Inventory management method and system based on demand prediction

Abstract

The invention relates to an inventory management method and system based on demand prediction. The method comprises the following steps: acquiring sales time sequence data, a point consumption curve, commodity shelf life data, external environment factors and other data; feature extraction is carried out on the integral consumption curve, and demand prediction processing is carried out based on sales time sequence data, integral urgency factors, regional integral density and external environment factors; calculating a safety stock threshold value of each warehouse according to the commodity demand prediction result and the integral urgency factor; and generating an optimal allocation path scheme based on the safe inventory threshold value and the current inventory state, generating a regional promotion scheme according to the commodity shelf life data, the meteorological data and the like, and executing the optimal allocation path scheme and the regional promotion scheme. According to the method, through demand prediction, inventory optimization allocation, promotion strategy collaboration and the like, the accuracy of inventory management is improved, complex demand fluctuation can be effectively dealt with, refined management and control of the inventory are realized, and the flexibility and adaptability of inventory management are enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of big data analysis and forecasting, and in particular relates to an inventory management method and system based on demand forecasting. Background Art

[0002] In modern business operations, inventory management is one of the core links in a company's supply chain system. With the intensification of market competition and the diversification of consumer demand, traditional inventory management models that rely on experience or single sales data are no longer able to cope with dynamic market changes. On the one hand, the point system is widely used in consumer scenarios, such as point redemption on e-commerce platforms and point deductions in physical supermarkets. Point consumption behavior is deeply correlated with product demand, but existing management methods do not fully tap the value of point data, making it difficult to accurately predict point-driven demand fluctuations. On the other hand, the impact of external factors such as weather and holidays on product demand is becoming increasingly significant. For example, extreme weather affects the demand for fresh food, and holidays stimulate gift consumption. However, inventory allocation and promotion strategies lack the support of multi-source data integration, which can easily lead to inventory backlogs or stockouts, increasing business operating costs. Summary of the Invention

[0003] Based on this, it is necessary to provide an inventory management method and system based on demand forecasting to address the above technical problems, aiming to improve the accuracy of demand forecasting and enhance the accuracy and real-time performance of inventory management.

[0004] In a first aspect, the present application provides an inventory management method based on demand forecasting, comprising:

[0005] Acquire multi-source data, including sales time series data, point consumption curves, product shelf life data, and external environmental factors. The point consumption curve is calculated based on the desensitized time series data of total point consumption in the regional dimension. External environmental factors include meteorological data and the national statutory holiday calendar.

[0006] Dynamic feature extraction is performed on the integral consumption curve to obtain the integral urgency factor and regional integral density. Based on sales time series data, integral urgency factor, regional integral density, and external environmental factors, demand forecasting is performed through a deep spatiotemporal integral network to generate commodity demand forecast results within a preset period.

[0007] Based on the commodity demand forecast results and the integral urgency factor, dynamic safety stock calculation is performed to obtain the safety stock threshold of each warehouse;

[0008] Based on the safety stock threshold and current inventory status, the ant colony algorithm is used to optimize inventory allocation and generate the optimal allocation path plan. Based on product shelf life data, meteorological data, national statutory holiday calendar and regional point density, regional promotion strategies are generated and processed to obtain regional promotion plans. The optimal allocation path plan and regional promotion plans are then executed.

[0009] In one embodiment, based on sales time series data, point urgency factors, regional point density, and external environmental factors, a deep spatiotemporal integration network is used to perform demand forecasting to generate product demand forecast results within a preset period, including:

[0010] Preprocess the sales time series data to construct a multidimensional feature tensor, which includes historical sales volume, price fluctuations, and seasonal factors;

[0011] Perform meteorological feature conversion processing based on meteorological data to obtain a normalized meteorological feature vector, and perform date feature encoding based on the national statutory holiday calendar to obtain a holiday feature vector. The normalized meteorological feature vector includes temperature, humidity, and precipitation;

[0012] The multidimensional feature tensor, normalized meteorological feature vector, holiday feature vector, integral urgency factor and regional integral density are input into the deep spatiotemporal integration network, including:

[0013] According to the multidimensional feature tensor, normalized meteorological feature vector, holiday feature vector, integral urgency factor and regional integral density, the time series feature extraction module based on the long short-term memory network in the deep spatiotemporal integration network is used to obtain the time series hidden state matrix.

[0014] The time series hidden state matrix, normalized meteorological feature vector, and holiday feature vector are concatenated, dynamic weights are calculated through the attention mechanism, and the concatenated features are weighted fused based on the dynamic weights to obtain a fused feature tensor.

[0015] Based on the fused feature tensor, demand forecasting is performed through the fully connected layer to obtain the unit period commodity demand forecast value;

[0016] According to the unit period commodity demand forecast value, multi-period forecast processing is performed through the sliding window iteration mechanism to predict the commodity demand forecast sequence within the preset period and obtain the commodity demand forecast result.

[0017] In one embodiment, based on the safety stock threshold and the current stock status, an ant colony algorithm is used to optimize stock transfers and generate an optimal transfer path solution, including:

[0018] Calculate inventory matching based on the current inventory status and safety stock threshold of each warehouse;

[0019] Construct a dual-objective optimization model, which includes a transportation cost objective function and a time penalty objective function. The transportation cost objective function is calculated based on the transportation distance and allocation quantity between warehouses, while the time penalty objective function is calculated based on the estimated transportation time and the overtime penalty coefficient. The weight distribution of the transportation cost objective function and the time penalty objective function is dynamically adjusted according to the integral urgency factor and the preset urgency threshold.

[0020] The ant colony algorithm is used to solve the dual-objective optimization model. In the pheromone update process, the ant colony algorithm generates the Pareto optimal solution set through iterative search based on transportation cost and inventory matching.

[0021] The solution with the highest comprehensive evaluation value is selected from the Pareto optimal solution set as the optimal allocation path solution. The comprehensive evaluation value is calculated by weighted integration of transportation cost, time penalty and inventory matching degree.

[0022] In one embodiment, a regional promotion strategy is generated based on product shelf life data, weather data, national statutory holiday calendar, and regional point density to obtain a regional promotion plan, including:

[0023] Based on the regional point density and product shelf life data, the expiration urgency and point redemption degree are calculated. The expiration urgency is calculated based on the difference between the remaining shelf life of the product and the preset critical value, and the point redemption degree is calculated based on the ratio of the regional point density to the preset point redemption threshold.

[0024] The normalized meteorological feature vector, holiday feature vector, expiration urgency, and point redemption degree are input into the meta-reinforcement learning system. The initial promotion strategy is generated through state space construction, action space definition, and reward function evaluation. The state space includes environmental characteristics, product expiration status, and regional point level, and the action space includes a combination of discount rate, point leverage ratio, and cash payment ratio.

[0025] Based on the preset promotion goals, the initial promotion strategy is optimized with multiple objectives. Combined with the regional point consumption rules and commodity cost data, the discount rate and point leverage ratio are adjusted to obtain a regional promotion plan. The preset promotion goals include sales growth threshold, point consumption rate threshold and inventory turnover rate threshold.

[0026] In one embodiment, a dynamic safety stock calculation process is performed based on the product demand forecast results and the integral urgency factor to obtain the safety stock threshold of each warehouse, including:

[0027] Obtain sensitivity coefficients based on commodity category attributes in commodity demand forecast results;

[0028] Based on the product demand forecast results and sensitivity coefficients, the demand standard deviation and replenishment lead time are calculated. Combined with the preset service level coefficient and integral urgency factor, the basic safety stock value is calculated using a preset formula.

[0029] Calculate the inventory correction factor based on the historical out-of-stock data and current inventory turnover rate of each warehouse;

[0030] The basic safety stock value is modified by the inventory correction coefficient to obtain the safety stock threshold.

[0031] In one embodiment, dynamic feature extraction processing is performed on the integral consumption curve to obtain the integral urgency factor and the regional integral density, including:

[0032] Based on the distribution of the remaining validity period of points in the point consumption curve, a timeliness weight calculation is performed to obtain the regional average remaining validity period of points;

[0033] Calculate the urgency factor based on the remaining validity period of the regional average points to obtain the points urgency factor;

[0034] Obtain the number of active users and calculate the ratio of the total regional point consumption to the number of active users in the point consumption curve to obtain the regional point density.

[0035] In one embodiment, the integral urgency factor is calculated by the following formula:

[0036]

[0037] Where Φ is the integral urgency factor, T avg is the remaining validity period of the regional average integral, β is the attenuation coefficient, k is the response steepness coefficient, T critical is the critical value of the integral validity period, w i is the weight coefficient of the i-th category of goods, I i is the total points of the i-th category of goods.

[0038] In a second aspect, the present application also provides an inventory management system based on demand forecasting, comprising:

[0039] The multi-source data acquisition module is used to obtain multi-source data, including sales time series data, point consumption curves, product shelf life data, and external environmental factors. The point consumption curve is calculated based on the desensitized time series data of total point consumption in the regional dimension. External environmental factors include meteorological data and the national statutory holiday calendar.

[0040] The integral feature extraction and demand forecasting module is used to perform dynamic feature extraction processing on the integral consumption curve to obtain the integral urgency factor and regional integral density. Based on the sales time series data, the integral urgency factor, the regional integral density and external environmental factors, the module performs demand forecasting processing through the deep spatiotemporal integral network to generate the product demand forecast results within the preset period.

[0041] The dynamic safety stock calculation module is used to perform dynamic safety stock calculations based on the product demand forecast results and the integral urgency factor to obtain the safety stock threshold of each warehouse;

[0042] The inventory allocation and promotion planning module is used to optimize inventory allocation based on the safety stock threshold and current inventory status through the ant colony algorithm, generate the optimal allocation path plan, and generate regional promotion strategies based on product shelf life data, meteorological data, national statutory holiday calendar and regional point density to obtain regional promotion plans and implement the optimal allocation path plan and regional promotion plans.

[0043] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the first aspect when executing the computer program.

[0044] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps in the first aspect when the computer program is processed.

[0045] The aforementioned demand forecasting-based inventory management method and system utilizes multi-source data, including sales time series, point consumption, product expiration dates, weather conditions, and holidays. The system then calculates the point consumption curve based on a desensitized regional time series of total points. This method and system comprehensively integrates business operation data with external influences, providing a data foundation for subsequent inventory management analysis. Secondly, the system extracts the point urgency factor and regional point density from the point consumption curve. Demand forecasting is performed using a deep spatiotemporal integration network, combining this multi-source data. By mining the value of point data and the correlations between multiple data points, the system accurately captures demand fluctuations driven by points and influenced by the environment, improving the accuracy and foresight of demand forecasts. Furthermore, the system dynamically calculates safety stock thresholds based on demand forecasts and point urgency factors. This allows for a more flexible and reasonable setting, adapting to demand fluctuations and point consumption pressures, and avoiding the risks of overstocking or stockouts associated with traditional fixed inventory models. Finally, the system optimizes allocation paths based on safety stock and current inventory, generates and executes regional promotion plans, and effectively stimulates consumption while balancing cost and efficiency.

[0046] Compared with traditional inventory management methods, this method not only improves the accuracy of inventory management, but also effectively responds to complex demand fluctuations caused by factors such as point consumption and external environmental changes, realizes refined inventory management and control, and enhances the flexibility and adaptability of inventory management through technical means such as multi-source data fusion, demand forecasting, dynamic inventory calculation, inventory optimization and allocation, and promotion strategy coordination. It can also effectively respond to complex demand fluctuations caused by factors such as point consumption and external environmental changes, realize refined inventory management and control, and enhances the flexibility and adaptability of inventory management. It can quickly respond to market changes and optimize inventory layout and promotion strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A flow chart of an inventory management method based on demand forecasting provided by an exemplary embodiment of the present invention;

[0049] Figure 2 A schematic structural diagram of an inventory management system based on demand forecasting provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0051] In one embodiment, Figure 1 As shown, a method for inventory management based on demand forecasting is provided. This embodiment uses the method applied to a terminal as an example for illustration. It is understood that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0052] S101: Acquire multi-source data, which includes sales time series data, point consumption curve, product shelf life data, and external environmental factors. The point consumption curve is calculated based on the desensitized regional dimension point consumption total time series data. External environmental factors include meteorological data and the national statutory holiday calendar.

[0053] Specifically, sales time series data can include product categories, historical sales, price fluctuations, and promotional records. For example, a community supermarket's daily milk sales data for the past three months and price adjustment records during weekend promotions can be used. Furthermore, user point consumption data within geographic areas, such as communities and business districts, can be aggregated and anonymized to remove personal identifiers and retain only regional statistics, creating a time series curve to generate a point consumption curve. For example, in a region, supermarket points can be synchronized with property management accounts. Community supermarket points can be linked to the community's property management payment system, and the total points used to offset property fees for all users each month can be obtained. After anonymization, regional point consumption time series data can be generated. Specifically, in this scenario, a custom system can be implemented where every 100 yuan spent at a supermarket is accrued 7 points, with 1 point equal to 1 yuan in property fee deductions. Insufficient points can be topped up, and excess points can be withdrawn or redeemed for goods and services. Furthermore, when points are negative, supermarket membership benefits, such as discounts and point accumulation, can be frozen, and the parking system can be linked to restrict vehicle access for users in arrears. In addition, the rolling validity period of points can be 12 months, with automatic reminders for use to avoid sedimentation. The shelf life data of goods can record the shelf life information of each product in the region. And by analyzing external environmental factors such as meteorological data and national statutory holiday calendars, their impact on the demand for goods in the region can be analyzed, such as changes in demand for certain types of goods under specific weather conditions or consumption peaks during holidays. Indicatively, in this scenario, regular health lectures such as health preservation and chronic disease management can be held, combined with recommendations for healthy products, and certain points can be obtained by signing up for health lectures.

[0054] S102: Dynamic feature extraction is performed on the integral consumption curve to obtain the integral urgency factor and regional integral density. Based on the sales time series data, integral urgency factor, regional integral density and external environmental factors, demand forecasting is performed through a deep spatiotemporal integral network to generate commodity demand forecast results within a preset period.

[0055] Specifically, the points urgency factor can be calculated by analyzing the short-term fluctuations of the points consumption curve within a region. For example, when points consumption in a region increases significantly as the points expiration date approaches, it indicates that there is an urgency for points consumption in the region. This points urgency factor reflects the overall points consumption trend in the region. The regional points density can be calculated by analyzing the long-term distribution of the points consumption curve within the region, which can reflect the level of active points consumption in the region. For example, the points density in a certain commercial district may be higher than that in other areas, which means that points consumption in this area is more frequent. Combining the above factors and features with sales time series data and external environmental factors can be processed through a deep spatiotemporal integral network. This network can analyze the spatiotemporal characteristics of consumption data within a region, predict changes in commodity demand in the future, and provide a scientific basis for inventory management.

[0056] S103: Perform dynamic safety stock calculation based on the commodity demand forecast result and the integral urgency factor to obtain the safety stock threshold of each warehouse.

[0057] Specifically, the safety stock threshold for each warehouse can be calculated based on the demand forecast results and the points urgency factor obtained in step S102. For example, if the forecast shows that demand for a certain product in a certain area will surge in the near future due to the urgency of point consumption, the safety stock of the product can be appropriately increased to ensure the stability of inventory supply.

[0058] S104: Based on the safety stock threshold and current inventory status, the ant colony algorithm is used to optimize inventory allocation and generate an optimal allocation path plan. Based on product shelf life data, meteorological data, national statutory holiday calendar, and regional point density, regional promotion strategies are generated to obtain regional promotion plans, and the optimal allocation path plan and regional promotion plans are executed.

[0059] Specifically, an ant colony algorithm can be used to plan the optimal allocation route based on each warehouse's safety stock threshold and current inventory status. For example, when inventory in a certain area falls below the safety threshold, goods can be automatically allocated from nearby areas, and the optimal transportation route can be planned. Furthermore, promotional plans can be developed by combining product expiration date data, weather data, holiday calendars, and regional point density. For example, when a product is nearing its expiration date and point consumption is active in a region, a "points + cash" promotion can be launched to reduce inventory backlogs while stimulating consumption.

[0060] The above method, by acquiring data such as sales time series data, point consumption curves, and external environmental factors, provides comprehensive data support for subsequent inventory management decisions. Secondly, feature extraction is performed on the point consumption curve, further exploring the intrinsic value of point data. Demand forecasting is also performed using a deep spatiotemporal integral network, incorporating factors such as temporal variation, spatial differences, and point-driven factors. This results in more accurate product demand forecasts, effectively addressing demand fluctuations caused by point consumption and external environmental changes. Furthermore, by incorporating a point urgency factor into the calculation, the demand uncertainty caused by point consumption is fully accounted for, making the setting of safety stock levels at each warehouse more flexible and adaptable, thereby avoiding inventory overstocks or stock-outs. Inventory allocation optimization based on this safety stock threshold and current inventory status further optimizes the geographical distribution of inventory and ensures balanced inventory levels across warehouses. Furthermore, regional promotion strategies are generated based on product expiration dates and meteorological data, effectively stimulating consumption and reducing inventory losses caused by external environmental fluctuations or product expiration. Finally, by implementing the optimal allocation path plan and regional promotion plan, dynamic optimization of inventory management and precise marketing can be achieved, improving the efficiency and scientific nature of inventory management.

[0061] In one embodiment, dynamic feature extraction processing is performed on the integral consumption curve to obtain the integral urgency factor and regional integral density, including:

[0062] Based on the distribution of the remaining validity period of points in the point consumption curve, a timeliness weight calculation is performed to obtain the regional average remaining validity period of points;

[0063] Calculate the urgency factor based on the remaining validity period of the regional average points to obtain the points urgency factor;

[0064] Obtain the number of active users and calculate the ratio of the total regional point consumption to the number of active users in the point consumption curve to obtain the regional point density.

[0065] Specifically, the distribution of the remaining validity period of points based on the point consumption curve can be grouped and counted by the remaining validity period. Subsequently, the timeliness weight can be calculated based on the proportion of points with different remaining validity periods, and the regional average remaining validity period of points can be obtained. And based on the regional average remaining validity period of points, if the regional average remaining validity period of points is short, it indicates that a large number of points in the region are about to expire, that is, consumers may use points in a concentrated manner in a short period of time, so the value of the points urgency factor can be increased accordingly. This factor can effectively reflect the consumption peak that may be caused by the expiration of points in the region, and help predict the growth trend of point consumption in the short term.

[0066] Schematically, the integral urgency factor can be calculated by the following formula:

[0067]

[0068] Where Φ is the integral urgency factor, T avg is the remaining validity period of the regional average integral, β is the attenuation coefficient, k is the response steepness coefficient, T critical is the critical value of the integral validity period, w i is the weight coefficient of the i-th category of goods, I i is the total points of the i-th category of goods.

[0069] Furthermore, the number of active users in the area can be obtained, that is, users who have recently consumed or accumulated points. Then, by calculating the total point consumption in the area and ratioing it with the number of active users, the regional point density can be obtained. This regional point density can indicate the average amount of points consumed by each active user in the area, reflecting the overall level of point consumption activity in the area. Through this indicator, the popularity of point consumption in different regions can be compared, providing data support for subsequent inventory management, promotion strategy formulation, etc. For example, areas with high point density need to monitor the inventory of related products more closely to cope with possible consumption peaks.

[0070] In one embodiment, based on sales time series data, point urgency factors, regional point density, and external environmental factors, a deep spatiotemporal integration network is used to perform demand forecasting to generate product demand forecast results within a preset period, including:

[0071] Preprocess the sales time series data to construct a multidimensional feature tensor, which includes historical sales volume, price fluctuations, and seasonal factors;

[0072] Perform meteorological feature conversion processing based on meteorological data to obtain a normalized meteorological feature vector, and perform date feature encoding based on the national statutory holiday calendar to obtain a holiday feature vector. The normalized meteorological feature vector includes temperature, humidity, and precipitation;

[0073] The multidimensional feature tensor, normalized meteorological feature vector, holiday feature vector, integral urgency factor and regional integral density are input into the deep spatiotemporal integration network, including:

[0074] According to the multidimensional feature tensor, normalized meteorological feature vector, holiday feature vector, integral urgency factor and regional integral density, the time series feature extraction module based on the long short-term memory network in the deep spatiotemporal integration network is used to obtain the time series hidden state matrix.

[0075] The time series hidden state matrix, normalized meteorological feature vector, and holiday feature vector are concatenated, dynamic weights are calculated through the attention mechanism, and the concatenated features are weighted fused based on the dynamic weights to obtain a fused feature tensor.

[0076] Based on the fused feature tensor, demand forecasting is performed through the fully connected layer to obtain the unit period commodity demand forecast value;

[0077] According to the unit period commodity demand forecast value, multi-period forecast processing is performed through the sliding window iteration mechanism to predict the commodity demand forecast sequence within the preset period and obtain the commodity demand forecast result.

[0078] Specifically, sales time series data can be preprocessed through data cleaning, missing value processing, and outlier detection to ensure data quality. This data can then be analyzed for historical sales, price fluctuations caused by promotional discounts, and seasonal sales differences to construct a multidimensional feature tensor. Weather conditions also have a direct impact on demand for certain products, such as increased demand for umbrellas on rainy days. Furthermore, holidays are often accompanied by specific consumer demands, such as for holiday food or gifts. Therefore, the multidimensional feature tensor, normalized weather feature vectors, holiday feature vectors, integral urgency factors, and regional integral density can be input into a deep spatiotemporal integration network to further analyze product demand within a preset future period. Long-short-term memory networks (LSTMs) can capture temporal dependencies in data. Using a feature extraction module based on LSTMs, temporal features can be extracted from the input data to produce a temporal hidden state matrix.

[0079] By concatenating the time series hidden state matrix, normalized meteorological feature vectors, and holiday feature vectors and calculating weights based on an attention mechanism, the deep spatiotemporal integration network can identify which features are more important in the prediction process, thereby improving prediction accuracy and generating dynamic weights. These weights reflect the impact of different external factors on product demand. The concatenated features are weighted and fused based on the dynamic weights to generate a fused feature tensor. This is then passed through a fully connected layer for demand forecasting, yielding a unit-period product demand forecast. This forecast reflects the product demand for the next period after accounting for relevant factors. Finally, using a sliding window iteration mechanism, the unit-period forecast can be extended to multiple periods, generating a series of product demand forecasts for a preset period.

[0080] In one embodiment, a dynamic safety stock calculation process is performed based on the commodity demand forecast results and the integral urgency factor to obtain the safety stock threshold of each warehouse, including:

[0081] Obtain sensitivity coefficients based on commodity category attributes in commodity demand forecast results;

[0082] Based on the product demand forecast results and sensitivity coefficients, the demand standard deviation and replenishment lead time are calculated. Combined with the preset service level coefficient and integral urgency factor, the basic safety stock value is calculated using a preset formula.

[0083] Calculate the inventory correction factor based on the historical out-of-stock data and current inventory turnover rate of each warehouse;

[0084] The basic safety stock value is modified by the inventory correction coefficient to obtain the safety stock threshold.

[0085] Specifically, for each product category in the demand forecast results, a sensitivity coefficient is derived based on its category attributes. This sensitivity coefficient reflects the sensitivity of demand for each product to changes in points. For example, demand for fresh produce may fluctuate rapidly due to point-based promotions, and therefore may have a higher sensitivity coefficient. The standard deviation of demand for each product is then calculated based on the product demand forecast results and sensitivity coefficients. This standard deviation reflects the volatility of demand. This can also be used to determine the replenishment lead time—the time it takes from an order to the product's arrival at the warehouse. The preset service level coefficient is a pre-set indicator used to ensure the probability of meeting customer demand within a specific timeframe. Furthermore, a base safety stock value can be calculated using a pre-set formula, combined with the point urgency factor. For example, if the point urgency factor indicates a possible increase in point consumption in the short term, the base safety stock value can be increased accordingly to accommodate the increase in demand. Furthermore, historical out-of-stock data for each warehouse can be used to analyze inventory rationality. For example, if a warehouse experienced two out-of-stocks of fresh produce last month, with an out-of-stock rate of 5%, this suggests weak inventory management. The inventory turnover rate can also be calculated by comparing monthly sales to average inventory. Based on each warehouse's historical out-of-stock data and current inventory turnover rate, a corresponding inventory correction factor can be calculated. This correction factor is then used to adjust the base safety stock value to reflect the specific warehouse's actual situation and determine the safety stock threshold, enabling better response to market changes and reducing inventory overstock.

[0086] In one embodiment, based on the safety stock threshold and the current inventory status, an ant colony algorithm is used to optimize inventory transfers and generate an optimal transfer path solution, including:

[0087] Calculate inventory matching based on the current inventory status and safety stock threshold of each warehouse;

[0088] Construct a dual-objective optimization model, which includes a transportation cost objective function and a time penalty objective function. The transportation cost objective function is calculated based on the transportation distance and allocation quantity between warehouses, while the time penalty objective function is calculated based on the estimated transportation time and the overtime penalty coefficient. The weight distribution of the transportation cost objective function and the time penalty objective function is dynamically adjusted according to the integral urgency factor and the preset urgency threshold.

[0089] The ant colony algorithm is used to solve the dual-objective optimization model. In the pheromone update process, the ant colony algorithm generates the Pareto optimal solution set through iterative search based on transportation cost and inventory matching.

[0090] The solution with the highest comprehensive evaluation value is selected from the Pareto optimal solution set as the optimal allocation path solution. The comprehensive evaluation value is calculated by weighted integration of transportation cost, time penalty and inventory matching degree.

[0091] Specifically, by comparing each warehouse's current inventory level with a safety stock threshold calculated based on demand forecasts and an integral urgency factor, the inventory fit of each warehouse can be calculated. The dual-objective optimization model aims to minimize transportation costs and time penalties. Transportation costs are calculated based on the transportation distance between warehouses, the number of transfers, and the unit transportation cost, while time penalties are calculated based on the estimated transportation time, the overtime penalty coefficient, and the number of transfers. When the integral urgency factor exceeds the preset urgency threshold, it indicates that the credits for a region are about to expire and inventory needs to be transferred more quickly to meet demand. Therefore, the weight of the time penalty objective function is increased. The ant colony algorithm, an optimization algorithm that mimics the foraging behavior of ants in nature, finds the optimal solution through pheromone updates and iterative search. This algorithm solves the dual-objective optimization model. During the pheromone update process, the pheromone intensity is adjusted based on transportation costs and inventory fit. Through iterative search, a Pareto optimal solution set is generated. This Pareto optimal solution set contains solutions that strike a balance between the two objectives. Furthermore, the comprehensive value of each solution in the solution set can be calculated by weighted integration of transportation costs, time penalties, and inventory matching, and the solution with the highest comprehensive evaluation value is selected as the optimal allocation path solution. This comprehensive value reflects the overall performance of the allocation solution on multiple objective functions.

[0092] In one embodiment, a regional promotion strategy is generated based on product shelf life data, weather data, national statutory holiday calendar, and regional point density to obtain a regional promotion plan, including:

[0093] Based on the regional point density and product shelf life data, the expiration urgency and point redemption degree are calculated. The expiration urgency is calculated based on the difference between the remaining shelf life of the product and the preset critical value, and the point redemption degree is calculated based on the ratio of the regional point density to the preset point redemption threshold.

[0094] The normalized meteorological feature vector, holiday feature vector, expiration urgency, and point redemption degree are input into the meta-reinforcement learning system. The initial promotion strategy is generated through state space construction, action space definition, and reward function evaluation. The state space includes environmental characteristics, product expiration status, and regional point level, and the action space includes a combination of discount rate, point leverage ratio, and cash payment ratio.

[0095] Based on the preset promotion goals, the initial promotion strategy is optimized with multiple objectives. Combined with the regional point consumption rules and commodity cost data, the discount rate and point leverage ratio are adjusted to obtain a regional promotion plan. The preset promotion goals include sales growth threshold, point consumption rate threshold and inventory turnover rate threshold.

[0096] Specifically, by calculating the difference between the remaining shelf life of each type of product and the preset critical value, the expiration urgency can be obtained. The expiration urgency reflects the urgency of how long the product needs to be sold to avoid expiration losses. For example, for products with a short shelf life, such as milk or bread, the expiration urgency will be higher, and promotions need to be given priority. By calculating the ratio of the regional point density to the preset point redemption threshold, the point redemption degree can be obtained. The point redemption degree reflects the active use of points in the region and the potential demand for point redemption. For example, if the point density in a certain area is high, it means that consumers in the area are more inclined to use points for consumption, and promotions with high point discount ratios can be launched.

[0097] Specifically, to account for the impact of external factors such as weather and holidays on product sales, the normalized weather feature vector, holiday feature vector, expiration urgency, and point redemption can be input into the meta-reinforcement learning system to construct a state space, define an action space, and set a reward function. Environmental features in the state space can be quantified by weather and holidays, product expiration status can be reflected by expiration urgency, and regional point levels can be quantified by point redemption. The action space, on the other hand, can generate a variety of options for promotional activities based on different combinations of discount rates, point leverage ratios, and cash payment ratios. For example, a high discount rate combined with a low point leverage ratio can be used to attract price-sensitive consumers, or a low discount rate combined with a high point leverage ratio can be used to encourage users to use their points. The reward function serves as a criterion for evaluating the quality of generated promotional strategies. Positive rewards can be given when sales increase, while negative feedback can be given when sales decrease. This allows for continuous optimization and ultimately results in a more effective initial promotional strategy.

[0098] Furthermore, sales growth thresholds, point consumption rate thresholds, and inventory turnover rate thresholds can be used as preset promotional targets to perform multi-objective optimization on the initial promotion strategy. Regional point consumption rules and product cost data can be combined to precisely adjust the discount rate and point leverage ratio. For example, if near-expiry milk sales need to be boosted, and its product cost data can be the purchase cost data, if the purchase price of milk is 15 yuan per piece, the promotion plan could be 20 points + 10 yuan, and the regional point consumption rule could be 1 point equal to 1 yuan, resulting in a total profit of 30 yuan, meeting profitability requirements and increasing sales. If sales under the initial strategy do not reach the growth threshold, and product costs allow, the discount rate can be appropriately increased. If the point consumption rate is not up to standard, the point leverage ratio can be adjusted to increase the points redemption incentive. Finally, a regional promotion plan that meets promotional needs can be obtained. For example, this regional promotion plan could adopt a joint marketing model between property management and supermarkets, where points can be used not only for product purchases but also for property management fees. Membership tiers can also be established. Points earned through supermarkets can be used to upgrade property management membership tiers, providing additional benefits such as increased parking discounts.

[0099] Furthermore, this regional promotional plan could also employ a point-based redemption model for takeout delivery services or community group buying. Points can be used to offset a portion of the takeout order and delivery fees, or to offset group purchase costs. Furthermore, upon successful group purchase, the group leader earns 20 points, while each group member earns 10 points, thereby encouraging consumption. For group purchase delivery, choosing self-pickup earns double points, further increasing utilization at the pick-up point and reducing delivery costs.

[0100] like Figure 2 As shown, based on the same inventive concept, the embodiment of the present application also provides a demand forecasting inventory management system 200 for implementing the above-mentioned demand forecasting inventory management method. The implementation solution provided by this system is similar to the implementation solution described in the above-mentioned method. Therefore, the specific limitations of one or more embodiments of the demand forecasting inventory management system provided below can be found in the above-mentioned limitations of the demand forecasting inventory management method, and will not be repeated here. The system includes:

[0101] Multi-source data acquisition module 201 is used to obtain multi-source data, including sales time series data, point consumption curves, product shelf life data, and external environmental factors. The point consumption curve is calculated based on the desensitized regional dimension point consumption total time series data. External environmental factors include meteorological data and the national statutory holiday calendar.

[0102] The integral feature extraction and demand forecasting module 202 is used to perform dynamic feature extraction processing on the integral consumption curve to obtain the integral urgency factor and regional integral density. Based on the sales time series data, the integral urgency factor, the regional integral density and external environmental factors, the module performs demand forecasting processing through a deep spatiotemporal integral network to generate a product demand forecast result within a preset period.

[0103] Dynamic safety stock calculation module 203, used to perform dynamic safety stock calculation processing based on the commodity demand forecast results and the integral urgency factor to obtain the safety stock threshold of each warehouse;

[0104] The inventory allocation and promotion planning module 204 is used to optimize inventory allocation based on the safety stock threshold and current inventory status through the ant colony algorithm to generate the optimal allocation path plan. It is also used to generate regional promotion strategies based on product shelf life data, meteorological data, national statutory holiday calendar and regional point density to obtain regional promotion plans and execute the optimal allocation path plan and regional promotion plans.

[0105] In the above system, the multi-source data acquisition module 201 acquires data including sales time series data, point consumption curves, product shelf life data, and external environmental factors, providing an accurate data foundation for subsequent demand forecasting and inventory management. The point feature extraction and demand forecasting module 202 dynamically extracts features from the point consumption curve to identify point urgency factors and regional point density. Combining sales time series data with external environmental factors, it performs demand forecasting using a deep spatiotemporal integration network, generating product demand forecasts for a preset period, providing a scientific basis for inventory management and promotional activities. The dynamic safety stock calculation module 203 dynamically calculates safety stock thresholds based on product demand forecasts and point urgency factors, determining the minimum inventory level each warehouse should maintain, making inventory management more flexible and adaptable to market demand fluctuations. The inventory allocation and promotion planning module 204 optimizes inventory allocation plans using an ant colony algorithm and formulates promotional strategies based on factors such as product shelf life, weather conditions, holidays, and point density. By implementing optimal allocation paths and regional promotion plans, it not only reduces inventory backlogs but also stimulates consumption and increases sales.

[0106] In an exemplary embodiment, the present invention further provides a computer device comprising a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the demand forecast-based inventory management method of the present application. A multi-core processor is preferred to improve the system's parallel processing capabilities. The memory provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate large amounts of supply information and computing tasks.

[0107] In an exemplary embodiment, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the demand forecast-based inventory management method of the present application.

[0108] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A method for inventory management based on demand forecasting, characterized in that: The method comprises: Acquire multi-source data, including sales time series data, point consumption curves, product shelf life data, and external environmental factors. The point consumption curves are calculated based on the desensitized time series data of total point consumption in the regional dimension. The external environmental factors include meteorological data and the national statutory holiday calendar. Dynamic feature extraction is performed on the point consumption curve to obtain a point urgency factor and a regional point density, and based on the sales time series data, the point urgency factor, the regional point density, and the external environmental factors, demand forecasting is performed using a deep spatiotemporal integration network to generate a product demand forecast result within a preset period; Performing dynamic safety stock calculation based on the commodity demand forecast result and the integral urgency factor to obtain a safety stock threshold for each warehouse; Based on the safety stock threshold and the current inventory status, an ant colony algorithm is used to optimize inventory allocation and generate an optimal allocation path plan. In addition, a regional promotion strategy is generated based on the product shelf life data, the meteorological data, the national statutory holiday calendar, and the regional point density to obtain a regional promotion plan. The optimal allocation path plan and the regional promotion plan are then executed.

2. The method according to claim 1, characterized in that The method of performing demand forecasting processing based on the sales time series data, the integral urgency factor, the regional integral density, and the external environmental factors through a deep spatiotemporal integral network to generate a commodity demand forecast result within a preset period includes: Preprocessing the sales time series data to construct a multidimensional feature tensor, wherein the multidimensional feature tensor includes historical sales volume, price fluctuations, and seasonal factors; Performing meteorological feature conversion processing on the meteorological data to obtain a normalized meteorological feature vector, and performing date feature encoding based on the national statutory holiday calendar to obtain a holiday feature vector, wherein the normalized meteorological feature vector includes temperature, humidity, and precipitation; Inputting the multidimensional feature tensor, the normalized meteorological feature vector, the holiday feature vector, the integral urgency factor, and the regional integral density into the deep spatiotemporal integration network includes: According to the multidimensional feature tensor, the normalized meteorological feature vector, the holiday feature vector, the integral urgency factor, and the regional integral density, a feature extraction module based on a long short-term memory network in the deep spatiotemporal integration network is used to perform temporal feature extraction processing to obtain a temporal hidden state matrix; concatenating the temporal hidden state matrix, the normalized meteorological feature vector, and the holiday feature vector, calculating dynamic weights through an attention mechanism, and performing weighted fusion on the concatenated features based on the dynamic weights to obtain a fused feature tensor; Based on the fused feature tensor, demand forecasting is performed through a fully connected layer to obtain a unit period commodity demand forecast value; According to the unit period commodity demand forecast value, a multi-period forecast process is performed through a sliding window iteration mechanism to predict the commodity demand forecast sequence within the preset period to obtain the commodity demand forecast result.

3. The method according to claim 1, characterized in that The method of performing inventory allocation optimization processing based on the safety stock threshold and the current inventory status by using an ant colony algorithm to generate an optimal allocation path solution includes: Calculating an inventory matching degree according to the current inventory status of each warehouse and the safety stock threshold; Constructing a dual-objective optimization model, the dual-objective optimization model including a transportation cost objective function and a time penalty objective function, wherein the transportation cost objective function is calculated based on the transportation distance and the allocation quantity between each warehouse, and the time penalty objective function is calculated based on the estimated transportation time and the overtime penalty coefficient, wherein the weight distribution of the transportation cost objective function and the time penalty objective function is dynamically adjusted according to the integral urgency factor and the preset urgency threshold; An ant colony algorithm is used to solve the dual-objective optimization model, wherein the ant colony algorithm generates a Pareto optimal solution set through iterative search based on transportation cost and inventory matching during the pheromone update process; The solution with the highest comprehensive evaluation value is selected from the Pareto optimal solution set as the optimal allocation path solution, and the comprehensive evaluation value is calculated by weighted integration of transportation cost, time penalty and inventory matching degree.

4. The method according to claim 2, characterized in that The generating process of a regional promotion strategy based on the product shelf life data, the meteorological data, the national statutory holiday calendar, and the regional point density to obtain a regional promotion plan includes: Calculating the expiration urgency and point redemption degree based on the regional point density and the product shelf life data, wherein the expiration urgency is calculated based on the difference between the remaining shelf life of the product and a preset critical value, and the point redemption degree is calculated based on the ratio of the regional point density to a preset point redemption threshold; Inputting the normalized meteorological feature vector, the holiday feature vector, the expiration urgency, and the point redemption degree into a meta-reinforcement learning system, generating an initial promotion strategy through state space construction, action space definition, and reward function evaluation, wherein the state space includes environmental characteristics, product expiration status, and regional point level, and the action space includes a combination of discount rate, point leverage ratio, and cash payment ratio; The initial promotion strategy is subjected to multi-objective optimization processing according to preset promotional targets. The discount rate and the point leverage ratio are adjusted in combination with regional point consumption rules and commodity cost data to obtain the regional promotion plan. The preset promotional targets include a sales growth threshold, a point consumption rate threshold, and an inventory turnover rate threshold.

5. The method according to claim 1, characterized in that The dynamic safety stock calculation process is performed based on the commodity demand forecast result and the integral urgency factor to obtain the safety stock threshold of each warehouse, including: Obtaining a sensitivity coefficient based on a commodity category attribute in the commodity demand forecast result; Based on the commodity demand forecast result and the sensitivity coefficient, the demand standard deviation and replenishment lead time are calculated, and the basic safety stock value is calculated using a preset formula in combination with the preset service level coefficient and the integral urgency factor; Calculating an inventory correction factor based on historical warehouse out-of-stock data and current inventory turnover rate for each of the warehouses; The basic safety stock value is corrected by the inventory correction coefficient to obtain the safety stock threshold.

6. The method according to claim 1, characterized in that The dynamic feature extraction process is performed on the integral consumption curve to obtain the integral urgency factor and the regional integral density, including: Based on the distribution of the remaining validity period of points in the point consumption curve, a timeliness weight calculation process is performed to obtain the regional average remaining validity period of points; Calculate the urgency factor based on the remaining validity period of the average points in the region to obtain the points urgency factor; The number of active users is obtained, and the ratio of the total regional point consumption in the point consumption curve to the number of active users is calculated to obtain the regional point density.

7. The method according to claim 6, characterized in that The integral urgency factor is calculated by the following formula: Where Φ is the integral urgency factor, T avg is the remaining validity period of the average integral of the region, β is the attenuation coefficient, k is the response steepness coefficient, T critical is the critical value of the integral validity period, w i is the weight coefficient of the i-th category of goods, I i is the total points of the i-th category of goods.

8. A demand forecasting inventory management system, characterized in that: The system comprises: A multi-source data acquisition module is used to acquire multi-source data, including sales time series data, point consumption curves, product shelf life data, and external environmental factors. The point consumption curves are calculated based on the desensitized time series data of total point consumption in the regional dimension. The external environmental factors include meteorological data and the national statutory holiday calendar. An integral feature extraction and demand forecasting module, configured to perform dynamic feature extraction processing on the integral consumption curve to obtain an integral urgency factor and a regional integral density, and perform demand forecasting processing using a deep spatiotemporal integral network based on the sales time series data, the integral urgency factor, the regional integral density, and the external environmental factors, to generate a commodity demand forecast result within a preset period; A dynamic safety stock calculation module is used to perform dynamic safety stock calculation processing based on the commodity demand forecast result and the integral urgency factor to obtain the safety stock threshold of each warehouse; The inventory allocation and promotion planning module is used to optimize inventory allocation based on the safety stock threshold and the current inventory status through an ant colony algorithm to generate an optimal allocation path plan, and to generate a regional promotion strategy based on the product shelf life data, the meteorological data, the national statutory holiday calendar and the regional point density to obtain a regional promotion plan, and to execute the optimal allocation path plan and the regional promotion plan.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images