. Similarly, when a particular game machine or children's amusement ride becomes popular in a specific area, it will drive a simultaneous increase in demand for related consumables or merchandise at multiple surrounding stores. In this embodiment of the invention, for any SKU, within a preset rolling time window, the Pearson correlation coefficient between the demand for that SKU among stores within the region is calculated. The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between the sales quantity sequences of two stores (for the same SKU) within the same time window, reflecting the synchronicity of demand changes between stores (e.g., whether sales will surge simultaneously due to holidays). Simultaneously, the demand overflow rate is calculated, which refers to the extent to which unmet demand is transferred (overflowed) to other stores when a store experiences stockouts. It is calculated as follows: when the inventory fulfillment rate of the first store is lower than a preset threshold (i.e., stockouts or customer overflow occur), the sales increment of the second store within the same time window is the ratio of this overflow amount (calculated by taking the median of historical sales volume in the same scenario where the historical inventory fulfillment rate is greater than 95% as the historical expected demand, and calculating the difference between the historical expected demand and the actual sales quantity of the store). These two indicators are dynamically updated as the time window slides, thus forming a collaborative temporal feature vector that reflects regional collaborative relationships.

[0026] By concatenating and fusing the collaborative time-series feature vector with the demand feature vector based on historical data of a single store, the business format collaborative feature vector constructed in this invention enables the time-series prediction model to simultaneously consider both "single-store operational patterns" and "regional customer flow linkage effects," effectively improving the accuracy of demand prediction in complex scenarios within the entertainment industry and providing more reliable data support for subsequent intelligent procurement and allocation decisions. The time-series feature data includes demand dates and holiday identifiers, while the business scenario feature data includes the number of promotions effective on that day, discount levels, average equipment utilization, and the number of members visiting the store.

[0027] According to an embodiment of the present invention, the step of obtaining a time-series prediction model by training a model based on business format synergy feature vectors, and outputting the predicted demand for multiple future days and the prediction confidence interval based on the time-series prediction model, includes: Using the business format collaboration feature vector as input and the actual demand within a preset future time period as label, a supervised learning training sample set is generated; A time series prediction model is trained using a pre-defined LSTM algorithm based on a supervised learning training sample set. Obtain the business collaboration feature vector for the current preset time period, input it into the time series prediction model, and output the predicted demand for the next few days and the prediction confidence interval.

[0028] It should be noted that, in this embodiment of the invention, the construction of the time-series prediction model is based on a supervised learning framework. Specifically, the model input is a business format collaboration feature vector, and the actual demand within a preset future time period is used as the training label, thereby forming a supervised learning training sample set. Compared to traditional methods that only use historical data from a single store, this invention incorporates the collaborative temporal features reflecting the collaborative relationship between stores into the input, enabling the model to learn the overall regional demand change pattern. The model is trained using a Long Short-Term Memory (LSTM) network, which can effectively capture long-term dependencies and short-term fluctuations in time-series data. This effectively addresses business format demand prediction scenarios with significant temporality, nonlinearity, and the influence of external factors (such as holidays and promotional activities). Meanwhile, the model output includes demand forecasts for the next few days (point forecasts) and prediction confidence intervals. The prediction confidence intervals are generated using the Monte Carlo Dropout method. Specifically, the Dropout layer is retained during the inference phase, multiple forward propagations are performed, and the quantiles of the multiple outputs are used as the upper and lower bounds of the confidence interval. The prediction confidence intervals reflect the model's assessment of the uncertainty of the prediction results, enabling subsequent procurement decisions to be dynamically adjusted based on risk assessment. This achieves a better balance between reducing inventory holding costs and avoiding stockout losses, realizing refined and intelligent management of the supply chain.

[0029] According to an embodiment of the present invention, calculating the out-of-stock risk score and the slow-moving stock risk score for each SKU-store combination includes: For each SKU-store combination, obtain the current inventory of the SKU, and calculate the predicted inventory for the next few days based on the current inventory and the demand forecast for the next few days. If the current inventory or the predicted inventory for any future day is less than the preset safety stock threshold, the difference between the safety stock threshold and the minimum predicted inventory is calculated and divided by the safety stock threshold. The stockout risk score is determined based on the preset range to which the obtained ratio belongs. If both the current inventory level and the projected inventory level for the next few days are greater than the maximum inventory threshold, then the difference between the current inventory level and the maximum inventory threshold is calculated and divided by the maximum inventory threshold. The unsold inventory risk score is determined based on the preset range to which the obtained ratio belongs.

[0030] It should be noted that for each SKU-store combination, the current inventory level of the SKU is obtained. Based on the current inventory level and the output multi-day demand forecast, the predicted inventory level for each future time point is calculated daily. Specifically, the predicted inventory level for day t = current inventory level - sum of demand forecasts from day 1 to day t. If the current inventory level or the predicted inventory level for any future day is less than the preset safety stock threshold, the difference between the safety stock threshold and the minimum predicted inventory level is calculated and divided by the safety stock threshold. The stockout risk score is determined based on the preset interval to which the resulting ratio belongs. The minimum predicted inventory level is the minimum value among the current inventory level and all future predicted inventory levels. The preset interval can be set as follows: a ratio within (0, 0.2] scores 1 point, within (0.2, 0.5] scores 2 points, within (0.5, 1.0] scores 3 points, and greater than 1.0 scores 4 points. A higher score indicates a more severe stockout risk. Furthermore, if the current inventory level is between the safety stock threshold and the maximum inventory threshold, both the stockout risk score and the slow-moving inventory risk score are 0.

[0031] According to an embodiment of the present invention, the prioritization of each SKU-store combination based on out-of-stock risk score and slow-moving inventory risk score includes: If the stockout risk score of a certain SKU-store combination is greater than the first preset threshold, it is marked as a high-priority object, and it is determined whether the slow-moving risk score of other stores with the same SKU is greater than the second preset threshold. If yes, then generate a transfer candidate plan; otherwise, do not generate a transfer candidate plan. If the slow-moving risk score of a certain SKU-store combination is greater than the second preset threshold, it is marked as a medium priority object, and it is determined whether the stockout risk score of other stores with the same SKU is greater than the first preset threshold. If yes, then generate a transfer candidate plan; otherwise, do not generate a transfer candidate plan. If the stockout risk score of a certain SKU-store combination is less than or equal to the first preset threshold and the slow-moving risk score is less than or equal to the second preset threshold, it is marked as a low-priority object.

[0032] It should be noted that by using a multi-level priority grading mechanism based on stockout risk scores and slow-moving inventory risk scores, SKU-store combinations are divided into high priority, medium priority, and low priority, enabling differentiated handling of inventory issues with different levels of urgency. For high-priority stockout items, the system proactively determines whether there are available sources of slow-moving inventory that can be allocated, achieving intelligent diversion of allocation and procurement. This ensures that inventory resources within the chain network are utilized preferentially, avoids unnecessary procurement expenditures, and effectively reduces overall supply chain costs.

[0033] According to an embodiment of the present invention, generating the allocation candidate scheme includes: Stores with a slow-moving inventory risk score greater than the second preset threshold are marked as potential transferees; Stores with out-of-stock risk scores greater than the first preset threshold are marked as potential transferees; Calculate the shortage amount based on the multi-day demand forecast and confidence interval of the potential recipient; Calculate the amount that can be transferred out based on the current inventory of the potential transferor and its own multi-day demand forecast. The smaller value between the out-of-stock quantity and the available transfer quantity is taken as the suggested transfer quantity, and transfer candidate plans are generated, including: SKU, transferring-out store, transferring-in store, and suggested transfer quantity.

[0034] It should be noted that by combining the predicted confidence interval to calculate the stockout amount for the receiving store, the uncertainty of future demand is fully considered, giving the allocation decision a risk buffer capability. Specifically, for the receiving store (stockout store), the upper limit of the confidence interval of the daily demand forecast is taken as a conservative demand estimate. The cumulative demand is calculated daily and compared with the current inventory and in-transit inventory. If the inventory balance on a certain day is less than zero, the gap for that day is the stockout amount. The sum of the stockout amounts over the next few days is the total stockout amount for that store. The wider the confidence interval and the higher the upper limit, the more conservative the calculated stockout amount, and the more sufficient the allocation amount. At the same time, the smaller value between the stockout amount and the available allocation amount is used as the suggested allocation amount, which not only meets the demand gap of the receiving store, but also ensures the future supply of the sending store, avoiding the risk of new stockouts after the allocation.

[0035] According to an embodiment of the present invention, for SKU-store combinations of high-priority objects for which no transfer candidate plans have been generated, or for which transfer candidate plans have been generated but the transfer volume is insufficient to cover all stockouts, a procurement plan is generated using an optimization model based on multi-day demand forecasts and forecast confidence intervals, including: Obtain basic parameter data for each SKU in each store to be optimized, including cost parameters, constraint parameters, current inventory level, and transferred quantity; Using basic parameter data and the predicted demand and confidence interval for a future preset period as model inputs, a mixed integer programming model is constructed with the optimization objective of minimizing total cost, which includes procurement cost, warehousing and holding cost, and stockout loss cost. The constraints of the model include: minimum order quantity constraint, supplier delivery cycle constraint, warehouse capacity constraint, and maximum inventory turnover days constraint. The mixed-integer programming model is solved using a mixed-integer programming solver to obtain a suggested procurement plan, including suggested procurement quantity, suggested procurement time, and suggested procurement scheme.

[0036] It should be noted that by constructing a mixed-integer programming model with the objective of minimizing total cost, a balance is comprehensively considered among procurement costs, warehousing and holding costs, and stockout loss costs. Multiple constraints are introduced, including minimum order quantity (i.e., a single purchase quantity not less than the supplier's minimum order quantity), supplier delivery cycle (i.e., the procurement arrival time not earlier than the supplier's delivery cycle), warehouse capacity (i.e., the inventory at the end of each period does not exceed the warehouse capacity limit), and inventory turnover days. This ensures that the procurement plan achieves optimal cost while meeting actual business constraints. Furthermore, by using the forecast confidence interval as model input, the uncertainty of future demand is quantified, giving procurement decisions a risk buffer and avoiding over- or under-purchasing problems caused by demand fluctuations.

[0037] According to an embodiment of the present invention, it further includes: If the trigger scenario field in the demand metadata set is displayed as a promotion scenario, then extract the historical demand records containing the promotion identifier field from the demand metadata set to construct the promotion event dataset; The promotion event dataset includes promotion type, discount level, participating SKUs, and promotion start and end time; The SKU-store combination that participated in the promotion was marked as the experimental group, and the same SKU in the same store that did not participate in the same promotion during the same period was marked as the control group. Based on the changes in demand before and after the promotion in the experimental and control groups, the double difference method was used to calculate the promotion elasticity coefficient of each SKU under each promotion type in each preset discount range, and a three-dimensional promotion elasticity coefficient matrix with promotion type, SKU and discount range as dimensions was generated. Obtain the promotion type and discount level corresponding to the SKU currently participating in the promotion, map the discount level to the corresponding discount level range, and find the corresponding promotion elasticity coefficient in the three-dimensional promotion elasticity coefficient matrix. Multiply the multi-day demand forecast output by the time series forecast model by the promotion elasticity coefficient to obtain the expected demand increment. The expected increase in demand is superimposed on the multi-day demand forecast output by the time-series forecast model to generate the final demand forecast after adjustment for promotional effects.

[0038] It should be noted that, in a specific embodiment of the present invention, when the trigger scenario field in the demand metadata set displays a promotion scenario, a promotion effect correction process is executed to eliminate the confusing impact of promotional activities on demand forecasting. First, all records containing promotional identifiers are extracted from historical demand data to construct a promotional event dataset, which includes fields such as promotion type, discount level, participating SKUs, and promotion start and end times. Then, the net causal effect of the promotional activity is estimated using the difference-in-differences method: the SKU-store combination participating in the promotion is labeled as the experimental group, and the same SKUs in the same type of stores that did not participate in the same type of promotion during the same period are labeled as the control group. The change in demand before and after the promotional activity is calculated for both groups, and the difference between the two is taken as the net promotional effect value of that SKU under the promotional conditions, i.e., the promotional elasticity coefficient. To differentiate the varying impacts of different promotional incentive intensities on demand increments, discount levels are discretized into several intervals according to preset rules (e.g., [0, 20%), [20%, 40%), [40%, 60%), [60%, 80%), [80%, 100%]). For each SKU under each promotional type, a promotional elasticity coefficient is calculated for each discount level interval. This results in a three-dimensional promotional elasticity coefficient matrix with "promotional type × SKU × discount level interval" as the three-dimensional index and the corresponding promotional elasticity coefficient stored at each intersection. During the forecasting phase, the promotional type and discount level of the currently participating SKU are obtained, the discount level is mapped to the corresponding interval, and the promotional elasticity coefficient for that SKU is found in the three-dimensional matrix. Multiplying this coefficient by the basic demand forecast output by the time-series forecasting model yields the expected demand increment. This increment is then added to the basic forecast to generate the final demand forecast after adjustment for the promotional effect. This invention can separate and quantify the net causal contribution of promotions to demand from historical data, effectively improving the accuracy of demand forecasting in promotional scenarios, thus providing a more reliable data foundation for subsequent replenishment decisions.

[0039] This invention also discloses a demand-driven business supply chain collaborative optimization system, including a memory and a processor. The memory stores a demand-driven business supply chain collaborative optimization method program, which, when executed by the processor, performs the following steps: Acquire historical demand data for goods and generate business format collaboration feature vectors; A time-series prediction model is obtained by training the model based on the feature vector of business format collaboration, and the predicted demand for the next few days and the prediction confidence interval are output based on the time-series prediction model. For each SKU-store combination, calculate the out-of-stock risk score and the slow-moving risk score; Prioritize each SKU-store combination based on out-of-stock risk score and slow-moving risk score; For SKU-store combinations that have not generated transfer candidate plans among high-priority objects, or have generated transfer candidate plans but whose transfer volume is insufficient to cover all out-of-stock quantities, a procurement plan is generated using an optimization model based on the demand forecast for the next few days and the forecast confidence interval.

[0040] It should be noted that demand forecasting is performed by constructing feature vectors that integrate the collaborative relationships between multiple business locations, risk assessment is conducted by combining dynamic inventory simulation, and an optimal replenishment plan is generated based on priority grading and mixed integer programming. This addresses the problems of insufficient forecasting accuracy, delayed risk warning, and coarse replenishment decision-making in existing technologies. In this embodiment of the invention, SKU (Stock Keeping Unit) refers to the smallest independently manageable inventory unit of a product, such as a specific specification of souvenir, a model of consumable, or a package deal.

[0041] According to an embodiment of the present invention, the step of acquiring historical demand data for goods and generating a business format collaboration feature vector includes: Obtain historical demand data for products, including time, store identifier, SKU, sales quantity, and trigger scenario fields, and generate a demand metadata dataset; The demand metadata is aggregated at a preset time granularity to generate the original demand time series; Extract demand feature vectors from the original demand time series, including historical actual demand volume, time feature data and business scenario feature data within a preset time window; Using a preset length of rolling time window as the statistical unit, the system slides forward step by step. Within each time window, the Pearson correlation coefficient and demand overflow rate between the demand for the same SKU in each store in the region are calculated, generating a collaborative time-series feature vector that changes dynamically over time. The demand feature vector and the collaborative time-series feature vector are concatenated and merged to form the business format collaborative feature vector.

[0042] It's important to note that traditional entertainment industry supply chain demand forecasting typically treats each store or park as an independent forecasting unit, building forecasting models solely based on historical sales data from individual points. However, in real-world entertainment scenarios, particularly in experiential and visitor-driven sectors like arcades, children's playgrounds, and scenic areas, significant demand linkages exist between stores or parks. For instance, during holidays or peak periods, when arcade A or children's playground A reaches its capacity limit, customers turned away will turn to nearby stores of the same brand or type, causing an abnormal surge in foot traffic and demand for related merchandise (such as game tokens, beverages, and toys) at store B. Similarly, when a particular game machine or children's amusement ride becomes popular in a specific area, it will drive a simultaneous increase in demand for related consumables or merchandise at multiple surrounding stores. In this embodiment of the invention, for any SKU, within a preset rolling time window, the Pearson correlation coefficient between the demand for that SKU among stores within the region is calculated. The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between the sales quantity sequences of two stores (for the same SKU) within the same time window, reflecting the synchronicity of demand changes between stores (e.g., whether sales will surge simultaneously due to holidays). Simultaneously, the demand overflow rate is calculated, which refers to the extent to which unmet demand is transferred (overflowed) to other stores when a store experiences stockouts. It is calculated as follows: when the inventory fulfillment rate of the first store is lower than a preset threshold (i.e., stockouts or customer overflow occur), the sales increment of the second store within the same time window is the ratio of this overflow amount (calculated by taking the median of historical sales volume in the same scenario where the historical inventory fulfillment rate is greater than 95% as the historical expected demand, and calculating the difference between the historical expected demand and the actual sales quantity of the store). These two indicators are dynamically updated as the time window slides, thus forming a collaborative temporal feature vector that reflects regional collaborative relationships.

[0043] By concatenating and fusing the collaborative time-series feature vector with the demand feature vector based on historical data of a single store, the business format collaborative feature vector constructed in this invention enables the time-series prediction model to simultaneously consider both "single-store operational patterns" and "regional customer flow linkage effects," effectively improving the accuracy of demand prediction in complex scenarios within the entertainment industry and providing more reliable data support for subsequent intelligent procurement and allocation decisions. The time-series feature data includes demand dates and holiday identifiers, while the business scenario feature data includes the number of promotions effective on that day, discount levels, average equipment utilization, and the number of members visiting the store.

[0044] According to an embodiment of the present invention, the step of obtaining a time-series prediction model by training a model based on business format synergy feature vectors, and outputting the predicted demand for multiple future days and the prediction confidence interval based on the time-series prediction model, includes: Using the business format collaboration feature vector as input and the actual demand within a preset future time period as label, a supervised learning training sample set is generated; A time series prediction model is trained using a pre-defined LSTM algorithm based on a supervised learning training sample set. Obtain the business collaboration feature vector for the current preset time period, input it into the time series prediction model, and output the predicted demand for the next few days and the prediction confidence interval.

[0045] It should be noted that, in this embodiment of the invention, the construction of the time-series prediction model is based on a supervised learning framework. Specifically, the model input is a business format collaboration feature vector, and the actual demand within a preset future time period is used as the training label, thereby forming a supervised learning training sample set. Compared to traditional methods that only use historical data from a single store, this invention incorporates the collaborative temporal features reflecting the collaborative relationship between stores into the input, enabling the model to learn the overall regional demand change pattern. The model is trained using a Long Short-Term Memory (LSTM) network, which can effectively capture long-term dependencies and short-term fluctuations in time-series data. This effectively addresses business format demand prediction scenarios with significant temporality, nonlinearity, and the influence of external factors (such as holidays and promotional activities). Meanwhile, the model output includes demand forecasts for the next few days (point forecasts) and prediction confidence intervals. The prediction confidence intervals are generated using the Monte Carlo Dropout method. Specifically, the Dropout layer is retained during the inference phase, multiple forward propagations are performed, and the quantiles of the multiple outputs are used as the upper and lower bounds of the confidence interval. The prediction confidence intervals reflect the model's assessment of the uncertainty of the prediction results, enabling subsequent procurement decisions to be dynamically adjusted based on risk assessment. This achieves a better balance between reducing inventory holding costs and avoiding stockout losses, realizing refined and intelligent management of the supply chain.

[0046] According to an embodiment of the present invention, calculating the out-of-stock risk score and the slow-moving stock risk score for each SKU-store combination includes: For each SKU-store combination, obtain the current inventory of the SKU, and calculate the predicted inventory for the next few days based on the current inventory and the demand forecast for the next few days. If the current inventory or the predicted inventory for any future day is less than the preset safety stock threshold, the difference between the safety stock threshold and the minimum predicted inventory is calculated and divided by the safety stock threshold. The stockout risk score is determined based on the preset range to which the obtained ratio belongs. If both the current inventory level and the projected inventory level for the next few days are greater than the maximum inventory threshold, then the difference between the current inventory level and the maximum inventory threshold is calculated and divided by the maximum inventory threshold. The unsold inventory risk score is determined based on the preset range to which the obtained ratio belongs.

[0047] It should be noted that for each SKU-store combination, the current inventory level of the SKU is obtained. Based on the current inventory level and the output multi-day demand forecast, the predicted inventory level for each future time point is calculated daily. Specifically, the predicted inventory level for day t = current inventory level - sum of demand forecasts from day 1 to day t. If the current inventory level or the predicted inventory level for any future day is less than the preset safety stock threshold, the difference between the safety stock threshold and the minimum predicted inventory level is calculated and divided by the safety stock threshold. The stockout risk score is determined based on the preset interval to which the resulting ratio belongs. The minimum predicted inventory level is the minimum value among the current inventory level and all future predicted inventory levels. The preset interval can be set as follows: a ratio within (0, 0.2] scores 1 point, within (0.2, 0.5] scores 2 points, within (0.5, 1.0] scores 3 points, and greater than 1.0 scores 4 points. A higher score indicates a more severe stockout risk. Furthermore, if the current inventory level is between the safety stock threshold and the maximum inventory threshold, both the stockout risk score and the slow-moving inventory risk score are 0.

[0048] According to an embodiment of the present invention, the prioritization of each SKU-store combination based on out-of-stock risk score and slow-moving inventory risk score includes: If the stockout risk score of a certain SKU-store combination is greater than the first preset threshold, it is marked as a high-priority object, and it is determined whether the slow-moving risk score of other stores with the same SKU is greater than the second preset threshold. If yes, then generate a transfer candidate plan; otherwise, do not generate a transfer candidate plan. If the slow-moving risk score of a certain SKU-store combination is greater than the second preset threshold, it is marked as a medium priority object, and it is determined whether the stockout risk score of other stores with the same SKU is greater than the first preset threshold. If yes, then generate a transfer candidate plan; otherwise, do not generate a transfer candidate plan. If the stockout risk score of a certain SKU-store combination is less than or equal to the first preset threshold and the slow-moving risk score is less than or equal to the second preset threshold, it is marked as a low-priority object.

[0049] It should be noted that by using a multi-level priority grading mechanism based on stockout risk scores and slow-moving inventory risk scores, SKU-store combinations are divided into high priority, medium priority, and low priority, enabling differentiated handling of inventory issues with different levels of urgency. For high-priority stockout items, the system proactively determines whether there are available sources of slow-moving inventory that can be allocated, achieving intelligent diversion of allocation and procurement. This ensures that inventory resources within the chain network are utilized preferentially, avoids unnecessary procurement expenditures, and effectively reduces overall supply chain costs.

[0050] According to an embodiment of the present invention, generating the allocation candidate scheme includes: Stores with a slow-moving inventory risk score greater than the second preset threshold are marked as potential transferees; Stores with out-of-stock risk scores greater than the first preset threshold are marked as potential transferees; Calculate the shortage amount based on the multi-day demand forecast and confidence interval of the potential recipient; Calculate the amount that can be transferred out based on the current inventory of the potential transferor and its own multi-day demand forecast. The smaller value between the out-of-stock quantity and the available transfer quantity is taken as the suggested transfer quantity, and transfer candidate plans are generated, including: SKU, transferring-out store, transferring-in store, and suggested transfer quantity.

[0051] It should be noted that by combining the predicted confidence interval to calculate the stockout amount for the receiving store, the uncertainty of future demand is fully considered, giving the allocation decision a risk buffer capability. Specifically, for the receiving store (stockout store), the upper limit of the confidence interval of the daily demand forecast is taken as a conservative demand estimate. The cumulative demand is calculated daily and compared with the current inventory and in-transit inventory. If the inventory balance on a certain day is less than zero, the gap for that day is the stockout amount. The sum of the stockout amounts over the next few days is the total stockout amount for that store. The wider the confidence interval and the higher the upper limit, the more conservative the calculated stockout amount, and the more sufficient the allocation amount. At the same time, the smaller value between the stockout amount and the available allocation amount is used as the suggested allocation amount, which not only meets the demand gap of the receiving store, but also ensures the future supply of the sending store, avoiding the risk of new stockouts after the allocation.

[0052] According to an embodiment of the present invention, for SKU-store combinations of high-priority objects for which no transfer candidate plans have been generated, or for which transfer candidate plans have been generated but the transfer volume is insufficient to cover all stockouts, a procurement plan is generated using an optimization model based on multi-day demand forecasts and forecast confidence intervals, including: Obtain basic parameter data for each SKU in each store to be optimized, including cost parameters, constraint parameters, current inventory level, and transferred quantity; Using basic parameter data and the predicted demand and confidence interval for a future preset period as model inputs, a mixed integer programming model is constructed with the optimization objective of minimizing total cost, which includes procurement cost, warehousing and holding cost, and stockout loss cost. The constraints of the model include: minimum order quantity constraint, supplier delivery cycle constraint, warehouse capacity constraint, and maximum inventory turnover days constraint. The mixed-integer programming model is solved using a mixed-integer programming solver to obtain a suggested procurement plan, including suggested procurement quantity, suggested procurement time, and suggested procurement scheme.

[0053] It should be noted that by constructing a mixed-integer programming model with the objective of minimizing total cost, a balance is comprehensively considered among procurement costs, warehousing and holding costs, and stockout loss costs. Multiple constraints are introduced, including minimum order quantity (i.e., a single purchase quantity not less than the supplier's minimum order quantity), supplier delivery cycle (i.e., the procurement arrival time not earlier than the supplier's delivery cycle), warehouse capacity (i.e., the inventory at the end of each period does not exceed the warehouse capacity limit), and inventory turnover days. This ensures that the procurement plan achieves optimal cost while meeting actual business constraints. Furthermore, by using the forecast confidence interval as model input, the uncertainty of future demand is quantified, giving procurement decisions a risk buffer and avoiding over- or under-purchasing problems caused by demand fluctuations.

[0054] According to an embodiment of the present invention, it further includes: If the trigger scenario field in the demand metadata set is displayed as a promotion scenario, then extract the historical demand records containing the promotion identifier field from the demand metadata set to construct the promotion event dataset; The promotion event dataset includes promotion type, discount level, participating SKUs, and promotion start and end time; The SKU-store combination that participated in the promotion was marked as the experimental group, and the same SKU in the same store that did not participate in the same promotion during the same period was marked as the control group. Based on the changes in demand before and after the promotion in the experimental and control groups, the double difference method was used to calculate the promotion elasticity coefficient of each SKU under each promotion type in each preset discount range, and a three-dimensional promotion elasticity coefficient matrix with promotion type, SKU and discount range as dimensions was generated. Obtain the promotion type and discount level corresponding to the SKU currently participating in the promotion, map the discount level to the corresponding discount level range, and find the corresponding promotion elasticity coefficient in the three-dimensional promotion elasticity coefficient matrix. Multiply the multi-day demand forecast output by the time series forecast model by the promotion elasticity coefficient to obtain the expected demand increment. The expected increase in demand is superimposed on the multi-day demand forecast output by the time-series forecast model to generate the final demand forecast after adjustment for promotional effects.

[0055] It should be noted that, in a specific embodiment of the present invention, when the trigger scenario field in the demand metadata set displays a promotion scenario, a promotion effect correction process is executed to eliminate the confusing impact of promotional activities on demand forecasting. First, all records containing promotional identifiers are extracted from historical demand data to construct a promotional event dataset, which includes fields such as promotion type, discount level, participating SKUs, and promotion start and end times. Then, the net causal effect of the promotional activity is estimated using the difference-in-differences method: the SKU-store combination participating in the promotion is labeled as the experimental group, and the same SKUs in the same type of stores that did not participate in the same type of promotion during the same period are labeled as the control group. The change in demand before and after the promotional activity is calculated for both groups, and the difference between the two is taken as the net promotional effect value of that SKU under the promotional conditions, i.e., the promotional elasticity coefficient. To differentiate the varying impacts of different promotional incentive intensities on demand increments, discount levels are discretized into several intervals according to preset rules (e.g., [0, 20%), [20%, 40%), [40%, 60%), [60%, 80%), [80%, 100%]). For each SKU under each promotional type, a promotional elasticity coefficient is calculated for each discount level interval. This results in a three-dimensional promotional elasticity coefficient matrix with "promotional type × SKU × discount level interval" as the three-dimensional index and the corresponding promotional elasticity coefficient stored at each intersection. During the forecasting phase, the promotional type and discount level of the currently participating SKU are obtained, the discount level is mapped to the corresponding interval, and the promotional elasticity coefficient for that SKU is found in the three-dimensional matrix. Multiplying this coefficient by the basic demand forecast output by the time-series forecasting model yields the expected demand increment. This increment is then added to the basic forecast to generate the final demand forecast after adjustment for the promotional effect. This invention can separate and quantify the net causal contribution of promotions to demand from historical data, effectively improving the accuracy of demand forecasting in promotional scenarios, thus providing a more reliable data foundation for subsequent replenishment decisions.

[0056] A third aspect of the present invention provides a readable storage medium storing a demand-driven business supply chain collaborative optimization method program, wherein when the demand-driven business supply chain collaborative optimization method program is executed by a processor, it implements the steps of the demand-driven business supply chain collaborative optimization method as described in any of the preceding claims.

[0057] The present invention discloses a demand-driven business supply chain collaborative optimization method, system and medium, which forecasts demand by constructing feature vectors that integrate collaborative relationships among multiple business locations, conducts risk assessment by combining dynamic inventory simulation, and generates the optimal replenishment plan based on priority grading and mixed integer programming, so as to solve the problems of insufficient forecast accuracy, delayed risk warning and crude replenishment decision-making in the prior art.

[0058] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0059] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0060] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0061] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0062] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A demand-driven business supply chain collaborative optimization method, characterized in that, Includes the following steps: Acquire historical demand data for goods and generate business format collaboration feature vectors; A time-series prediction model is obtained by training the model based on the feature vector of business format collaboration, and the predicted demand for the next few days and the prediction confidence interval are output based on the time-series prediction model. For each SKU-store combination, calculate the out-of-stock risk score and the slow-moving risk score; Prioritize each SKU-store combination based on out-of-stock risk score and slow-moving risk score; For SKU-store combinations that have not generated transfer candidate plans among high-priority objects, or have generated transfer candidate plans but whose transfer volume is insufficient to cover all out-of-stock quantities, a procurement plan is generated using an optimization model based on the demand forecast for the next few days and the forecast confidence interval.

2. The demand-driven business supply chain collaborative optimization method according to claim 1, characterized in that, The process of acquiring historical demand data for goods and generating a business format collaboration feature vector includes: Obtain historical demand data for products, including time, store identifier, SKU, sales quantity, and trigger scenario fields, and generate a demand metadata dataset; The demand metadata is aggregated at a preset time granularity to generate the original demand time series; Extract demand feature vectors from the original demand time series, including historical actual demand volume, time feature data and business scenario feature data within a preset time window; Using a preset length of rolling time window as the statistical unit, the system slides forward step by step. Within each time window, the Pearson correlation coefficient and demand overflow rate between the demand for the same SKU in each store in the region are calculated, generating a collaborative time-series feature vector that changes dynamically over time. The demand feature vector and the collaborative time-series feature vector are concatenated and merged to form the business format collaborative feature vector.

3. The demand-driven business supply chain collaborative optimization method according to claim 2, characterized in that, The time-series prediction model is obtained by training the model based on the business format synergy feature vector. Based on the time-series prediction model, the predicted demand for the next few days and the prediction confidence interval are output, including: Using the business format collaboration feature vector as input and the actual demand within a preset future time period as label, a supervised learning training sample set is generated; A time series prediction model is trained using a pre-defined LSTM algorithm based on a supervised learning training sample set. Obtain the business collaboration feature vector for the current preset time period, input it into the time series prediction model, and output the predicted demand for the next few days and the prediction confidence interval.

4. The demand-driven business supply chain collaborative optimization method according to claim 3, characterized in that, For each SKU-store combination, the calculation of out-of-stock risk score and slow-moving risk score includes: For each SKU-store combination, obtain the current inventory of the SKU, and calculate the predicted inventory for the next few days based on the current inventory and the demand forecast for the next few days. If the current inventory or the predicted inventory for any future day is less than the preset safety stock threshold, the difference between the safety stock threshold and the minimum predicted inventory is calculated and divided by the safety stock threshold. The stockout risk score is determined based on the preset range to which the obtained ratio belongs. If both the current inventory level and the projected inventory level for the next few days are greater than the maximum inventory threshold, then the difference between the current inventory level and the maximum inventory threshold is calculated and divided by the maximum inventory threshold. The unsold inventory risk score is determined based on the preset range to which the obtained ratio belongs.

5. The demand-driven business supply chain collaborative optimization method according to claim 4, characterized in that, The prioritization of each SKU-store combination based on out-of-stock risk score and slow-moving inventory risk score includes: If the stockout risk score of a certain SKU-store combination is greater than the first preset threshold, it is marked as a high-priority object, and it is determined whether the slow-moving risk score of other stores with the same SKU is greater than the second preset threshold. If yes, then generate a transfer candidate plan; otherwise, do not generate a transfer candidate plan. If the slow-moving risk score of a certain SKU-store combination is greater than the second preset threshold, it is marked as a medium priority object, and it is determined whether the stockout risk score of other stores with the same SKU is greater than the first preset threshold. If yes, then generate a transfer candidate plan; otherwise, do not generate a transfer candidate plan. If the stockout risk score of a certain SKU-store combination is less than or equal to the first preset threshold and the slow-moving risk score is less than or equal to the second preset threshold, it is marked as a low-priority object.

6. The demand-driven business supply chain collaborative optimization method according to claim 5, characterized in that, The generation of allocation candidate schemes includes: Stores with a slow-moving inventory risk score greater than the second preset threshold are marked as potential transferees; Stores with out-of-stock risk scores greater than the first preset threshold are marked as potential transferees; Calculate the shortage amount based on the multi-day demand forecast and confidence interval of the potential recipient; Calculate the amount that can be transferred out based on the current inventory of the potential transferor and its own multi-day demand forecast. The smaller value between the out-of-stock quantity and the available transfer quantity is taken as the suggested transfer quantity, and transfer candidate plans are generated, including: SKU, transferring-out store, transferring-in store, and suggested transfer quantity.

7. The demand-driven business supply chain collaborative optimization method according to claim 6, characterized in that, For SKU-store combinations where no transfer candidate plans have been generated for high-priority objects, or where transfer candidate plans have been generated but the transfer volume is insufficient to cover all stockouts, a procurement plan is generated using an optimization model based on multi-day demand forecasts and forecast confidence intervals, including: Obtain basic parameter data for each SKU in each store to be optimized, including cost parameters, constraint parameters, current inventory level, and transferred quantity; Using basic parameter data and the predicted demand and confidence interval for a future preset period as model inputs, a mixed integer programming model is constructed with the optimization objective of minimizing total cost, which includes procurement cost, warehousing and holding cost, and stockout loss cost. The constraints of the model include: minimum order quantity constraint, supplier delivery cycle constraint, warehouse capacity constraint, and maximum inventory turnover days constraint. The mixed-integer programming model is solved using a mixed-integer programming solver to obtain a suggested procurement plan, including suggested procurement quantity, suggested procurement time, and suggested procurement scheme.

8. A demand-driven business supply chain collaborative optimization system, characterized in that: The system includes a memory and a processor. The memory stores a program for a demand-driven business supply chain collaborative optimization method. When the program for the demand-driven business supply chain collaborative optimization method is executed by the processor, it performs the following steps: Acquire historical demand data for goods and generate business format collaboration feature vectors; A time-series prediction model is obtained by training the model based on the feature vector of business format collaboration, and the predicted demand for the next few days and the prediction confidence interval are output based on the time-series prediction model. For each SKU-store combination, calculate the out-of-stock risk score and the slow-moving risk score; Prioritize each SKU-store combination based on out-of-stock risk score and slow-moving risk score; For SKU-store combinations that have not generated transfer candidate plans among high-priority objects, or have generated transfer candidate plans but whose transfer volume is insufficient to cover all out-of-stock quantities, a procurement plan is generated using an optimization model based on the demand forecast for the next few days and the forecast confidence interval.

9. The demand-driven business supply chain collaborative optimization system according to claim 8, characterized in that, The process of acquiring historical demand data for goods and generating a business format collaboration feature vector includes: Obtain historical demand data for products, including time, store identifier, SKU, sales quantity, and trigger scenario fields, and generate a demand metadata dataset; The demand metadata is aggregated at a preset time granularity to generate the original demand time series; Extract demand feature vectors from the original demand time series, including historical actual demand volume, time feature data and business scenario feature data within a preset time window; Using a preset length of rolling time window as the statistical unit, the system slides forward step by step. Within each time window, the Pearson correlation coefficient and demand overflow rate between the demand for the same SKU in each store in the region are calculated, generating a collaborative time-series feature vector that changes dynamically over time. The demand feature vector and the collaborative time-series feature vector are concatenated and merged to form the business format collaborative feature vector.

10. The demand-driven business supply chain collaborative optimization system according to claim 9, characterized in that, The time-series prediction model is obtained by training the model based on the business format synergy feature vector. Based on the time-series prediction model, the predicted demand for the next few days and the prediction confidence interval are output, including: Using the business format collaboration feature vector as input and the actual demand within a preset future time period as label, a supervised learning training sample set is generated; A time series prediction model is trained using a pre-defined LSTM algorithm based on a supervised learning training sample set. Obtain the business collaboration feature vector for the current preset time period, input it into the time series prediction model, and output the predicted demand for the next few days and the prediction confidence interval.

Images