Retail inventory optimization method based on full-link management

Abstract

The invention relates to the technical field of inventory optimization, and discloses a retail inventory optimization method based on full-link management, which comprises the following steps: collecting historical data to construct a full-link graph structure, and predicting the sales demand of each node in a future period based on a graph convolutional network and a time sequence model; acquiring real-time data of each link in a whole link, and determining an initial safety stock, a replenishment period and an order quantity based on the whole link graph structure and in combination with the sales demand data; and based on an inventory optimization algorithm, iteratively solving the optimal safe inventory, the replenishment cycle and the order quantity, and executing an inventory optimization operation according to an optimization result. According to the method, the graph convolutional network and the quantum optimization algorithm are innovatively combined, dynamic inventory optimization based on full-link management is realized, the intelligent level of inventory management is improved, and the method has remarkable advantages in precision, response speed and cost control; a brand-new and efficient inventory optimization solution is provided for retail enterprises.

Description

Technical Field

[0001] The present invention relates to the technical field of inventory optimization, and in particular to a retail inventory optimization method based on full-link management. Background Art

[0002] With the continuous development of the retail industry and supply chain management, inventory management plays a vital role in ensuring the supply of goods and reducing costs. Traditional inventory management methods usually rely on static inventory strategies, which often fail to fully consider factors such as dynamically changing market demand, supply chain fluctuations, and transportation time. With the application of big data technology and machine learning methods, more and more retail companies have begun to adopt data-driven inventory optimization strategies. Although traditional inventory optimization algorithms, such as economic order quantity (EOQ), just-in-time inventory control (JIT), and management methods based on demand forecasting, can effectively reduce inventory holding costs and out-of-stock risks, these methods often do not fully consider the complex relationships and real-time changes of various links in the supply chain. In addition, traditional methods are slow to respond to factors such as sales demand fluctuations, market trends, seasonal fluctuations, and promotional activities, and often rely on manual adjustments, resulting in poor flexibility and response speed of inventory management.

[0003] With the continuous development of graph neural networks (GNN) and quantum optimization algorithms, new inventory management methods have emerged. Graph convolutional networks (GCN) can effectively capture the complex relationships between supply chain nodes, while quantum optimization algorithms can find the optimal solution in a more efficient way in the solution space of large-scale problems. By combining these advanced technologies, inventory strategies can be optimized more accurately, and safety stocks, replenishment cycles, and order quantities can be adjusted dynamically, thereby achieving more efficient inventory management. Therefore, the present invention proposes a retail inventory optimization method based on full-link management. Summary of the invention

[0004] In response to the problems in the related technology, the present invention proposes a retail inventory optimization method based on full-link management to overcome the above-mentioned technical problems existing in the existing related technology.

[0005] To this end, the specific technical solution adopted by the present invention is as follows:

[0006] A retail inventory optimization method based on full-link management includes the following steps:

[0007] S1. Collect historical data, build a full-link graph structure, and predict the sales demand of each node in the future period based on the graph convolutional network and time series model;

[0008] S2. Obtain real-time data from each link in the entire chain, and determine the initial safety stock, replenishment cycle, and order quantity based on the full-chain graph structure and combined with sales demand data;

[0009] S3. Based on the inventory optimization algorithm, iteratively solve the optimal safety stock, replenishment cycle and order quantity, and perform inventory optimization operations based on the optimization results.

[0010] Preferably, the collecting of historical data, constructing a full-link graph structure, and predicting the sales demand of each node in the future period based on a graph convolutional network and a time series model include the following steps:

[0011] S11. Collect data on historical sales, market trends, seasonal fluctuations, and promotional activities, and build a full-link graph structure, where the nodes in the full-link graph structure include suppliers, warehouses, stores, and transportation hubs, and the edges include logistics paths, order flows, and inventory dependencies;

[0012] S12. Based on the full-link graph structure, the graph convolutional network is applied to each layer of the graph. By propagating the features of adjacent nodes, the spatiotemporal dependencies between nodes are learned. The graph convolutional network is used to learn the embedded representation of each node and convert the relationship between nodes into a low-dimensional vector representation.

[0013] S13. Model the historical data of the node in combination with the time series model, use the node embedding generated by the graph convolutional network as the input feature of the time series model, combine the historical sales data, market trends, seasonal fluctuations, and time series data of promotional activities with the graph structure features output by the graph convolutional network, and perform training to obtain a trained multimodal prediction model;

[0014] S14. Utilize the trained multimodal prediction model, combined with current sales data, market trends, seasonal fluctuations, and promotional activity data, to predict the sales demand of each node in the future period.

[0015] Preferably, the acquisition of real-time data of each link in the full link, and determination of the initial safety stock, replenishment cycle and order quantity based on the full link graph structure and combined with sales demand data comprises the following steps:

[0016] S21. Obtain real-time data on sales, inventory, supply and transportation status of each link in the whole chain, and dynamically update the whole chain graph structure;

[0017] S22. Embed the updated full-link graph based on the graph convolutional network, learn the implicit dependencies between nodes, and generate the initial safety stock, replenishment cycle and order quantity based on sales demand data.

[0018] Preferably, embedding the updated full-link graph based on the graph convolutional network, learning the implicit dependencies between nodes, and generating the initial safety stock, replenishment cycle and order quantity in combination with sales demand data includes the following steps:

[0019] S221, the graph convolutional network combines the features of each node with the features of its neighboring nodes through a convolution operation to obtain an embedded representation of the node, wherein the embedded node representation includes the local features of the node and its relationship with its neighboring nodes;

[0020] S222. Graph convolutional networks use adjacency relationships to propagate dependencies between nodes. Through graph convolution operations, each node integrates the information of its neighboring nodes to capture the complex correlations between nodes in the supply chain.

[0021] S223, the graph convolutional network learns and quantifies the implicit dependencies between nodes through the training process, and generates an embedded representation containing the dependencies between the nodes;

[0022] S224. Combining sales demand forecast, inventory status and dependencies between supply chain nodes, the graph convolutional network embedding representation is used to generate the initial safety stock, replenishment cycle and order quantity of each node.

[0023] Preferably, the method of combining sales demand forecast, inventory status and dependencies between supply chain nodes and using graph convolutional network embedding representation to generate initial safety inventory, replenishment cycle and order quantity of each node includes:

[0024] Combining sales demand forecast, inventory status and dependencies between supply chain nodes, the initial safety stock of each node is generated using graph convolutional network embedding representation.

[0025] Based on the node embedding representation generated by the graph convolutional network, combined with sales demand fluctuations, inventory consumption speed and transportation time efficiency factors, the replenishment cycle of each node is calculated;

[0026] Calculate the order quantity for each node based on the initial safety stock, replenishment cycle and predicted sales demand of each node.

[0027] As a preferred method, the calculation formula for the replenishment cycle of each node is:

[0028]

[0029] Where, T r (i) represents the replenishment cycle of node i, I avg (i) represents the average inventory of node i, D avg (i) represents the average daily demand of node i, L lead (i) represents the replenishment lead time of node i, β represents the constant factor used to adjust the impact of the embedded representation on the replenishment cycle and order quantity, and E i Represents the embedding representation of node i generated by the graph convolutional network.

[0030] As a preferred method, the calculation formula for the order quantity of each node is:

[0031] Qo(i)=(Davg(i)×Ti(i))×(1+β·Ei)+Si-Ic(i);

[0032] In the formula, Q o (i) represents the order quantity of node i, S i represents the safety stock of node i, I c (i) represents the current inventory of node i.

[0033] Preferably, the method of iteratively solving the optimal safety stock, replenishment cycle and order quantity based on the inventory optimization algorithm, and performing the inventory optimization operation according to the optimization result includes the following steps:

[0034] S31. Based on the quantum optimization algorithm, combined with the initial safety stock, replenishment cycle and order quantity, iteratively solves the optimal safety stock strategy, the replenishment path based on graph topology and the dynamic order quantity;

[0035] S32. Execute inventory transfer, replenishment and unsalable product clearance operations between warehouses and stores based on the optimal safety stock strategy, the replenishment path based on graph topology and the dynamic order quantity.

[0036] Preferably, the method based on the quantum optimization algorithm, combining the initial safety stock, replenishment cycle and order quantity, iteratively solving the optimal safety stock strategy, the replenishment path based on the graph topology and the dynamic order quantity includes the following steps:

[0037] S311. Taking minimizing the total cost of inventory management as the optimization goal, the safety stock, replenishment cycle and order quantity of each node are decision variables, and the constraints of inventory level, replenishment cycle, transportation time and budget limit are determined;

[0038] S312. Convert the objective function of inventory management into a quantum optimization problem, define a quantum Hamiltonian to represent the objective function, and map each decision variable to a quantum bit;

[0039] S313, setting the quantum bit to a uniform superposition state, randomly initializing the state of the quantum bit, and constructing a quantum circuit through quantum gate operations;

[0040] S314, converting the inventory holding cost, out-of-stock cost and transportation cost of the inventory management objective function into sub-items of the Hamiltonian to construct the overall quantum Hamiltonian;

[0041] S315. Based on the quantum approximate optimization algorithm, the Hamiltonian is optimized through quantum circuits, and the optimal solution is iteratively solved to obtain the optimal safety stock for each node;

[0042] S315. Utilize the shortest path optimization algorithm and the full-link graph structure to optimize the replenishment path between nodes and obtain a replenishment path based on the graph topology;

[0043] S316. Calculate the dynamic order quantity of each node based on the optimal safety stock, replenishment cycle and sales demand of each node.

[0044] Preferably, the expression of the quantum Hamiltonian is:

[0045] Hcost=∑i(Cstock(i)+Cstockout(i)+Ctransport(i));

[0046] In the formula, H cost represents the quantum Hamiltonian, C stock (i) represents the inventory holding cost of node i, C stockout (i) represents the out-of-stock cost of node i, C transport (i) represents the transportation cost of node i.

[0047] Compared with the prior art, the present invention provides a retail inventory optimization method based on full-link management, which has the following beneficial effects:

[0048] (1) The present invention innovatively combines graph convolutional networks and quantum optimization algorithms to achieve dynamic inventory optimization based on full-link management, which not only improves the intelligence level of inventory management, but also shows significant advantages in accuracy, response speed and cost control, providing retail enterprises with a new and efficient inventory optimization solution.

[0049] (2) The present invention can effectively capture the spatiotemporal dependencies between nodes in the supply chain through the embedded representation of the graph convolutional network, thereby calculating more accurate safety stock, replenishment cycle and order quantity for each node. In addition, combined with the quantum optimization algorithm, it can efficiently solve the optimal solution in inventory management, reduce manual intervention, and improve the flexibility and automation of the inventory management system.

[0050] (3) Through the construction of the full-link graph structure and dynamic data update, the present invention can obtain the sales, inventory, transportation and other status information of each link in real time, so as to dynamically adjust the inventory strategy to ensure that each node in the supply chain can maintain an appropriate inventory level and avoid inventory shortages or excesses; in addition, the shortest path optimization and replenishment path optimization based on the quantum optimization algorithm make the replenishment process more efficient and reduce unnecessary transportation costs and time delays. At the same time, accurate safety inventory calculation and dynamically adjusted order quantity effectively reduce the risk of out-of-stock and ensure the continuity of commodity supply.

[0051] (4) By combining graph convolutional networks with time series models, the present invention can flexibly adjust inventory strategies based on market trends, seasonal fluctuations, and the impact of promotional activities to adapt to complex market environment changes. In addition, the solution method based on the quantum optimization algorithm enables the technical solution to maintain high performance in a large-scale supply chain environment and has strong scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0053] Figure 1 4 is a flowchart of a retail inventory optimization method based on full-link management according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] According to an embodiment of the present invention, a retail inventory optimization method based on full-link management is provided.

[0056] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a retail inventory optimization method based on full-link management is provided, comprising the following steps:

[0057] S1. Collect historical data, build a full-link graph structure, and predict the sales demand of each node in the future period based on the graph convolutional network and time series model;

[0058] The collecting of historical data, building a full-link graph structure, and predicting the sales demand of each node in the future period based on the graph convolutional network and the time series model include the following steps:

[0059] S11. Collect data on historical sales, market trends, seasonal fluctuations, and promotional activities, and build a full-link graph structure, where the nodes in the full-link graph structure include suppliers, warehouses, stores, and transportation hubs, and the edges include logistics paths, order flows, and inventory dependencies;

[0060] Specifically, 1) data collection and collation include:

[0061] Historical sales data: collect sales data over the past period of time, including sales volume, inventory changes, etc. of each store, warehouse, and supplier. These data can be obtained through channels such as POS systems, warehouse management systems, and supply chain management systems;

[0062] Market trends: Collect macro market data, analyze seasonal changes, industry trends, consumer behavior and other factors, and understand the changing patterns of market demand;

[0063] Seasonal fluctuations: Obtain the impact of different seasons, holidays and other factors on sales. Seasonal data can be analyzed through historical sales data, weather data, holiday promotions, etc.

[0064] Promotional activity data: Collect relevant data on promotional activities, understand the role of promotional activities in driving sales, and extract information such as the time, frequency, and type of promotional activities;

[0065] 2) Building a full-link graph structure includes:

[0066] Define nodes and edges:

[0067] Node: A key entity in the supply chain, such as suppliers, warehouses, stores, transportation hubs, etc. Each node represents an independent supply chain link;

[0068] Edge: the relationship between nodes, such as logistics path (delivery route from warehouse to store), order flow (order path from supplier to warehouse), inventory dependency (inventory transfer between store and warehouse), etc.

[0069] Dynamic characteristics of graph structure: In addition to static information (such as fixed relationships between nodes), the graph structure should be dynamic and able to be updated with factors such as transportation delays, inventory fluctuations, and order changes. The attributes of each node and edge (such as inventory, transportation time, demand, etc.) will be adjusted with changes in actual operations;

[0070] S12. Based on the full-link graph structure, the graph convolutional network is applied to each layer of the graph. By propagating the features of adjacent nodes, the spatiotemporal dependencies between nodes are learned. The graph convolutional network is used to learn the embedded representation of each node and convert the relationship between nodes into a low-dimensional vector representation.

[0071] S13. Model the historical data of the node in combination with the time series model, use the node embedding generated by the graph convolutional network as the input feature of the time series model, combine the historical sales data, market trends, seasonal fluctuations, and time series data of promotional activities with the graph structure features output by the graph convolutional network, and perform training to obtain a trained multimodal prediction model; specifically including:

[0072] 1) Graph Convolutional Network (GCN) Applications:

[0073] GCN builds graph convolutional layers: Based on the full-link graph structure, GCN is applied to each layer of the graph. GCN learns the spatiotemporal dependencies between nodes by propagating the features of adjacent nodes. For example, the sales demand of a store is not only affected by its own historical sales, but may also be affected by the inventory status and transportation delays of neighboring stores or warehouses. Through GCN, the model can capture the complex dependencies between nodes, thereby improving the prediction of sales demand;

[0074] Node embedding: GCN learns the embedding representation of each node through graph convolution operations, converting the relationship between nodes into low-dimensional vector representations. These vectors can represent the node's historical demand pattern, inventory demand, and dependencies with other nodes;

[0075] 2) Time series models (such as ARIMA, LSTM):

[0076] Fusion time series forecasting: In order to handle the dynamic characteristics of nodes changing over time, the historical data of the nodes are modeled by combining time series models (such as LSTM, ARIMA, etc.). The time series model predicts the demand changes of each node in the future period by capturing time series patterns such as sales trends, seasonal fluctuations and promotion effects;

[0077] Fusion mechanism: The node embedding generated by GCN is used as the input feature of the time series model. The time series data such as historical sales data, seasonal fluctuations, and promotional activities are combined with the graph structure features output by GCN. The model’s learning of the spatiotemporal dependencies between nodes is strengthened through fusion mechanisms (such as weighted averaging and splicing).

[0078] 3) Model training

[0079] In the training phase, historical sales data, market trends, seasonal fluctuations and other input data are used, and the graph convolutional network (GCN) is used to learn the dependencies and spatiotemporal patterns between nodes. The time series model is responsible for learning the temporal regularity of the data and capturing the changing trend of sales demand over time;

[0080] Loss function design: Design a suitable loss function to measure the accuracy of the prediction results, combine node embedding and time series data, and optimize the overall model;

[0081] S14. Using the trained multimodal prediction model, combined with current sales data, market trends, seasonal fluctuations, and promotion activity data, predict the sales demand of each node in the future period;

[0082] S2. Obtain real-time data from each link in the entire chain, and determine the initial safety stock, replenishment cycle, and order quantity based on the full-chain graph structure and combined with sales demand data;

[0083] The acquisition of real-time data of each link in the full chain, based on the full chain graph structure, and combined with sales demand data to determine the initial safety stock, replenishment cycle and order quantity includes the following steps:

[0084] S21. Obtain real-time data on sales, inventory, supply and transportation status of each link in the whole chain, and dynamically update the whole chain graph structure;

[0085] The graph structure is dynamically updated as follows:

[0086] As sales, inventory, transportation and other statuses change, the supply chain graph structure will be dynamically updated. For example, when the inventory level of a node (such as a warehouse) changes, the corresponding graph edge (such as the inventory transfer path from the warehouse to the store) will adjust its weight according to the inventory situation;

[0087] At the same time, if there are events such as transportation delays and supply chain disruptions, the graph structure and edge weights will be dynamically adjusted based on these abnormal events to ensure that the graph structure always reflects the latest status of the supply chain;

[0088] S22. Embed the updated full-link graph based on the graph convolutional network, learn the implicit dependencies between nodes, and generate the initial safety stock, replenishment cycle and order quantity based on sales demand data;

[0089] Specifically, embedding the updated full-link graph based on the graph convolutional network, learning the implicit dependencies between nodes, and generating the initial safety stock, replenishment cycle and order quantity in combination with sales demand data includes the following steps:

[0090] S221. The graph convolutional network combines the features of each node with the features of its neighboring nodes through a convolution operation to obtain an embedded representation of the node, wherein the embedded node representation includes the local features of the node and its relationship with its neighboring nodes; specifically, it includes:

[0091] Use graph convolutional networks (GCNs) to embed the entire supply chain graph. GCNs combine the features of each node (such as inventory, sales, shipping delays, etc.) with the features of its neighboring nodes through convolution operations to obtain an embedded representation of the node. The embedded node representation includes the local features of the node (such as inventory levels) and its relationship with neighboring nodes (such as suppliers, warehouses, stores, etc.);

[0092] S222. The graph convolutional network uses the adjacency relationship to propagate the dependency relationship between nodes. Through the graph convolution operation, each node integrates the information of its adjacent nodes to capture the complex correlation between nodes in the supply chain; specifically, it includes:

[0093] GCN uses adjacency relationships to propagate dependencies between nodes. Through graph convolution operations, each node can integrate the information of its neighboring nodes, thereby better capturing the complex correlations between nodes in the supply chain (for example, demand transmission across regional stores, inventory dependencies between warehouses and stores, etc.); after being processed by GCN, the embedding vector of each node can represent the state of the node in the entire supply chain, including its responsiveness to sales demand, flexibility in inventory adjustment, etc. In this way, GCN can effectively learn the implicit dependencies between nodes;

[0094] S223. The graph convolutional network learns and quantifies the implicit dependencies between nodes through the training process, and generates an embedded representation containing the dependencies between the nodes; specifically, it includes:

[0095] 1) Dependency modeling: GCN can capture complex dependencies between nodes through graph convolution operations. These implicit dependencies include:

[0096] Cross-node demand transmission: For example, demand changes in a store may be affected by factors such as inventory levels and replenishment cycles in other stores or regional warehouses;

[0097] The mutual influence of inventory and supply: The inventory changes of a warehouse are not only determined by the sales of the warehouse, but also affected by factors such as the supplier's delivery cycle and transportation delays;

[0098] Impact of logistics constraints: Delays in transportation routes, capacity constraints, and other factors can affect inventory flow and replenishment efficiency throughout the supply chain;

[0099] 2) Dependency learning: GCN learns and quantifies these dependencies through the training process, generating an embedded representation containing the dependencies between nodes to support subsequent inventory management and replenishment decisions;

[0100] S224. Combining sales demand forecast, inventory status and dependencies between supply chain nodes, the graph convolutional network embedding representation is used to generate the initial safety stock, replenishment cycle and order quantity of each node;

[0101] Specifically, the method combines sales demand forecast, inventory status and dependencies between supply chain nodes, and uses graph convolutional network embedding representation to generate the initial safety stock, replenishment cycle and order quantity of each node, including:

[0102] Combining sales demand forecast, inventory status and dependencies between supply chain nodes, the initial safety stock of each node is generated using graph convolutional network embedding representation.

[0103] Based on the node embedding representation generated by the graph convolutional network, combined with sales demand fluctuations, inventory consumption speed and transportation time efficiency factors, the replenishment cycle of each node is calculated;

[0104] The calculation formula for the replenishment cycle of each node is:

[0105]

[0106] Where, T r (i) represents the replenishment cycle of node i, I avg (i) represents the average inventory of node i, D avg (i) represents the average daily demand of node i, L lead (i) represents the replenishment lead time of node i, β represents the constant factor used to adjust the impact of the embedded representation on the replenishment cycle and order quantity, and E i Represents the embedded representation of node i generated by the graph convolutional network;

[0107] Calculate the order quantity for each node based on the initial safety stock, replenishment cycle, and predicted sales demand of each node;

[0108] The calculation formula for the order quantity of each node is:

[0109] Qo(i)=(Davg(i)×Ti(i))×(1+β·Ei)+Si-Ic(i);

[0110] In the formula, Q o (i) represents the order quantity of node i, S i represents the safety stock of node i, I c (i) represents the current inventory of node i;

[0111] S3. Based on the inventory optimization algorithm, iteratively solve the optimal safety stock, replenishment cycle and order quantity, and perform inventory optimization operations according to the optimization results;

[0112] The method of iteratively solving the optimal safety stock, replenishment cycle and order quantity based on the inventory optimization algorithm and performing the inventory optimization operation according to the optimization result includes the following steps:

[0113] S31. Based on the quantum optimization algorithm, combined with the initial safety stock, replenishment cycle and order quantity, iteratively solves the optimal safety stock strategy, the replenishment path based on graph topology and the dynamic order quantity;

[0114] Specifically, the method based on the quantum optimization algorithm, combining the initial safety stock, replenishment cycle and order quantity, iteratively solving the optimal safety stock strategy, the replenishment path based on the graph topology and the dynamic order quantity includes the following steps:

[0115] S311. Taking minimizing the total cost of inventory management as the optimization goal, the safety stock, replenishment cycle and order quantity of each node are decision variables, and the constraints of inventory level, replenishment cycle, transportation time and budget limit are determined;

[0116] S312. Convert the objective function of inventory management into a quantum optimization problem, define a quantum Hamiltonian to represent the objective function, and map each decision variable to a quantum bit (map the decision variables such as safety inventory, replenishment cycle, and order quantity of each node to quantum bits. Each decision variable will occupy one or more quantum bits, and the state of these quantum bits determines the current inventory configuration);

[0117] S313, setting the quantum bit to a uniform superposition state, randomly initializing the state of the quantum bit, and constructing a quantum circuit through quantum gate operations;

[0118] The initial state of a quantum circuit is usually set to a uniform superposition state to cover all possible inventory configurations. Through quantum gate operations, the state of the quantum bit is initialized to this superposition state;

[0119] S314, converting the inventory holding cost, out-of-stock cost and transportation cost of the inventory management objective function into sub-items of the Hamiltonian to construct the overall quantum Hamiltonian;

[0120] The expression of the quantum Hamiltonian is:

[0121] Hcost=∑i(Cstock(i)+Cstockout(i)+Ctransport(i));

[0122] In the formula, H cost represents the quantum Hamiltonian, C stock (i) represents the inventory holding cost of node i, C stockout (i) represents the out-of-stock cost of node i, C transport (i) represents the transportation cost of node i;

[0123] S315. Based on the quantum approximate optimization algorithm, the Hamiltonian is optimized through quantum circuits, and the optimal solution is iteratively solved to obtain the optimal safety stock for each node; specifically, it includes:

[0124] 1) Application of Quantum Approximate Optimization Algorithm (QAOA)

[0125] The quantum approximate optimization algorithm optimizes the state of the quantum bit through a series of parameterized quantum gates, thereby minimizing the Hamiltonian. The core idea of ​​QAOA is to gradually adjust the state of the quantum bit through multiple iterations of the quantum circuit so that it tends to the lowest energy state, that is, to optimize the optimal solution of the objective function;

[0126] Steps of QAOA:

[0127] Initialization: The initial state of a quantum circuit is usually a uniform superposition state;

[0128] Apply Hamiltonian: adjust the energy according to the Hamiltonian of the target function through quantum gate operation;

[0129] Parameter optimization: Through parameterized quantum gates (such as rotating gate RZ), the state of the quantum bit is adjusted so that the expected value of the Hamiltonian gradually decreases;

[0130] Measurement: After each iteration, the quantum bits are measured to obtain a solution. This solution is the current inventory strategy (safety stock, replenishment cycle, and order quantity for each node);

[0131] 2) Iterative Optimization Process

[0132] QAOA is an iterative process, and each iteration reduces the expected value of the Hamiltonian by adjusting the parameters in the quantum circuit. After each iteration, the quantum bits are measured to obtain the current solution and calculate the value of the objective function. After multiple iterations of optimization, the final solution is the optimal solution;

[0133] Iterations: After each iteration of optimization, the state of the quantum bits is measured to obtain an inventory configuration plan;

[0134] Optimal solution: Through the optimization process, it gradually approaches the optimal safety stock configuration. In the end, the solution given by QAOA corresponds to the optimal safety stock amount for each node;

[0135] 3) Output the optimal safety stock

[0136] The output of the quantum optimization algorithm is the state of the qubit, which, after multiple measurements, can determine the optimal safety stock for each node. The measurement results of the qubit will correspond to the safety stock configuration of each node;

[0137] S315. Using the shortest path optimization algorithm and combining it with the full-link graph structure, optimize the replenishment path between nodes to obtain a replenishment path based on the graph topology; specifically including:

[0138] 1) Apply the shortest path optimization algorithm (Dijkstra algorithm)

[0139] Suppose that Dijkstra algorithm is used to optimize the replenishment path. Dijkstra algorithm is a classic algorithm for dealing with the single-source shortest path problem in weighted graphs, and is applicable to all graphs with non-negative edge weights.

[0140] Dijkstra algorithm steps:

[0141] Initialization: Set a source node (such as a warehouse or supplier) and assign an initial distance to each node. The initial distance of the source node is 0, and the initial distances of other nodes are infinite;

[0142] Select the node with the smallest distance among the unprocessed nodes: From the unprocessed nodes, select a node with the smallest distance as the current node;

[0143] Update the distance of adjacent nodes: For each adjacent node of the current node, check whether a shorter path can be obtained through the current node. If it can, update the distance of the adjacent node;

[0144] Mark the current node as processed: Mark the current node as processed and no longer update the path

[0145] Repeat the above steps until all nodes are processed and finally the shortest path from the source node to all other nodes is obtained;

[0146] 2) Optimize replenishment path

[0147] Use the Dijkstra algorithm to calculate the shortest path from the source node (warehouse or supplier) to all target nodes (stores, etc.) to obtain the replenishment path based on the graph topology;

[0148] Path selection: Select the optimal path from the warehouse or supplier to the store based on the shortest path algorithm;

[0149] Minimize transportation costs: Ensure the efficiency of the replenishment process by optimizing the route to minimize the total transportation cost or time;

[0150] 3) Dynamically adjust the replenishment path

[0151] Real-time data feedback: Based on real-time transportation data (such as delays, traffic, etc.), the replenishment path can be adjusted dynamically. By continuously updating the edge weights in the graph structure (for example, transportation delays, cost changes, etc.), the shortest path is recalculated;

[0152] Multi-path optimization: If multiple paths can reach the same target node, the optimal solution of multiple candidate paths can be calculated and the path that maximizes cost-effectiveness can be selected;

[0153] 4) Output replenishment path

[0154] Finally, the replenishment path obtained based on the shortest path algorithm will be used as the optimization result. The best path for each replenishment link is output, indicating the shortest transportation path between the warehouse and the store;

[0155] S316. Calculate the dynamic order quantity of each node according to the optimal safety stock quantity, replenishment cycle and sales demand of each node;

[0156] S32. Execute inventory transfer, replenishment and unsalable product clearance operations between warehouses and stores based on the optimal safety stock strategy, the replenishment path based on graph topology and the dynamic order quantity.

[0157] To sum up, with the help of the above-mentioned technical scheme of the present invention, the present invention realizes dynamic inventory optimization based on full-link management by innovatively combining graph convolutional networks and quantum optimization algorithms, which not only improves the intelligence level of inventory management, but also shows significant advantages in accuracy, response speed and cost control, providing retail enterprises with a new and efficient inventory optimization solution.

[0158] At the same time, the present invention can effectively capture the spatiotemporal dependencies between nodes in the supply chain through the embedded representation of the graph convolutional network, thereby calculating more accurate safety stock, replenishment cycle and order quantity for each node. In addition, combined with the quantum optimization algorithm, it can efficiently solve the optimal solution in inventory management, reduce manual intervention, and improve the flexibility and automation of the inventory management system.

[0159] At the same time, the present invention can obtain the sales, inventory, transportation and other status information of each link in real time through the construction of the full-link graph structure and dynamic data update, so as to dynamically adjust the inventory strategy to ensure that each node in the supply chain can maintain an appropriate inventory level and avoid inventory shortages or excesses; in addition, the shortest path optimization and replenishment path optimization based on the quantum optimization algorithm make the replenishment process more efficient and reduce unnecessary transportation costs and time delays. At the same time, accurate safety inventory calculation and dynamically adjusted order quantity effectively reduce the risk of out-of-stock and ensure the continuity of commodity supply.

[0160] At the same time, by combining graph convolutional networks with time series models, the present invention can flexibly adjust inventory strategies based on market trends, seasonal fluctuations, and the impact of promotional activities to adapt to complex market changes. In addition, the solution method based on the quantum optimization algorithm enables the technical solution to maintain high performance in a large-scale supply chain environment and has strong scalability.

[0161] The technical features of the above-mentioned embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. A person of ordinary skill in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes the steps described in the above method, and the storage medium, such as ROM / RAM, a disk, an optical disk, etc.

[0162] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A retail inventory optimization method based on full-link management, characterized in that: The following steps are involved: S1. Collect historical data, build a full-link graph structure, and predict the sales demand of each node in the future period based on the graph convolutional network and time series model; S2. Obtain real-time data from each link in the entire chain, and determine the initial safety stock, replenishment cycle, and order quantity based on the full-chain graph structure and combined with sales demand data; S3. Based on the inventory optimization algorithm, iteratively solve the optimal safety stock, replenishment cycle and order quantity, and perform inventory optimization operations based on the optimization results.

2. A retail inventory optimization method based on full-link management according to claim 1, characterized in that: The collecting of historical data, building a full-link graph structure, and predicting the sales demand of each node in the future period based on the graph convolutional network and the time series model include the following steps: S11. Collect data on historical sales, market trends, seasonal fluctuations, and promotional activities, and build a full-link graph structure, where the nodes in the full-link graph structure include suppliers, warehouses, stores, and transportation hubs, and the edges include logistics paths, order flows, and inventory dependencies; S12. Based on the full-link graph structure, the graph convolutional network is applied to each layer of the graph. By propagating the features of adjacent nodes, the spatiotemporal dependencies between nodes are learned. The graph convolutional network is used to learn the embedded representation of each node and convert the relationship between nodes into a low-dimensional vector representation. S13. Model the historical data of the node in combination with the time series model, use the node embedding generated by the graph convolutional network as the input feature of the time series model, combine the historical sales data, market trends, seasonal fluctuations, and time series data of promotional activities with the graph structure features output by the graph convolutional network, and perform training to obtain a trained multimodal prediction model; S14. Utilize the trained multimodal prediction model, combined with current sales data, market trends, seasonal fluctuations, and promotional activity data, to predict the sales demand of each node in the future period.

3. The retail inventory optimization method based on full-link management according to claim 1 is characterized in that: The acquisition of real-time data of each link in the full chain, based on the full chain graph structure, and combined with sales demand data to determine the initial safety stock, replenishment cycle and order quantity includes the following steps: S21. Obtain real-time data on sales, inventory, supply and transportation status of each link in the whole chain, and dynamically update the whole chain graph structure; S22. Embed the updated full-link graph based on the graph convolutional network, learn the implicit dependencies between nodes, and generate the initial safety stock, replenishment cycle and order quantity based on sales demand data.

4. A retail inventory optimization method based on full-link management according to claim 3, characterized in that: The method of embedding the updated full-link graph based on the graph convolutional network, learning the implicit dependency relationship between nodes, and generating the initial safety stock, replenishment cycle and order quantity in combination with sales demand data includes the following steps: S221, the graph convolutional network combines the features of each node with the features of its neighboring nodes through a convolution operation to obtain an embedded representation of the node, wherein the embedded node representation includes the local features of the node and its relationship with its neighboring nodes; S222. Graph convolutional networks use adjacency relationships to propagate dependencies between nodes. Through graph convolution operations, each node integrates the information of its neighboring nodes to capture the complex correlations between nodes in the supply chain. S223, the graph convolutional network learns and quantifies the implicit dependencies between nodes through the training process, and generates an embedded representation containing the dependencies between the nodes; S224. Combining sales demand forecast, inventory status and dependencies between supply chain nodes, the graph convolutional network embedding representation is used to generate the initial safety stock, replenishment cycle and order quantity of each node.

5. A retail inventory optimization method based on full-link management according to claim 4, characterized in that: The method combines sales demand forecast, inventory status and dependencies between supply chain nodes and uses graph convolutional network embedding to represent the generation of initial safety inventory, replenishment cycle and order quantity for each node, including: Combining sales demand forecast, inventory status and dependencies between supply chain nodes, the initial safety stock of each node is generated using graph convolutional network embedding representation. Based on the node embedding representation generated by the graph convolutional network, combined with sales demand fluctuations, inventory consumption speed and transportation time factors, the replenishment cycle of each node is calculated; Calculate the order quantity for each node based on the initial safety stock, replenishment cycle and predicted sales demand of each node.

6. A retail inventory optimization method based on full-link management according to claim 5, characterized in that: The calculation formula for the replenishment cycle of each node is: Where, T r (i) represents the replenishment cycle of node i, I avg (i) represents the average inventory of node i, D avg (i) represents the average daily demand of node i, L lead (i) represents the replenishment lead time of node i, β represents the constant factor used to adjust the impact of the embedded representation on the replenishment cycle and order quantity, and E i Represents the embedding representation of node i generated by the graph convolutional network.

7. A retail inventory optimization method based on full-link management according to claim 6, characterized in that: The calculation formula for the order quantity of each node is: Qo(i)=(Davg(i)×Ti(i))×(1+β·Ei)+Si-Ic(i); In the formula, Q o (i) represents the order quantity of node i, S i represents the safety stock of node i, I c (i) represents the current inventory of node i.

8. The method for optimizing retail inventory based on full-link management according to claim 1, characterized in that: The method of iteratively solving the optimal safety stock, replenishment cycle and order quantity based on the inventory optimization algorithm and performing the inventory optimization operation according to the optimization result includes the following steps: S31. Based on the quantum optimization algorithm, combined with the initial safety stock, replenishment cycle and order quantity, iteratively solves the optimal safety stock strategy, the replenishment path based on graph topology and the dynamic order quantity; S32. Execute inventory transfer, replenishment and unsalable product clearance operations between warehouses and stores based on the optimal safety stock strategy, the replenishment path based on graph topology and the dynamic order quantity.

9. A retail inventory optimization method based on full-link management according to claim 8, characterized in that: The method based on the quantum optimization algorithm, combining the initial safety stock, replenishment cycle and order quantity, iteratively solving the optimal safety stock strategy, the replenishment path based on the graph topology and the dynamic order quantity includes the following steps: S311. Taking minimizing the total cost of inventory management as the optimization goal, the safety stock, replenishment cycle and order quantity of each node are decision variables, and the constraints of inventory level, replenishment cycle, transportation time and budget limit are determined; S312. Convert the objective function of inventory management into a quantum optimization problem, define a quantum Hamiltonian to represent the objective function, and map each decision variable to a quantum bit; S313, setting the quantum bit to a uniform superposition state, randomly initializing the state of the quantum bit, and constructing a quantum circuit through quantum gate operations; S314, converting the inventory holding cost, out-of-stock cost and transportation cost of the inventory management objective function into sub-items of the Hamiltonian to construct the overall quantum Hamiltonian; S315. Based on the quantum approximate optimization algorithm, the Hamiltonian is optimized through quantum circuits, and the optimal solution is iteratively solved to obtain the optimal safety stock for each node; S315. Utilize the shortest path optimization algorithm and the full-link graph structure to optimize the replenishment path between nodes and obtain a replenishment path based on the graph topology; S316. Calculate the dynamic order quantity of each node based on the optimal safety stock, replenishment cycle and sales demand of each node.

10. A retail inventory optimization method based on full-link management according to claim 9, characterized in that: The expression of the quantum Hamiltonian is: Hcost=∑i(Cstock(i)+Cstockout(i)+Ctransport(i)); In the formula, H cost represents the quantum Hamiltonian, C stock (i) represents the inventory holding cost of node i, C stockout (i) represents the out-of-stock cost of node i, C transport (i) represents the transportation cost of node i.

Images