Multi-node time series data fusion-oriented replenishment demand dynamic prediction method and system

Abstract

The invention provides a multi-node time sequence data fusion-oriented replenishment demand dynamic prediction method and system, and relates to the technical field of data synchronization, and the method comprises the steps: deploying a plurality of lightweight store agents at the background of a plurality of store cashier terminals; according to a commodity sales volume accumulation calculation result, performing virtual inventory updating of the plurality of chain stores to obtain a plurality of groups of real-time commodity inventory data; carrying out inventory demand prediction according to the multiple groups of time sequence commodity inventory data and the multiple groups of real-time commodity inventory data to obtain multiple groups of commodity replenishment time and multiple groups of commodity replenishment prediction quantity; performing supply chain dynamic fitting, and outputting a replenishment distribution route; and the supply chain end responds to the replenishment delivery route to perform dynamic replenishment of the plurality of chain stores. The technical problems that in the prior art, sales and inventory of chain stores are mutually independent, a unified real-time data synchronization mechanism is lacked, dynamic replenishment optimization is difficult to carry out according to the sales and inventory data of the whole chain, and inventory overstock or frequent occurrence of stockout is caused are solved.

Description

Technical Field

[0001] This invention relates to the field of data synchronization technology, specifically to a method and system for dynamic prediction of replenishment demand for multi-node time-series data fusion. Background Technology

[0002] In multi-store chain operations, the number of stores and the variety of inventory are numerous, and the dynamic changes in product sales are more complex. Many chain stores still use independently managed sales and inventory systems without a unified data sharing mechanism. This independent operation means that the data of each store exists in isolation and cannot be synchronized or managed in real time with the data of other stores. As a result, each store can only make inventory management and replenishment decisions based on its own isolated sales data, making it difficult to dynamically optimize replenishment based on the sales and inventory data of the entire chain. This leads to frequent inventory backlogs or stockouts, increases management difficulty, and reduces supply chain efficiency. Summary of the Invention

[0003] This application provides a method and system for dynamic prediction of replenishment demand based on multi-node time-series data fusion. It aims to solve the technical problem in the prior art that the sales and inventory of chain stores are independent of each other, lack a unified real-time data synchronization mechanism, and are difficult to optimize dynamic replenishment based on the sales and inventory data of the entire chain, resulting in frequent inventory backlogs or stockouts.

[0004] The first aspect disclosed in this application provides a method for dynamic prediction of replenishment demand based on multi-node time-series data fusion. The method includes: deploying multiple lightweight store agent agents in the backend of multiple store POS terminals across multiple chain stores, wherein each lightweight store agent agent is responsible for real-time capture and encryption of POS transaction data; after receiving multiple POS transaction data pushed in real-time by the multiple lightweight store agent agents via MQTT, a central server performs the following steps: Step a: updating the virtual inventory of the multiple chain stores based on the cumulative calculation result of the sales volume of the multiple POS transaction data, obtaining multiple sets of real-time product inventory data; Step b: predicting inventory demand based on the multiple sets of time-series product inventory data obtained through backtracking and the multiple sets of real-time product inventory data, obtaining multiple sets of product replenishment times and multiple sets of product replenishment forecast quantities; Step c: dynamically fitting the supply chain based on the multiple sets of product replenishment times and multiple sets of product replenishment forecast quantities, outputting replenishment delivery routes; and the supply chain responds to the replenishment delivery routes issued by the central server to dynamically replenish the multiple chain stores.

[0005] The second aspect of this application discloses a dynamic replenishment demand forecasting system for multi-node time-series data fusion. The system is used in the aforementioned dynamic replenishment demand forecasting method for multi-node time-series data fusion. The system includes: a deployment module for deploying multiple lightweight store agent agents in the backend of multiple store POS terminals across multiple chain stores, wherein the lightweight store agent agents are responsible for real-time capture and encryption of POS transaction data; and a data processing module where a central server receives multiple POS transaction data pushed in real-time by the multiple lightweight store agent agents via MQTT and executes the following steps: a virtual inventory update unit for executing step a: based on the multiple... The system calculates the cumulative sales volume of goods based on the cash register transaction data, updates the virtual inventory of the multiple chain stores, and obtains multiple sets of real-time goods inventory data. An inventory demand forecasting unit performs step b: based on the backtracked multiple sets of time-series goods inventory data and the multiple sets of real-time goods inventory data, it forecasts inventory demand to obtain multiple sets of goods replenishment times and multiple sets of goods replenishment forecast quantities. A dynamic fitting unit performs step c: based on the multiple sets of goods replenishment times and multiple sets of goods replenishment forecast quantities, it performs dynamic supply chain fitting and outputs replenishment delivery routes. A dynamic replenishment module, on the supply chain side, responds to the replenishment delivery routes issued by the central server to dynamically replenish the multiple chain stores.

[0006] One or more technical solutions provided in this application have at least the following beneficial effects: By deploying lightweight store agent data across multiple chain stores, real-time capture and encryption of cash register transaction data were achieved. The lightweight store agent features low resource consumption and efficient data transmission, ensuring data timeliness and security. After receiving multiple cash register transaction data pushes from the lightweight store agent data, the central server first performs a virtual inventory update. By accumulating sales volume, it can promptly update the inventory status of each store, ensuring the real-time accuracy of inventory information and providing precise data for subsequent replenishment decisions. Then, using time-series and real-time inventory data, inventory demand forecasting is performed. Based on historical sales trends and current inventory status, the system predicts the demand for goods. By predicting replenishment time and quantity, this method can identify which products are about to run out of stock in advance, ensuring timely replenishment and avoiding inventory shortages or surpluses. Finally, based on the predicted replenishment time and quantity, the supply chain is dynamically fitted, and the optimal replenishment and delivery route is determined through optimization algorithms. This process integrates factors such as the geographical location of each store and the replenishment priority of goods, generating efficient delivery routes that maximize transportation efficiency and resource utilization while reducing delivery costs and time. After the replenishment and delivery route is determined by the central server, the supply chain responds and initiates replenishment execution. Through this dynamic replenishment mechanism, chain stores can quickly replenish goods when inventory approaches the threshold, ensuring sufficient supply and improving the flexibility and responsiveness of inventory management.

[0007] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the replenishment demand dynamic prediction method based on multi-node time series data fusion provided in an embodiment of this application.

[0009] Figure 2 This is a schematic diagram of the structure of a dynamic prediction system for replenishment demand based on multi-node time-series data fusion provided in an embodiment of this application.

[0010] Figure labeling: Deployment module 10, data processing module 20, virtual inventory update unit 21, inventory demand forecasting unit 22, dynamic fitting unit 23, dynamic replenishment module 30. Detailed Implementation

[0011] This application provides a method and system for dynamic prediction of replenishment demand based on multi-node time-series data fusion. It solves the technical problem in the prior art that the sales and inventory of chain stores are independent of each other, lack a unified real-time data synchronization mechanism, and are difficult to dynamically optimize replenishment based on the sales and inventory data of the entire chain, resulting in frequent inventory backlogs or stockouts.

[0012] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0013] Example 1, as Figure 1 As shown in the embodiments of this application, a method for dynamic prediction of replenishment demand based on multi-node time-series data fusion is provided, the method comprising: Multiple lightweight store agent agents are deployed in the back-end of multiple store POS terminals in multiple chain stores. The lightweight store agent agents are responsible for capturing and encrypting POS transaction data in real time.

[0014] For multiple chain stores, a lightweight store agent is deployed in the backend of each store's POS terminal. This lightweight store agent is responsible for the following tasks: whenever a transaction occurs at a store, the lightweight store agent monitors and captures transaction data in real time in the backend, such as sales amount, product code, quantity, etc. Simultaneously, to ensure the security and privacy protection of transaction data, the lightweight store agent encrypts the POS transaction data to ensure that the data is not leaked or tampered with during transmission. The deployment of the lightweight store agent ensures the system's efficiency and security. Because these lightweight store agents only process and transmit data and do not perform complex calculations, their resource consumption is low, making them suitable for deployment on a large number of devices in a distributed system.

[0015] After receiving multiple POS transaction data pushed in real time via MQTT from the multiple lightweight store agent agents, the central server performs the following steps: Step a: Based on the cumulative calculation results of the sales volume of the multiple cash register transaction data, update the virtual inventory of the multiple chain stores to obtain multiple sets of real-time product inventory data; Step b: Based on the multiple sets of time-series product inventory data obtained from backtracking and the multiple sets of real-time product inventory data, perform inventory demand forecasting to obtain multiple sets of product replenishment time and multiple sets of product replenishment forecast quantity; Step c: Based on the multiple sets of product replenishment time and multiple sets of product replenishment forecast quantity, perform dynamic fitting of the supply chain and output replenishment and delivery routes.

[0016] The deployed lightweight store agent pushes the corresponding store's cash register transaction data to the central server in real time via the MQTT protocol. MQTT is a lightweight messaging protocol suitable for real-time, low-latency communication. After receiving the cash register transaction data, the central server performs the following steps: Based on multiple POS transaction data from various chain stores, for each product, the sales volume from different chain stores is summed to obtain the total sales volume of the product across all chain stores. This process is based on product codes, meaning that sales transaction data is aggregated using product codes to ensure the accuracy of the total sales volume for each product. The virtual inventory for each chain store is updated based on the total sales volume. For each chain store, the virtual inventory represents the actual inventory of the products. Based on the sales volume and current inventory, inventory changes are estimated, and the virtual inventory of each product across different chain stores is automatically calculated. This serves as the basis for providing data for subsequent inventory demand forecasting, i.e., multiple sets of real-time product inventory data.

[0017] By reviewing historical data, we can obtain historical changes in merchandise inventory across multiple chain stores, resulting in multiple sets of time-series merchandise inventory data. This historical data includes sales trends, inventory changes, and promotional impacts for each product over a past period. Based on this historical data, we use a time series analysis model to predict future demand for each product, including future sales volume and inventory change trends. Based on the time series analysis results and multiple sets of real-time merchandise inventory data, we can predict inventory demand and obtain multiple sets of replenishment times and replenishment forecasts.

[0018] Multiple sets of product replenishment times and multiple sets of product replenishment forecasts are input into the supply chain optimization algorithm to generate appropriate replenishment and delivery plans, aiming to optimize supply chain efficiency. Based on the optimization results, the replenishment and delivery routes for each product are output. Each replenishment and delivery route includes the stores that need replenishment, the required replenishment quantity for each store, and the corresponding delivery time window, ensuring that the stores' inventory replenishment is timely and efficient.

[0019] The supply chain responds to the replenishment and delivery routes issued by the central server to dynamically replenish the multiple chain stores.

[0020] The supply chain can be a warehousing or third-party logistics company. The supply chain dynamically replenishes goods based on the generated replenishment and delivery routes. In other words, the supply chain initiates the actual delivery operation based on the replenishment and delivery routes issued by the central server, including scheduling vehicles, personnel, inventory and other resources to ensure that the replenishment needs of stores are met within the specified time window.

[0021] Furthermore, based on the replenishment times and replenishment forecasts for the multiple sets of goods, a dynamic supply chain fit is performed to output a replenishment and delivery route. The method includes: Based on the consistency of product coding, single-category product clusters are extracted from the replenishment times and replenishment forecasts of multiple groups of products to obtain P groups of replenishment times and P groups of replenishment forecasts for P types of products to be replenished; the replenishment times of the P groups of products are used as delivery time window constraints, and low-energy routes with limited load capacity are fitted according to the geographical distribution of the multiple chain stores to output P single-category routes, where each store along each single-category route is associated with the replenishment forecast of the corresponding single-category product; the P single-category routes are optimized based on spatiotemporal fusion to output the replenishment delivery route.

[0022] All products are categorized using product codes, which represent product categories, specifications, and other characteristics. Consistent product codes require products to belong to the same category. Based on product code consistency, all products requiring replenishment are divided into several product clusters. Each product cluster contains one or more products, and the replenishment time and replenishment forecast quantity are the same for each product cluster. Assuming there are P types of products to be replenished, then there will be P product clusters, where P is a positive integer. Each product cluster has a corresponding product replenishment time and product replenishment forecast quantity. The product replenishment time indicates the point in time when the inventory of this type of product needs to be replenished, and the product replenishment forecast quantity indicates the quantity of this type of product that needs to be replenished.

[0023] The replenishment time of group P goods is used as a delivery time window constraint, meaning that each product cluster must be delivered within a specific time period. For example, the replenishment time for a certain product cluster may be set from the 1st to the 5th of each month, while the replenishment times for other product clusters may not overlap. Based on the geographical distribution of multiple chain stores, including the city and region where each store is located, preliminary delivery routes are selected to ensure that each chain store receives replenishment in the shortest possible time. Simultaneously, considering the limited load capacity of each delivery vehicle, the load capacity and volume restrictions of each delivery route are taken into account during route planning to avoid overloading during transportation and maximize the utilization efficiency of delivery vehicles.

[0024] Based on delivery time window constraints and limited load capacity, low-energy routes are fitted, that is, the routes that save the most energy and cost are selected. The analysis is combined with path optimization algorithms to reduce fuel consumption, shorten transportation distance and improve delivery efficiency. Based on the above factors, corresponding single-category routes are generated for each product cluster. Each single-category route corresponds to a product cluster, and each single-category route includes the replenishment quantity of that product cluster in each store.

[0025] Spatiotemporal fusion optimization refers to the comprehensive analysis of constraints in both time and space dimensions to optimize delivery routes. Since replenishment times for multiple product clusters may overlap, the overlapping portions of time windows for each product cluster are integrated to reduce resource waste and increase route efficiency. Delivery routes for different product clusters may pass through the same or nearby stores; by analyzing the geographical locations of each store, overlapping portions of routes are optimized to avoid duplicate deliveries and reduce transportation costs. Based on spatiotemporal fusion analysis, P single-category routes are merged and optimized to reduce redundancy between individual product category routes, thereby improving delivery efficiency. For example, adjacent portions of delivery paths for multiple product clusters are merged into a larger, more efficient delivery route. After optimization, the final replenishment delivery route is output. This route includes the delivery time, delivery sequence, stores passed through, and the required replenishment quantity for each product cluster. This replenishment delivery route maximizes transportation efficiency, ensures stores receive replenishment within the specified time, and reduces costs and energy consumption.

[0026] Furthermore, the method for optimizing the P single-category routes based on spatiotemporal fusion and outputting the replenishment and delivery routes includes: After enumerating and combining the P single-category routes, the results are performed based on the overlap ratio of time windows and geographical proximity. Calculation of the spatiotemporal coupling degree of single-category routes, output. Spatiotemporal coupling degree; taking the P single-category routes as vertices, with the... Using spatiotemporal coupling degree as edge weight, construct a route coupling association topology; remove edges in the route coupling association topology that are less than a preset fusion threshold to obtain M connected subgraphs; based on the vertex composition of the M connected subgraphs, divide the P single-category routes into M groups of single-category routes; perform cross-route node insertion operations on the M groups of single-category routes to generate M fused routes, which serve as the replenishment and delivery routes.

[0027] Enumerate and combine P single-category routes, that is, combine all P single-category routes in pairs to obtain Each group of single-category routes is a pair of any two single-category routes. For example, if P=3, there are 3(3-1) / 2=3 combinations, namely: route 1 and route 2, route 1 and route 3, and route 2 and route 3.

[0028] For each set of single-category routes, a spatiotemporal coupling degree is calculated between the two routes. Specifically, this involves calculating the overlap ratio of their time windows, i.e., whether the replenishment times of the two routes overlap and the proportion of the overlap to the total time window. If the delivery times of the two routes are similar, or they pass through some stores within the same time period, their time window overlap ratio is high. Geographic proximity is also calculated, referring to the degree of geographical proximity of the delivery paths of the two routes. If the two routes pass through similar stores or are located in nearby geographical areas, their geographic proximity is high. Based on the time window overlap ratio and geographic proximity, the spatiotemporal coupling degree of this set of routes is output. The spatiotemporal coupling degree is obtained by weighted summation of the time window overlap ratio and geographic proximity. The weights reflect the importance of the time window overlap ratio and geographic proximity in the coupling degree calculation and are set according to specific requirements. Through the spatiotemporal coupling degree calculation for each set of single-category routes, the final output is... Spatiotemporal coupling degree.

[0029] Using P single-item category routes as vertices in the topology graph, each vertex represents a replenishment and delivery route for a product cluster. The calculated... The spatiotemporal coupling degree is used as the edge weight between vertices in the topology graph. That is, the spatiotemporal coupling degree between each two single-category routes is used as the weight of the edge between them, which represents the tightness between these two single-category routes. In this way, a weighted undirected topology graph is constructed, namely, a route coupling and association topology.

[0030] A preset fusion threshold is set based on business needs. Only when the spatiotemporal coupling between two routes exceeds this threshold are they considered to belong to the same group and eligible for fusion. Based on the calculated spatiotemporal coupling, edges with spatiotemporal coupling below the preset fusion threshold are removed. That is, for each pair of routes, if their spatiotemporal coupling is low, it indicates they are not strongly correlated in time or space, and the edges between them are deleted, meaning these two routes are not suitable for merging. After removing edges with low spatiotemporal coupling, the graph is left with M connected subgraphs, where M is a positive integer. Each connected subgraph represents a group of highly coupled single-category routes. These single-category routes can be merged or optimized. The vertices (i.e., single-category routes) in each connected subgraph can be considered as a delivery group, exhibiting strong spatiotemporal coupling and suitable for resource scheduling and route merging.

[0031] The P single-category routes are divided into M groups of single-category routes based on the vertices (i.e., single-category routes) in each connected subgraph. Each group of single-category routes contains routes that belong to the same connected subgraph. The spatiotemporal coupling between these routes is high enough that they can be processed and optimized together in subsequent operations. For example, if a connected subgraph contains route 1, route 2, and route 3, these three routes will be assigned to the same group. Routes within each group share resources and delivery windows, allowing for further merging and optimization.

[0032] Cross-route node insertion refers to optimizing route connectivity by inserting nodes (i.e., stores) when merging multiple single-product-category routes into a longer route. Each merged route includes delivery tasks from multiple stores, representing joint delivery of different product clusters. Specifically, assuming a single-product-category route contains multiple stores, a more optimized merged route is generated by inserting appropriate delivery nodes during the merging process. The cross-route node insertion operation includes: inserting appropriate stores between different routes to make the merged route smoother and avoid unnecessary detours; after inserting nodes, reordering the delivery sequence and optimizing the route to reduce transportation distance and time. The resulting M merged routes are used as replenishment delivery routes for the supply chain to execute actual delivery tasks.

[0033] Furthermore, the method involves performing cross-route node insertion operations on the M groups of single-category routes to generate M merged routes, which serve as the replenishment and delivery routes. Calculate the first detour increment of the first single-item category route and the second single-item category route; if the first detour increment is less than a preset increment scale, perform cross-route node insertion operation on the first single-item category route and the second single-item category route to generate a first iterative route; similarly calculate the second detour increment of the first iterative route and the third single-item category route; if the second detour increment is greater than a preset increment scale, treat the third single-item category route as an independent delivery route; similarly traverse the first group of single-item category routes, iteratively perform cross-route node insertion operation until a first integrated route and multiple independent delivery routes are generated; add the first integrated route and multiple independent delivery routes to the replenishment delivery route.

[0034] The first and second product category routes are any pair of product category routes in group M. As the current analysis object, the first and second product category routes are merged to simulate a new delivery route and calculate the first detour increment brought about by the merger. This first detour increment refers to the additional distance or time caused by the merger, that is, the increased distance or time of the merged route compared to the original routes. This is because the merged route becomes less direct or requires detouring through certain areas during the merger process.

[0035] The preset increment scale is a set threshold used to measure whether the increased detour increment when merging two routes is worthwhile. Only when the increased distance or time of the merged route is less than this preset increment scale is the merge considered worthwhile, and the merged route remains valid. If the first detour increment is less than the preset increment scale, a cross-route node insertion operation is performed on the first and second single-category routes. That is, a node (i.e., a store) from one single-category route is inserted into the other single-category route. This insertion operation ensures the continuity between the two routes and improves transportation efficiency. After inserting the node, the first iteration route is generated, which is the optimized path after merging.

[0036] The third single-category route is any route other than the first iteration route and the second iteration route. The first iteration route and the third single-category route are merged, and the merged path is simulated again to calculate the new second detour increment, similar to the detour increment calculation in the previous step. The purpose is to determine how much additional travel time or distance the merged path adds.

[0037] If the second detour increment is greater than the preset increment scale, the third single-item route will be treated as an independent delivery route instead of being merged with the previous first iteration route. In this way, the three single-item routes will not be integrated with the current route, but will run independently to avoid the increased detour distance being too long.

[0038] The first group of single-item routes is any one of the M groups of single-item routes. Within this first group, each single-item route is traversed sequentially using the steps described above for merging. After each operation, a new detour increment is calculated, and a cross-route node insertion operation is performed according to a preset increment scale to optimize the merged delivery path. If the merged detour increment is less than the preset increment scale, merging continues; if the detour increment is too large, the new single-item route is treated as an independent delivery route, and merging is stopped. After traversing all routes in the first group of single-item routes, the first merged route is obtained. This first merged route is the optimal delivery path after merging. Multiple independent delivery routes are also generated, each representing an unmerged single-item route.

[0039] The first integrated route and several independent delivery routes were added to the replenishment delivery route, which includes replenishment tasks for all stores, ensuring that each store can receive the goods it needs on time, while reducing the total delivery distance and time.

[0040] Furthermore, if the first detour increment is less than a preset increment scale, then a cross-route node insertion operation is performed on the first single-category route and the second single-category route to generate a first iterative route. The method includes: The feasibility of insertion is verified based on the compatibility of the storage temperature layer of the first and second single-category routes. If the verification is successful, the cross-route node insertion operation of the first and second single-category routes is performed to generate the first iterative route.

[0041] Each product has specific storage requirements, especially the need for transportation and storage at specific temperatures, such as refrigeration, ambient temperature, and freezing. Category-specific storage temperatures refer to the temperature or environmental conditions required for different products. The storage temperature requirements of products in the first and second product category routes are checked. For example, if the first product category route includes refrigerated products and the second product category route includes ambient temperature products, the compatibility of these two types of product storage temperatures is checked to determine if they can be delivered under the same transportation conditions. If the products in the two routes require the same or compatible storage conditions, they can be merged; otherwise, they cannot be merged.

[0042] After successful verification, a cross-route node insertion operation is performed. The core of this operation is to insert a delivery node (store) from one product category route into another. Node insertion is based on geographical location, delivery order, and product demand, ensuring that the merged route maintains product temperature compatibility while optimizing delivery routes and reducing detours and time delays. After node insertion, the first iteration route is generated. This optimized route represents the merged delivery path. The first iteration route integrates the first and second product category routes, reducing transportation costs and time through node insertion.

[0043] Furthermore, based on the cumulative calculation results of product sales volume from the multiple POS transaction data, the virtual inventory of the multiple chain stores is updated to obtain multiple sets of real-time product inventory data. The method includes: Based on the multiple cash register transaction data, the sales volume of goods is accumulated and calculated to obtain the accumulated sales information of multiple goods; multiple goods entry records of the multiple chain stores are retrieved; based on the multiple goods entry records and the accumulated sales information of multiple goods, the virtual inventory is updated according to the product code dimension to obtain the multiple sets of real-time goods inventory data.

[0044] The system aggregates the cash flow data from multiple chain stores, that is, it summarizes the sales volume of the same product in different chain stores to obtain the total sales volume of the product in all stores. Through this aggregation, the system obtains the cumulative sales information of each product, which is used for subsequent inventory updates and replenishment forecasts.

[0045] The system retrieves product receipt records from the inventory management systems of multiple chain stores. These records include information such as the receipt time, quantity, and batch number for each product.

[0046] The product code is a unique identifier for each product. Products are categorized and updated based on the product code. For each product, the cumulative sales information of the product is combined with the product's inventory record to calculate the real-time inventory of the product. The real-time inventory is the product's inventory minus the cumulative sales. Through calculation, the real-time product inventory data of each chain store is updated. The real-time product inventory data reflects the current inventory status of each chain store, including the current stock of each product.

[0047] Furthermore, based on multiple sets of time-series commodity inventory data obtained through backtracking and the multiple sets of real-time commodity inventory data, inventory demand forecasting is performed to obtain multiple sets of commodity replenishment times and multiple sets of commodity replenishment forecast quantities. The method includes: Using multiple sets of product codes from the multiple sets of real-time product inventory data as search criteria, multiple replenishment baselines are obtained. These replenishment baselines are then used to verify the basic replenishment demand of the multiple sets of real-time product inventory data. If the verification fails, an immediate replenishment instruction is triggered. If the verification passes, after backtracking multiple sets of time-series product inventory data based on the multiple sets of product codes, inventory demand is predicted based on the multiple sets of time-series product inventory data and the multiple sets of real-time product inventory data, resulting in the replenishment time and predicted replenishment quantity for each set of products.

[0048] The product code of each product is extracted and used as the keyword for retrieval. The product code is a unique identifier for the product and can help to accurately retrieve relevant information for each product. Using the product code, the replenishment baseline of the corresponding product can be retrieved from sources such as historical inventory data and supply chain management systems. The replenishment baseline represents the minimum inventory level of each product and is set based on historical sales, seasonal changes and market demand forecasts.

[0049] For each product, its current real-time inventory data is compared with the replenishment baseline, which represents the minimum inventory level. The real-time inventory data reflects the current inventory status. If the real-time inventory data is lower than the replenishment baseline, it means that the product needs to be replenished; if the real-time inventory data is higher than the replenishment baseline, it means that the product is in surplus and does not need to be replenished.

[0050] When real-time inventory data falls below the replenishment baseline, the verification fails, triggering an immediate replenishment instruction. This instruction is used to notify relevant departments such as the supply chain, warehouse, and logistics to perform replenishment operations. The immediate replenishment instruction includes information such as the specific product, quantity, and estimated replenishment time.

[0051] If the real-time inventory data is higher than the replenishment baseline, the verification passes. In this case, the time-series inventory data of the product is traced back according to the product code, that is, the historical inventory data. By analyzing the past inventory changes, the future demand and replenishment pattern of the product can be predicted.

[0052] Based on the backtested time-series merchandise inventory data, and combined with the current real-time merchandise inventory data, inventory demand is predicted. Specifically, by comparing past sales trends with the current inventory status, future replenishment demand is predicted, resulting in merchandise replenishment time and merchandise replenishment forecast quantity. The merchandise replenishment time is the point in time when the merchandise inventory is expected to fall below the replenishment baseline, and the merchandise replenishment forecast quantity is the quantity of merchandise that is expected to be replenished when the replenishment time arrives.

[0053] Furthermore, after backtracking multiple sets of time-series commodity inventory data based on the multiple sets of commodity codes, inventory demand forecasting is performed based on the multiple sets of time-series commodity inventory data and multiple sets of real-time commodity inventory data to obtain the replenishment time and predicted replenishment quantity of the multiple sets of commodities. The method includes: The inventory data of a first-time-series commodity is traced back based on the first commodity code; the inventory data of the first-time-series commodity is divided into multiple first-time-series data segments based on a preset short-period window; the multiple first-time-series data segments are used as training data to construct a first commodity consumption prediction network based on an LSTM model; the first replenishment baseline is used as the iterative convergence condition, and the real-time inventory data of the first commodity is used as the initial input. The first commodity consumption prediction network is run in a forward iterative manner. When the simulated inventory level drops to the first replenishment baseline, the simulated time is output as the first commodity replenishment time; a first replenishment cycle is predefined and used as the convergence condition. The first commodity consumption prediction network is run in a backward optimization manner. When the simulated time reaches the first replenishment cycle, the simulated inventory level is output as the first commodity replenishment prediction level.

[0054] The first product code is any one of multiple product codes. It is used as the current analysis object. The first product code is used as the query condition to retrieve the inventory data of the product in the past period from the database or historical inventory records, and obtain the first time series product inventory data. The first time series product inventory data contains the inventory changes of the product at several points in the past.

[0055] A preset short-cycle window refers to a relatively short time period, such as 7 days or 14 days, as the basic unit for data segmentation. The length of this preset short-cycle window can be set according to the sales cycle of the product and the frequency of inventory fluctuations. During the sliding window process, the preset short-cycle window is used to segment the inventory data of the first time series product. For example, assuming the window length is 7 days, the data from day 1 to day 7 is taken as one segment, the data from day 2 to day 8 is taken as the second segment, and so on. By sliding the window, multiple first-time series data segments are obtained, each segment representing the inventory change data within a preset short-cycle window.

[0056] Multiple time-series data segments are used as training data for an LSTM (Long Short-Term Memory) model. Each data segment contains inventory information for a product within a time window, used to predict the product's sales and inventory consumption in the future. The LSTM model is a recurrent neural network used for forecasting time-series data. LSTM can handle long-term dependencies and is suitable for product inventory and consumption forecasting. By inputting multiple time-series data segments into the LSTM model, the model learns the product's sales patterns, periodic changes, and trends, resulting in a first product consumption prediction network. This first product consumption prediction network is used to predict the product's consumption over a future period, thus providing a reference for replenishment decisions.

[0057] The first replenishment baseline represents the minimum inventory level of the product. Using the real-time inventory data of the first product as the starting state of the first product consumption prediction network, a forward iteration is performed. In each iteration, the first product consumption prediction network predicts the inventory change of the product in the future period and generates a new inventory value. The inventory change is simulated through forward iteration until the simulated inventory level drops to the first replenishment baseline. At this point, the simulated time point is output, representing the first product replenishment time when the product inventory drops to the replenishment baseline. The first product replenishment time represents the time node when the product inventory is insufficient and replenishment is required. This time point serves as a reference for subsequent replenishment operations.

[0058] The first replenishment cycle is a predefined time period, set based on factors such as the product's sales cycle and supply chain cycle. For example, the replenishment cycle could be 7 days, 14 days, etc., serving as the time frame for replenishment decisions. Backward optimization refers to calculating the product's inventory changes backward from the target time point (i.e., the end of the first replenishment cycle). In this way, it predicts whether the product's inventory will reach the replenishment baseline within the specified replenishment cycle. During this process, inventory changes are simulated until the simulated time reaches the first replenishment cycle. At this point, the simulated inventory level is output, which is the current inventory level of the product. Based on the difference between the simulated inventory level and the first replenishment baseline, the quantity of products that need to be replenished is calculated, i.e., the first product replenishment forecast, representing the quantity of products that need to be replenished at the end of the specified replenishment cycle.

[0059] Furthermore, based on the multiple cash register transaction data, the method involves cumulatively calculating the sales volume of goods to obtain cumulative sales information for multiple goods, including: After filtering out non-sales transaction records from the first cashier transaction data, the first cashier transaction data is decomposed based on the product code to obtain multiple time-series sales volume data for various products on sale; the multiple time-series sales volume data are accumulated and calculated, and the cumulative sales volume of multiple products is output as the cumulative sales information of the first product.

[0060] Non-sales transaction records include refund records, inventory records, gift records, and other non-sales related transaction data. These non-sales transaction records are removed from the first POS transaction data, retaining only the actual sales data. The product code is a unique identifier for each product. The first POS transaction data is decomposed using product codes, matching the product in each transaction with its corresponding product code to extract the sales data for each product within different time periods. For example, a transaction might include products A, B, and C; the sales data for each product is extracted separately. For each product, time-series sales volume data is generated based on the timestamp of the sales data, representing the number of products sold within a specific time period.

[0061] The cumulative sales volume of each product is calculated by summing the time-series sales data. This process can be performed at different time granularities such as daily, weekly, and monthly. For example, the cumulative sales volume of a product over the past 30 days is the cumulative sales volume of that product over 30 days. Finally, the cumulative sales volume of multiple products is output, which is the total sales volume of each product within a specified time period. These data are aggregated as the cumulative sales information of the first product and used for subsequent analysis such as inventory updates, replenishment decisions, and sales forecasting.

[0062] Example 2, based on the same inventive concept as the replenishment demand dynamic forecasting method for multi-node time-series data fusion in the aforementioned examples, such as... Figure 2 As shown in the figure, this application provides a dynamic replenishment demand forecasting system for multi-node time-series data fusion, the system comprising: Deployment module 10 is used to deploy multiple lightweight store agent agents in the back-end of multiple store POS terminals of multiple chain stores. Each lightweight store agent agent is responsible for real-time capture and encryption of POS transaction data. Data processing module 20, after the central server receives the multiple POS transaction data pushed in real-time by the multiple lightweight store agent agents via MQTT, executes the following steps: Virtual inventory update unit 21 is used to execute step a: based on the cumulative calculation result of the product sales volume of the multiple POS transaction data, update the virtual inventory of the multiple chain stores. Multiple sets of real-time commodity inventory data are obtained; the inventory demand forecasting unit 22 is used to execute step b: based on the multiple sets of time-series commodity inventory data obtained through backtracking and the multiple sets of real-time commodity inventory data, perform inventory demand forecasting to obtain multiple sets of commodity replenishment time and multiple sets of commodity replenishment forecast quantity; the dynamic fitting unit 23 is used to execute step c: perform supply chain dynamic fitting based on the multiple sets of commodity replenishment time and multiple sets of commodity replenishment forecast quantity, and output replenishment delivery routes; the dynamic replenishment module 30, the supply chain side responds to the replenishment delivery routes issued by the central server to perform dynamic replenishment of the multiple chain stores.

[0063] Furthermore, the dynamic fitting unit 23 is used to perform the following operation steps: Based on the consistency of product coding, single-category product clusters are extracted from the replenishment times and replenishment forecasts of multiple groups of products to obtain P groups of replenishment times and P groups of replenishment forecasts for P types of products to be replenished; the replenishment times of the P groups of products are used as delivery time window constraints, and low-energy routes with limited load capacity are fitted according to the geographical distribution of the multiple chain stores to output P single-category routes, where each store along each single-category route is associated with the replenishment forecast of the corresponding single-category product; the P single-category routes are optimized based on spatiotemporal fusion to output the replenishment delivery route.

[0064] Furthermore, the dynamic fitting unit 23 is used to perform the following operation steps: After enumerating and combining the P single-category routes, the results are performed based on the overlap ratio of time windows and geographical proximity. Calculation of the spatiotemporal coupling degree of single-category routes, output. Spatiotemporal coupling degree; taking the P single-category routes as vertices, with the... Using spatiotemporal coupling degree as edge weight, construct a route coupling association topology; remove edges in the route coupling association topology that are less than a preset fusion threshold to obtain M connected subgraphs; based on the vertex composition of the M connected subgraphs, divide the P single-category routes into M groups of single-category routes; perform cross-route node insertion operations on the M groups of single-category routes to generate M fused routes, which serve as the replenishment and delivery routes.

[0065] Furthermore, the dynamic fitting unit 23 is used to perform the following operation steps: Calculate the first detour increment of the first single-item category route and the second single-item category route; if the first detour increment is less than a preset increment scale, perform cross-route node insertion operation on the first single-item category route and the second single-item category route to generate a first iterative route; similarly calculate the second detour increment of the first iterative route and the third single-item category route; if the second detour increment is greater than a preset increment scale, treat the third single-item category route as an independent delivery route; similarly traverse the first group of single-item category routes, iteratively perform cross-route node insertion operation until a first integrated route and multiple independent delivery routes are generated; add the first integrated route and multiple independent delivery routes to the replenishment delivery route.

[0066] Furthermore, the dynamic fitting unit 23 is used to perform the following operation steps: The feasibility of insertion is verified based on the compatibility of the storage temperature layer of the first and second single-category routes. If the verification is successful, the cross-route node insertion operation of the first and second single-category routes is performed to generate the first iterative route.

[0067] Furthermore, the virtual inventory update unit 21 is used to perform the following operation steps: Based on the multiple cash register transaction data, the sales volume of goods is accumulated and calculated to obtain the accumulated sales information of multiple goods; multiple goods entry records of the multiple chain stores are retrieved; based on the multiple goods entry records and the accumulated sales information of multiple goods, the virtual inventory is updated according to the product code dimension to obtain the multiple sets of real-time goods inventory data.

[0068] Furthermore, the inventory demand forecasting unit 22 is used to perform the following operational steps: Using multiple sets of product codes from the multiple sets of real-time product inventory data as search criteria, multiple replenishment baselines are obtained. These replenishment baselines are then used to verify the basic replenishment demand of the multiple sets of real-time product inventory data. If the verification fails, an immediate replenishment instruction is triggered. If the verification passes, after backtracking multiple sets of time-series product inventory data based on the multiple sets of product codes, inventory demand is predicted based on the multiple sets of time-series product inventory data and the multiple sets of real-time product inventory data, resulting in the replenishment time and predicted replenishment quantity for each set of products.

[0069] Furthermore, the inventory demand forecasting unit 22 is used to perform the following operational steps: The inventory data of a first-time-series commodity is traced back based on the first commodity code; the inventory data of the first-time-series commodity is divided into multiple first-time-series data segments based on a preset short-period window; the multiple first-time-series data segments are used as training data to construct a first commodity consumption prediction network based on an LSTM model; the first replenishment baseline is used as the iterative convergence condition, and the real-time inventory data of the first commodity is used as the initial input. The first commodity consumption prediction network is run in a forward iterative manner. When the simulated inventory level drops to the first replenishment baseline, the simulated time is output as the first commodity replenishment time; a first replenishment cycle is predefined and used as the convergence condition. The first commodity consumption prediction network is run in a backward optimization manner. When the simulated time reaches the first replenishment cycle, the simulated inventory level is output as the first commodity replenishment prediction level.

[0070] Furthermore, the virtual inventory update unit 21 is used to perform the following operation steps: After filtering out non-sales transaction records from the first cashier transaction data, the first cashier transaction data is decomposed based on the product code to obtain multiple time-series sales volume data for various products on sale; the multiple time-series sales volume data are accumulated and calculated, and the cumulative sales volume of multiple products is output as the cumulative sales information of the first product.

[0071] Through the foregoing detailed description of the method for dynamic prediction of replenishment demand based on multi-node time-series data fusion, those skilled in the art can clearly understand the system for dynamic prediction of replenishment demand based on multi-node time-series data fusion in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to in the method section.

[0072] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for dynamic forecasting of replenishment demand based on multi-node time-series data fusion, characterized in that, The method includes: Multiple lightweight store agent agents are deployed in the back-end of the POS terminals of multiple chain stores. The lightweight store agent agents are responsible for capturing and encrypting the POS transaction data in real time. After receiving multiple POS transaction data pushed in real time via MQTT from the multiple lightweight store agent agents, the central server performs the following steps: Step a: Based on the cumulative calculation result of the sales volume of the products from the multiple cash register transaction data, update the virtual inventory of the multiple chain stores to obtain multiple sets of real-time product inventory data; Step b: Based on the backtracking of multiple sets of time-series commodity inventory data and the multiple sets of real-time commodity inventory data, perform inventory demand forecasting to obtain multiple sets of commodity replenishment time and multiple sets of commodity replenishment forecast quantity; Step c: Perform dynamic fitting of the supply chain based on the replenishment time and replenishment forecast of the multiple sets of goods, and output the replenishment and delivery route; The supply chain responds to the replenishment and delivery routes issued by the central server to dynamically replenish the multiple chain stores.

2. The method for dynamic prediction of replenishment demand based on multi-node time-series data fusion as described in claim 1, characterized in that, Based on the multiple sets of product replenishment times and multiple sets of product replenishment forecasts, a dynamic supply chain fitting is performed to output a replenishment delivery route. The method includes: Based on the consistency of product codes, extract single-category product clusters from the replenishment time and replenishment forecast of multiple groups of products to obtain P groups of replenishment time and P groups of replenishment forecast of P types of products to be replenished. Using the replenishment time of the P groups of goods as the delivery time window constraint, and fitting low-energy routes with limited load capacity based on the geographical distribution of the multiple chain stores, P single-category routes are output, where each store along each single-category route is associated with the corresponding single-category goods replenishment forecast. Based on spatiotemporal fusion optimization of the P single-category routes, the replenishment and delivery routes are output.

3. The method for dynamic prediction of replenishment demand based on multi-node time-series data fusion as described in claim 2, characterized in that, The method optimizes the routes for the P individual product categories based on spatiotemporal fusion and outputs the replenishment and delivery routes, including: After enumerating and combining the P single-category routes, the results are performed based on the overlap ratio of time windows and geographical proximity. Calculation of the spatiotemporal coupling degree of single-category routes, output. Spatiotemporal coupling degree; Using the P single-category routes as vertices, and the... Using spatiotemporal coupling degrees as edge weights, a route coupling and association topology is constructed. Remove edges smaller than a preset fusion threshold from the route coupling topology to obtain M connected subgraphs; Based on the vertex composition of the M connected subgraphs, the P single-category routes are divided into M groups of single-category routes; Perform a cross-route node insertion operation on the M groups of single-category routes to generate M merged routes, which will serve as the replenishment and delivery routes.

4. The method for dynamic forecasting of replenishment demand based on multi-node time-series data fusion as described in claim 3, characterized in that, The method involves performing a cross-route node insertion operation on the M groups of single-category routes to generate M merged routes, which serve as the replenishment and delivery routes. Calculate the first detour increment when merging the first single-category route and the second single-category route; If the first detour increment is less than the preset increment scale, then the cross-route node insertion operation of the first single-category route and the second single-category route is performed to generate the first iterative route; The second detour increment for the first iteration route and the third single-category route can be calculated by analogy; If the second detour increment is greater than the preset increment scale, then the third single-item route will be treated as an independent delivery route. By analogy, traverse the first group of single-category routes and iteratively perform cross-route node insertion operations until the first merged route and multiple independent delivery routes are generated. Add the first merged route and multiple independent delivery routes to the replenishment delivery route.

5. The method for dynamic prediction of replenishment demand based on multi-node time-series data fusion as described in claim 4, characterized in that, If the first detour increment is less than a preset increment scale, then the cross-route node insertion operation of the first single-category route and the second single-category route is performed to generate the first iterative route. The method includes: The feasibility of insertion is verified based on the compatibility of the product category storage temperature layer between the first and second product category routes. If the verification passes, the cross-route node insertion operation of the first single-category route and the second single-category route is performed to generate the first iterative route.

6. The method for dynamic forecasting of replenishment demand based on multi-node time-series data fusion as described in claim 1, characterized in that, Based on the cumulative sales volume calculation results of the multiple POS transaction data, the virtual inventory of the multiple chain stores is updated to obtain multiple sets of real-time product inventory data. The method includes: Based on the multiple cash register transaction data, the cumulative sales volume of goods is calculated to obtain the cumulative sales information of multiple goods. Retrieve multiple product inbound records from the aforementioned multiple chain stores; Based on the multiple product inbound records and the cumulative sales information of multiple products, the virtual inventory is updated according to the product code dimension to obtain the multiple sets of real-time product inventory data.

7. The method for dynamic forecasting of replenishment demand based on multi-node time-series data fusion as described in claim 1, characterized in that, Based on multiple sets of time-series commodity inventory data obtained through backtracking and the multiple sets of real-time commodity inventory data, inventory demand forecasting is performed to obtain multiple sets of commodity replenishment times and multiple sets of commodity replenishment forecast quantities. The method includes: Using the multiple sets of product codes from the multiple sets of real-time product inventory data as search criteria, multiple sets of replenishment baselines are obtained. The multiple replenishment baselines are used to verify the basic replenishment demand of the multiple sets of real-time commodity inventory data. If the verification fails, an immediate replenishment order will be triggered; If the verification passes, after backtracking multiple sets of time-series product inventory data based on the multiple sets of product codes, inventory demand is predicted based on the multiple sets of time-series product inventory data and multiple sets of real-time product inventory data to obtain the replenishment time and predicted replenishment quantity of the multiple sets of products.

8. The method for dynamic prediction of replenishment demand based on multi-node time-series data fusion as described in claim 7, characterized in that, After backtracking multiple sets of time-series commodity inventory data based on the multiple sets of commodity codes, inventory demand forecasting is performed based on the multiple sets of time-series commodity inventory data and multiple sets of real-time commodity inventory data to obtain the replenishment time and predicted replenishment quantity for the multiple sets of commodities. The method includes: Backtrack the first time-series product inventory data based on the first product code; The first time series commodity inventory data is divided into multiple first time series data segments by sliding segmentation based on a preset short period window. Using the multiple first time-series data segments as training data, a first commodity consumption prediction network is constructed based on the LSTM model; The first replenishment baseline is used as the convergence condition for the iteration, and the real-time inventory data of the first commodity is used as the initial input. The consumption prediction network of the first commodity is run in a forward iterative manner. When the simulated inventory drops to the first replenishment baseline, the simulation time is output as the replenishment time of the first commodity. A first replenishment cycle is predefined and used as a convergence condition. The first product consumption prediction network is run through backward optimization. When the simulation time reaches the first replenishment cycle, the simulated inventory is output as the first product replenishment prediction quantity.

9. The method for dynamic prediction of replenishment demand based on multi-node time-series data fusion as described in claim 6, characterized in that, Based on the multiple cash register transaction data, the cumulative sales volume of goods is calculated to obtain cumulative sales information for multiple goods. The method includes: After filtering out non-sales transaction records from the first cashier transaction data, the first cashier transaction data is decomposed based on the product code to obtain multiple time-series sales volume data for various products on sale. The multiple time-series sales data are accumulated and calculated, and the cumulative sales volume of multiple products is output as the cumulative sales information of the first product.

10. A dynamic replenishment demand forecasting system for multi-node time-series data fusion, characterized in that, The system is used to implement the replenishment demand dynamic forecasting method for multi-node time-series data fusion as described in any one of claims 1-9, the system comprising: The deployment module is used to deploy multiple lightweight store agent agents in the back-end of multiple store POS terminals in multiple chain stores. The lightweight store agent agents are responsible for capturing and encrypting POS transaction data in real time. The data processing module, after the central server receives multiple POS transaction data pushed in real time via MQTT from the multiple lightweight store agent agents, performs the following steps: The virtual inventory update unit is used to perform step a: based on the cumulative calculation result of the sales volume of the products in the multiple cash register data, update the virtual inventory of the multiple chain stores to obtain multiple sets of real-time product inventory data; The inventory demand forecasting unit is used to perform step b: based on the backtracking obtained multiple sets of time-series commodity inventory data and the multiple sets of real-time commodity inventory data, perform inventory demand forecasting to obtain multiple sets of commodity replenishment time and multiple sets of commodity replenishment forecast quantity; The dynamic fitting unit is used to perform step c: perform dynamic fitting of the supply chain based on the multiple sets of product replenishment time and multiple sets of product replenishment forecast quantity, and output the replenishment and delivery route. The dynamic replenishment module enables the supply chain to dynamically replenish the multiple chain stores in response to the replenishment and delivery routes issued by the central server.

Images