Intelligent supply chain inventory optimization method and system

Abstract

The application discloses an intelligent supply chain inventory optimization method and system, which comprises the following steps: updating the actual inventory quantity of a current node according to the in-transit quantity from an upstream node to a downstream node; proportionally distributing the actual demand quantity of the downstream node received by the current node according to the updated actual inventory quantity of the current node; predicting the actual demand quantity of the downstream node received by the current node and the supply quantity of the upstream node to the current node by using a long short-term memory neural network; judging the supply-demand change phase based on the prediction result, adopting a supply-demand acceleration situation unified mapping strategy and upstream supply constraints, improving multi-agent, adding an environment perception layer to generate a dynamic safety inventory strategy, and executing a final order decision. Through upstream and downstream prediction driving, the application generates a supply-demand acceleration situation unified mapping strategy, improves multi-agent, adds an environment perception layer, dynamically generates a safety inventory strategy to match demand fluctuation and supply capacity change, and overcomes the shortcomings of a traditional safety inventory strategy.

Description

Technical Field

[0001] This invention relates to the field of inventory decision-making in supply chain management, and specifically to an intelligent supply chain inventory optimization method and system. Background Technology

[0002] With the continuous growth of e-commerce retail, FMCG distribution, and pharmaceutical cold chain industries, the supply chain system faces increasingly complex supply-demand imbalances and inventory management challenges when encountering holidays, promotional activities, and regional demand fluctuations. Stable, efficient, and flexible inventory control capabilities have become crucial for ensuring service levels and cost competitiveness. However, current supply chain inventory management commonly employs fixed safety stock coefficients or single-agent decision-making methods, which struggle to flexibly address demand fluctuations at different nodes and upstream supply capacity fluctuations, frequently leading to stockouts, supply disruptions, or inventory backlogs in practice. Furthermore, existing forecasting and ordering mechanisms largely rely on static models, lacking measures to detect supply-demand mismatches, making it difficult to respond promptly to forecast errors and extreme fluctuations, impacting overall supply chain operational efficiency, resource utilization efficiency, and service levels. Therefore, there is an urgent need for an intelligent inventory optimization method that combines upstream and downstream forecasting, multi-agent dynamic safety stock strategy generation, and the perception of accelerated supply-demand changes to enhance supply chain collaboration, flexibility, and resilience, meeting the practical needs of continuous and stable operation in highly volatile industries.

[0003] To address the aforementioned issues, this invention utilizes upstream and downstream forecasting and multi-agent technology to dynamically generate safety stock strategies for each node to match changes in demand and supply capacity. For enterprises with multiple inventory points, a unified mapping strategy for accelerating supply and demand is implemented. By leveraging the dynamic interaction of agents to perceive changes in supply and demand rhythms and ordering feasibility constraints in real time, flexible resource allocation among different nodes is achieved, ensuring service rates and reducing stockout risks. This enables accurate supply chain forecasting, dynamic ordering, and flexible replenishment closed-loop optimization in a real-world volatile environment, effectively improving supply chain collaboration efficiency. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent supply chain inventory optimization method and system. The aim is to construct an intelligent inventory optimization system with a unified mapping strategy for accelerating supply and demand, in order to improve the inventory imbalance and sluggish response problems caused by insufficient flexibility in existing supply chain inventory management.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent supply chain inventory optimization method and system, the method comprising the following steps:

[0006] Step S1: Collect the actual inventory and order lead time of the current node within the enterprise inventory system information period, the actual demand received by the current node from the downstream node, the stock shortage owed by the current node to the downstream node, the amount in transit from the upstream node to the current node, and the amount supplied by the upstream node to the current node.

[0007] Step S2: Update the actual inventory of the current node based on the amount in transit from the upstream node to the current node;

[0008] Step S3: Based on the updated actual inventory of the current node, allocate the inventory proportionally according to the actual demand received by the current node from the downstream nodes.

[0009] Step S4: Use a long short-term memory neural network to predict the actual demand received by the current node from downstream nodes and the supply from upstream nodes to the current node.

[0010] Step S5: Based on the forecast results, determine the phase of supply and demand changes, and adopt a unified mapping strategy for accelerated supply and demand trends and upstream supply constraints.

[0011] Step S6: Combining the unified mapping strategy of supply and demand acceleration and upstream supply constraints, improve the multi-agent system, add an environmental perception layer to generate a dynamic safety stock strategy, and execute the final ordering decision.

[0012] Furthermore, in step S1, the actual inventory level and order lead time of the current node within the enterprise inventory system information period are collected, along with the actual demand received by the current node from downstream nodes, the stockout owed by the current node to downstream nodes, the amount in transit from upstream nodes to the current node, and the amount supplied by upstream nodes to the current node; the formulas are as follows:

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] ;

[0018] ;

[0019] Where j represents the current node, i represents the upstream node, and k represents the downstream node. This indicates the actual demand received by current node j from downstream node k in the t-th period; J represents the set of all current nodes, corresponding to the existence of multiple inventory points for the enterprise; This represents the set of downstream nodes served by the current node j; Indicates the total collection period; This represents the actual inventory level of node j in the t-th period; This represents the amount of stock shortage that current node j owes to downstream node k in the t-th period; This represents the amount of time that is in transit from upstream node i to current node j in the t-th period; Let represent the amount of goods supplied by upstream node i to current node j in the t-th period; I represents the set of upstream nodes that can supply current node j. This represents the order lead time of the current node j in the t-th period.

[0020] Furthermore, in step S2, the actual inventory of the current node is updated based on the amount in transit from the upstream node to the current node; the specific steps are as follows:

[0021] Step S21, receive incoming goods in transit. The lead time from upstream node i to current node j is... In the t-th period, the current node j receives the data from the node in the t-th period. The amount of goods sent by upstream node i in each cycle;

[0022] Step S22, update the in-transit inventory. After the current node j receives the goods, the in-transit inventory from the upstream node i to the current node j decreases; the formula is as follows:

[0023] ;

[0024] in, This represents the amount of time that is in transit from upstream node i to current node j during the (t-1)th period; Indicates the first The order quantity from the current node j to the upstream node i in each cycle;

[0025] Step S23: Update the actual inventory. After the current node j receives the goods, the actual inventory increases; the formula is as follows:

[0026] ;

[0027] in, This represents the actual inventory of node j in the (t-1)th period.

[0028] Furthermore, in step S3, the inventory of the current node is allocated proportionally according to the actual demand received by the current node from downstream nodes, based on the updated actual inventory level of the current node; the specific steps are as follows:

[0029] Step S31, Receive demand: In the t-th period, the current node j receives the actual demand from the downstream node k. ;

[0030] Step S32: Determine whether the updated actual inventory of the current node meets the actual demand received by the current node from the downstream node; the specific steps are as follows:

[0031] when When the current node's actual inventory meets the actual demand received by the current node from downstream nodes, update the current node's shipments to downstream nodes, the current node's actual inventory, and the current node's stockout to downstream nodes; the formula is as follows:

[0032] ;

[0033] ;

[0034] ;

[0035] in, This represents the amount of goods shipped from the current node j to the downstream node k during the t-th period.

[0036] when When this occurs, it indicates that the current node's actual inventory cannot meet the actual demand received by the current node from downstream nodes. For downstream nodes, the inventory is allocated proportionally based on the actual demand received by the current node from downstream nodes; the formula is as follows:

[0037] ;

[0038] in, This represents the amount of stock shortage that current node j owes to downstream node k in the (t-1)th period;

[0039] Update the current node's outstanding inventory to downstream nodes using the following formula:

[0040] ;

[0041] Update the current node's actual inventory to zero using the following formula:

[0042] .

[0043] Furthermore, in step S4, a long short-term memory neural network is used to predict the actual demand received by the current node from downstream nodes and the supply volume from upstream nodes to the current node; the specific steps are as follows:

[0044] Step S41: Obtain the actual demand input sequence and the supply quantity input sequence based on a sliding window. The sliding window length is w periods. Extract the actual demand quantity received by the current node j from the downstream node over the past w periods to obtain the actual demand input sequence, as shown in the following formula:

[0045] ;

[0046] in, Indicates the first During each cycle, the current node j receives the actual demand from the downstream node k. Indicates the first During each cycle, the current node j receives the actual demand from the downstream node k, where w represents the sliding window length.

[0047] The supply quantity input sequence is obtained by extracting the supply quantity from upstream node i to current node j over the past w periods, as shown in the following formula:

[0048] ;

[0049] in, They represent in The amount of goods supplied from upstream node i to current node j in each cycle, The amount of goods supplied from upstream node i to current node j in each cycle;

[0050] Step S42: Based on the obtained actual demand input sequence, use a long short-term memory neural network to predict the predicted demand received by the current node j from the downstream node k in the (t+1)th period, as shown in the following formula:

[0051] ;

[0052] in, This indicates the predicted demand received by the current node j from the downstream node k in the (t+1)th period. This represents a long short-term memory neural network;

[0053] Step S43: Based on the obtained supply quantity input sequence, use a long short-term memory neural network to predict the predicted supply quantity from upstream node i to current node j in the (t+1)th period; the formula is as follows:

[0054] ;

[0055] in, This represents the predicted supply volume from upstream node i to current node j in the (t+1)th period.

[0056] Furthermore, in step S5, the phase of supply and demand changes is determined based on the forecast results, and a unified mapping strategy for accelerated supply and demand trends and upstream supply constraints are adopted; the specific steps are as follows:

[0057] Step S51: Based on the changes in demand and supply within adjacent periods, calculate the phase of the demand change rhythm and the phase of the supply change rhythm, and obtain the phase difference between the supply and demand change rhythms, as shown in the following formula:

[0058] ;

[0059] in, This represents the phase difference in the supply and demand rhythm at the current node at the nth cycle, used to characterize the relationship between the rate of change of demand and supply over time; when When the rate of change in demand exceeds the rate of change in supply, there is a potential risk of shortages, indicating a strong demand phase; when When the rate of increase in demand matches the rate of increase in supply, supply is in a stable state, which is the equilibrium phase; when... When the rate of change in demand is less than the rate of change in supply, a supply redundancy occurs, which is a strong supply phase. Represents the arctangent function. This indicates an extremely small positive number that avoids a denominator of zero, and its value is 0.00001;

[0060] The above three strategies can be represented using a unified mapping strategy of accelerating supply and demand, as shown in the following formula:

[0061] ;

[0062] in, This represents the unified mapping strategy for the accelerating supply and demand situation at the current node j in the t-th period;

[0063] Step S52: Based on the actual demand received by the current node j from the downstream node k within the collected enterprise inventory system information period, calculate the historical average demand and historical demand standard deviation; the formula is as follows:

[0064] ;

[0065] ;

[0066] in, This indicates that at the t-th period, the current node j receives the historical average demand from the downstream node k. This indicates that at the t-th period, the current node j receives the historical demand standard deviation from the downstream node k.

[0067] Step S53: Calculate the initial safety stock for the current node j based on the calculated historical demand standard deviation, using the following formula:

[0068] ;

[0069] in, This represents the initial safety stock of the current node j; This represents the safety stock coefficient of the current node j. According to the traditional ordering strategy, the safety stock coefficient is set to 1.65.

[0070] Step S54: Calculate the initial target inventory of current node j based on its initial safety stock, using the following formula:

[0071] ;

[0072] in, This represents the initial target inventory level of the current node j in the t-th period;

[0073] Step S55: Calculate the initial order quantity from current node j to upstream node i based on the initial target inventory of current node j, using the following formula:

[0074] ;

[0075] in, This represents the initial order quantity from the current node j to the upstream node i in the t-th period, and max represents taking the maximum value;

[0076] Step S56: Since the supply quantity from upstream node i to current node j is limited by the supply capacity of upstream node i, the initial order quantity from current node j to upstream node i needs to be feasibility-adjusted, as shown in the following formula:

[0077] ;

[0078] in, This represents the initial feasible order quantity from the current node j to the upstream node i in the t-th period, and min indicates taking the minimum value. This represents the order quantity from the current node j to the upstream node i at the t-th period.

[0079] Furthermore, in step S6, the multi-agent system is improved by combining the unified mapping strategy of supply and demand acceleration and upstream supply constraints, adding an environmental perception layer to generate a dynamic safety stock strategy, and executing the final ordering decision; the specific steps are as follows:

[0080] Step S61: Deploy an agent for each node in the current node set J. The multi-agent A is defined as:

[0081] ;

[0082] Each agent The predicted demand of the current node j is the predicted demand amount received by the downstream node k in the (t+1)th period. This information is shared among all nodes in the current node set J to form shared prediction information, as shown in the following formula:

[0083] ;

[0084] in, Indicates origin from multiple agents Shared prediction information This represents all nodes in the current node set except for the current node j; This indicates the predicted demand received by node m from downstream node k in the (t+1)th period.

[0085] Step S62: Obtain the input state of the multi-agent at the t-th cycle, as shown in the following formula:

[0086] ;

[0087] in, This represents the input state of the multi-agent system in the t-th cycle;

[0088] Step S63: Set up a hidden layer for multiple agents, and add an environmental perception layer based on the unified mapping strategy of supply and demand acceleration, as shown in the following formula:

[0089] ;

[0090] ;

[0091] in, This represents the hidden layer of the multi-agent system during the t-th cycle; This represents the activation function. ; Represents the weights of the hidden layer neural network in a multi-agent system; This represents the bias term in a multi-agent hidden layer neural network. The environment perception layer of the multi-agent neural network is represented by the weights and biases of the hidden layer neural network. The historical demand of the previous t-1 cycles is used as input, and the optimal safety stock coefficient obtained by backtracking calculation is used as the supervision signal. The weights and biases of the hidden layer neural network are automatically updated along the gradient direction by iterating 100 times.

[0092] Step S64: Obtain the output state of the multi-agent system in the t-th cycle, as shown in the following formula:

[0093]

[0094] in, This represents the output state of the multi-agent system in the t-th cycle; This indicates that the output will not exceed the realistically feasible boundary during the current cycle. Represents the weights of the output layer neural network of a multi-agent system; The bias term of the output layer neural network of the agent is represented by the weights and bias terms of the multi-agent output layer neural network. The historical demand of the previous t-1 cycles is used as input, and the optimal safety stock coefficient obtained by backtracking calculation is used as the supervision signal. The weights and bias terms of the multi-agent output layer neural network are automatically updated along the gradient direction by iterating 100 times.

[0095] Step S65: Obtain the safety stock of the current node j through the output state of the multi-agent system, using the following formula:

[0096] ;

[0097] in, This represents the safety stock of node j in the t-th period;

[0098] Step S66: Before making an ordering decision, further adjust the target inventory level based on the actual supply in the current period. Recalculate the target inventory level for the current node j based on the safety stock level at the current node j, using the following formula:

[0099] ;

[0100] in, This represents the target inventory level of node j in the t-th period;

[0101] Step S67: Calculate the order quantity from current node j to upstream node i based on the target inventory quantity of current node j, using the following formula:

[0102] ;

[0103] in, This represents the order quantity from the current node j to the upstream node i in the t-th period, and max represents taking the maximum value;

[0104] Step S68: Since the supply quantity from upstream node i to current node j is limited by the supply capacity of upstream node i, the order quantity from current node j to upstream node i needs to be finally adjusted, as shown in the following formula:

[0105] ;

[0106] in, This represents the final order quantity from the current node j to the upstream node i at the t-th period.

[0107] An intelligent supply chain inventory optimization system, applied to the aforementioned intelligent supply chain inventory optimization method, the system comprising:

[0108] The data acquisition module is used to obtain actual demand, actual inventory, out-of-stock items, and items in transit.

[0109] The update module is used to update the actual status of the inventory;

[0110] The allocation module is used to match actual demand with actual inventory levels.

[0111] The forecasting module is used to predict the actual demand and supply capacity for the next cycle;

[0112] The phase module is used to generate a unified mapping strategy for the accelerating supply and demand situation;

[0113] The perception module is used to improve the explicit perception of the accelerating supply and demand situation by multiple agents;

[0114] The ordering module is used to generate order quantities based on forecast results;

[0115] The decision-making module is used to execute ordering decisions.

[0116] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the intelligent supply chain inventory optimization method.

[0117] A computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the aforementioned intelligent supply chain inventory optimization method.

[0118] Compared with existing technologies, the present invention has the following advantages:

[0119] (1) This invention improves multi-agent systems by driving upstream and downstream forecasts, adds an environmental perception layer, and dynamically generates a safety stock strategy, which can adjust the inventory level in real time according to demand fluctuations and changes in upstream supply capacity.

[0120] (2) This invention realizes that the supply of upstream nodes and the actual demand of downstream nodes are considered when making ordering decisions, so that the inventory strategy is no longer limited to the local optimum of a single node, and improves the operational efficiency and service capabilities of the upstream and downstream of the supply chain.

[0121] (3) Based on the prediction results, the present invention judges the phase of supply and demand changes and adopts a unified mapping strategy for the acceleration of supply and demand. It can cope with extreme fluctuation scenarios such as prediction errors, sudden increases in demand and supply interruptions, and reduce the risk of shortages.

[0122] (4) This invention achieves dynamic, flexible and globally collaborative optimization of inventory decision-making by using upstream and downstream forecasting, supply and demand change phase generation supply and demand acceleration situation unified mapping strategy and improved multi-agent, increasing environmental perception, dynamically executing inventory strategy considering supply constraints. Attached Figure Description

[0123] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0124] Please see Figure 1 The present invention provides a technical solution: an intelligent supply chain inventory optimization method, characterized by comprising the following steps:

[0125] Step S1: Collect the actual inventory and order lead time of the current node within the enterprise inventory system information period, the actual demand received by the current node from the downstream node, the stock shortage owed by the current node to the downstream node, the amount in transit from the upstream node to the current node, and the amount supplied by the upstream node to the current node.

[0126] Step S2: Update the actual inventory of the current node based on the amount in transit from the upstream node to the current node;

[0127] Step S3: Based on the updated actual inventory of the current node, allocate the inventory proportionally according to the actual demand received by the current node from the downstream nodes.

[0128] Step S4: Use a long short-term memory neural network to predict the actual demand received by the current node from downstream nodes and the supply from upstream nodes to the current node.

[0129] Step S5: Based on the forecast results, determine the phase of supply and demand changes, and adopt a unified mapping strategy for accelerated supply and demand trends and upstream supply constraints.

[0130] Step S6: Combining the unified mapping strategy of supply and demand acceleration and upstream supply constraints, improve the multi-agent system, add an environmental perception layer to generate a dynamic safety stock strategy, and execute the final ordering decision.

[0131] Furthermore, in step S1, the actual inventory level and order lead time of the current node within the enterprise inventory system information period are collected, along with the actual demand received by the current node from downstream nodes, the stockout owed by the current node to downstream nodes, the amount in transit from upstream nodes to the current node, and the amount supplied by upstream nodes to the current node; the formulas are as follows:

[0132] ;

[0133] ;

[0134] ;

[0135] ;

[0136] ;

[0137] ;

[0138] Where j represents the current node, i represents the upstream node, and k represents the downstream node. This indicates the actual demand received by current node j from downstream node k in the t-th period; J represents the set of all current nodes, corresponding to the existence of multiple inventory points for the enterprise; This represents the set of downstream nodes served by the current node j; Indicates the total collection period; This represents the actual inventory level of node j in the t-th period; This represents the amount of stock shortage that current node j owes to downstream node k in the t-th period; This represents the amount of time that is in transit from upstream node i to current node j in the t-th period; Let represent the amount of goods supplied by upstream node i to current node j in the t-th period; I represents the set of upstream nodes that can supply current node j. This represents the order lead time of the current node j in the t-th period.

[0139] Furthermore, in step S2, the actual inventory of the current node is updated based on the amount in transit from the upstream node to the current node; the specific steps are as follows:

[0140] Step S21, receive incoming goods in transit. The lead time from upstream node i to current node j is... In the t-th period, the current node j receives the data from the node in the t-th period. The amount of goods sent by upstream node i in each cycle;

[0141] Step S22, update the in-transit inventory. After the current node j receives the goods, the in-transit inventory from the upstream node i to the current node j decreases; the formula is as follows:

[0142] ;

[0143] in, This represents the amount of time that is in transit from upstream node i to current node j during the (t-1)th period; Indicates the first The order quantity from the current node j to the upstream node i in each cycle;

[0144] Step S23: Update the actual inventory. After the current node j receives the goods, the actual inventory increases; the formula is as follows:

[0145] ;

[0146] in, This represents the actual inventory of node j in the (t-1)th period.

[0147] Furthermore, in step S3, the inventory of the current node is allocated proportionally according to the actual demand received by the current node from downstream nodes, based on the updated actual inventory level of the current node; the specific steps are as follows:

[0148] Step S31, Receive demand: In the t-th period, the current node j receives the actual demand from the downstream node k. ;

[0149] Step S32: Determine whether the updated actual inventory of the current node meets the actual demand received by the current node from the downstream node; the specific steps are as follows:

[0150] when When the current node's actual inventory meets the actual demand received by the current node from downstream nodes, update the current node's shipments to downstream nodes, the current node's actual inventory, and the current node's stockout to downstream nodes; the formula is as follows:

[0151] ;

[0152] ;

[0153] ;

[0154] in, This represents the amount of goods shipped from the current node j to the downstream node k during the t-th period.

[0155] when When this occurs, it indicates that the current node's actual inventory cannot meet the actual demand received by the current node from downstream nodes. For downstream nodes, the inventory is allocated proportionally based on the actual demand received by the current node from downstream nodes; the formula is as follows:

[0156] ;

[0157] in, This represents the amount of stock shortage that current node j owes to downstream node k in the (t-1)th period;

[0158] Update the current node's outstanding inventory to downstream nodes using the following formula:

[0159] ;

[0160] Update the current node's actual inventory to zero using the following formula:

[0161] .

[0162] Furthermore, in step S4, a long short-term memory neural network is used to predict the actual demand received by the current node from downstream nodes and the supply volume from upstream nodes to the current node; the specific steps are as follows:

[0163] Step S41: Obtain the actual demand input sequence and the supply quantity input sequence based on a sliding window. The sliding window length is w periods. Extract the actual demand quantity received by the current node j from the downstream node over the past w periods to obtain the actual demand input sequence, as shown in the following formula:

[0164] ;

[0165] in, Indicates the first During each cycle, the current node j receives the actual demand from the downstream node k. Indicates the first During each cycle, the current node j receives the actual demand from the downstream node k, where w represents the sliding window length.

[0166] The supply quantity input sequence is obtained by extracting the supply quantity from upstream node i to current node j over the past w periods, as shown in the following formula:

[0167] ;

[0168] in, They represent in The amount of goods supplied from upstream node i to current node j in each cycle, The amount of goods supplied from upstream node i to current node j in each cycle;

[0169] Step S42: Based on the obtained actual demand input sequence, use a long short-term memory neural network to predict the predicted demand received by the current node j from the downstream node k in the (t+1)th period, as shown in the following formula:

[0170] ;

[0171] in, This indicates the predicted demand received by the current node j from the downstream node k in the (t+1)th period. This represents a long short-term memory neural network;

[0172] Step S43: Based on the obtained supply quantity input sequence, use a long short-term memory neural network to predict the predicted supply quantity from upstream node i to current node j in the (t+1)th period; the formula is as follows:

[0173] ;

[0174] in, This represents the predicted supply volume from upstream node i to current node j in the (t+1)th period.

[0175] Furthermore, in step S5, the phase of supply and demand changes is determined based on the forecast results, and a unified mapping strategy for accelerated supply and demand trends and upstream supply constraints are adopted; the specific steps are as follows:

[0176] Step S51: Based on the changes in demand and supply within adjacent periods, calculate the phase of the demand change rhythm and the phase of the supply change rhythm, and obtain the phase difference between the supply and demand change rhythms, as shown in the following formula:

[0177] ;

[0178] in, This represents the phase difference in the supply and demand rhythm at the current node at the nth cycle, used to characterize the relationship between the rate of change of demand and supply over time; when When the rate of change in demand exceeds the rate of change in supply, there is a potential risk of shortages, indicating a strong demand phase; when When the rate of increase in demand matches the rate of increase in supply, supply is in a stable state, which is the equilibrium phase; when... When the rate of change in demand is less than the rate of change in supply, a supply redundancy occurs, which is a strong supply phase. Represents the arctangent function. This indicates an extremely small positive number that avoids a denominator of zero, and its value is 0.00001;

[0179] The above three strategies can be represented using a unified mapping strategy of accelerating supply and demand, as shown in the following formula:

[0180] ;

[0181] in, This represents the unified mapping strategy for the accelerating supply and demand situation at the current node j in the t-th period;

[0182] Step S52: Based on the actual demand received by the current node j from the downstream node k within the collected enterprise inventory system information period, calculate the historical average demand and historical demand standard deviation; the formula is as follows:

[0183] ;

[0184] ;

[0185] in, This indicates that at the t-th period, the current node j receives the historical average demand from the downstream node k. This indicates that at the t-th period, the current node j receives the historical demand standard deviation from the downstream node k.

[0186] Step S53: Calculate the initial safety stock for the current node j based on the calculated historical demand standard deviation, using the following formula:

[0187] ;

[0188] in, This represents the initial safety stock of the current node j; This represents the safety stock coefficient of the current node j. According to the traditional ordering strategy, the safety stock coefficient is set to 1.65.

[0189] Step S54: Calculate the initial target inventory of current node j based on its initial safety stock, using the following formula:

[0190] ;

[0191] in, This represents the initial target inventory level of the current node j in the t-th period;

[0192] Step S55: Calculate the initial order quantity from current node j to upstream node i based on the initial target inventory of current node j, using the following formula:

[0193] ;

[0194] in, This represents the initial order quantity from the current node j to the upstream node i in the t-th period, and max represents taking the maximum value;

[0195] Step S56: Since the supply quantity from upstream node i to current node j is limited by the supply capacity of upstream node i, the initial order quantity from current node j to upstream node i needs to be feasibility-adjusted, as shown in the following formula:

[0196] ;

[0197] in, This represents the initial feasible order quantity from the current node j to the upstream node i in the t-th period, and min indicates taking the minimum value. This represents the order quantity from the current node j to the upstream node i at the t-th period.

[0198] Furthermore, in step S6, the multi-agent system is improved by combining the unified mapping strategy of supply and demand acceleration and upstream supply constraints, adding an environmental perception layer to generate a dynamic safety stock strategy, and executing the final ordering decision; the specific steps are as follows:

[0199] Step S61: Deploy an agent for each node in the current node set J. The multi-agent A is defined as:

[0200] ;

[0201] Each agent The predicted demand of the current node j is the predicted demand amount received by the downstream node k in the (t+1)th period. This information is shared among all nodes in the current node set J to form shared prediction information, as shown in the following formula:

[0202] ;

[0203] in, Indicates origin from multiple agents Shared prediction information This represents all nodes in the current node set except for the current node j; This indicates the predicted demand received by node m from downstream node k in the (t+1)th period.

[0204] Step S62: Obtain the input state of the multi-agent at the t-th cycle, as shown in the following formula:

[0205] ;

[0206] in, This represents the input state of the multi-agent system in the t-th cycle;

[0207] Step S63: Set up a hidden layer for multiple agents, and add an environmental perception layer based on the unified mapping strategy of supply and demand acceleration, as shown in the following formula:

[0208] ;

[0209] ;

[0210] in, This represents the hidden layer of the multi-agent system during the t-th cycle; This represents the activation function. ; Represents the weights of the hidden layer neural network in a multi-agent system; This represents the bias term in a multi-agent hidden layer neural network. The environment perception layer of the multi-agent neural network is represented by the weights and biases of the hidden layer neural network. The historical demand of the previous t-1 cycles is used as input, and the optimal safety stock coefficient obtained by backtracking calculation is used as the supervision signal. The weights and biases of the hidden layer neural network are automatically updated along the gradient direction by iterating 100 times.

[0211] Step S64: Obtain the output state of the multi-agent system in the t-th cycle, as shown in the following formula:

[0212]

[0213] in, This represents the output state of the multi-agent system in the t-th cycle; This indicates that the output will not exceed the realistically feasible boundary during the current cycle. Represents the weights of the output layer neural network of a multi-agent system; The bias term of the output layer neural network of the agent is represented by the weights and bias terms of the multi-agent output layer neural network. The historical demand of the previous t-1 cycles is used as input, and the optimal safety stock coefficient obtained by backtracking calculation is used as the supervision signal. The weights and bias terms of the multi-agent output layer neural network are automatically updated along the gradient direction by iterating 100 times.

[0214] Step S65: Obtain the safety stock of the current node j through the output state of the multi-agent system, using the following formula:

[0215] ;

[0216] in, This represents the safety stock of node j in the t-th period;

[0217] Step S66: Before making an ordering decision, further adjust the target inventory level based on the actual supply in the current period. Recalculate the target inventory level for the current node j based on the safety stock level at the current node j, using the following formula:

[0218] ;

[0219] in, This represents the target inventory level of node j in the t-th period;

[0220] Step S67: Calculate the order quantity from current node j to upstream node i based on the target inventory quantity of current node j, using the following formula:

[0221] ;

[0222] in, This represents the order quantity from the current node j to the upstream node i in the t-th period, and max represents taking the maximum value;

[0223] Step S68: Since the supply quantity from upstream node i to current node j is limited by the supply capacity of upstream node i, the order quantity from current node j to upstream node i needs to be finally adjusted, as shown in the following formula:

[0224] ;

[0225] in, This represents the final order quantity from the current node j to the upstream node i at the t-th period.

[0226] An intelligent supply chain inventory optimization system, applied to the aforementioned intelligent supply chain inventory optimization method, the system comprising:

[0227] The data acquisition module is used to obtain actual demand, actual inventory, out-of-stock items, and items in transit.

[0228] The update module is used to update the actual status of the inventory;

[0229] The allocation module is used to match actual demand with actual inventory levels.

[0230] The forecasting module is used to predict the actual demand and supply capacity for the next cycle;

[0231] The phase module is used to generate a unified mapping strategy for the accelerating supply and demand situation;

[0232] The perception module is used to improve the explicit perception of the accelerating supply and demand situation by multiple agents;

[0233] The ordering module is used to generate order quantities based on forecast results;

[0234] The decision-making module is used to execute ordering decisions.

[0235] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the intelligent supply chain inventory optimization method.

[0236] A computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the aforementioned intelligent supply chain inventory optimization method.

[0237] The following specific examples illustrate the inventory optimization effect of the invented method:

[0238] First, a period T=100 was set, and data on the actual demand of each node in the supply chain to its downstream nodes, as well as the actual inventory, stockout, and in-transit quantities of each node, were collected. To facilitate comparison with traditional methods, standard parameters were uniformly set: unit stockout cost was set at 7.5 yuan / unit, unit inventory holding cost at 1.5 yuan / unit, and lead time was predicted based on actual transportation records, with a fluctuation range of 1-3 periods. During the test, real demand was subject to a random disturbance of 20% of the mean fluctuation, used to flexibly allocate inventory in cases of prediction deviation or sudden demand surges to verify the inventory optimization effect of the method in real-world fluctuation scenarios.

[0239] Table 1. Experimental results of the present invention and conventional methods

[0240]

[0241] The results in Table 1 show that, under real-world demand and supply fluctuation scenarios, the method of this invention, through dynamic prediction, judgment of the phase of supply and demand changes to generate a unified mapping strategy for accelerated supply and demand trends, and intelligent safety stock adjustment, achieves a reduction in total inventory cost of RMB 1894.04 (approximately 16.2%) compared to traditional methods. The service rate increases from 87.5% to 91.67%, and the stockout rate decreases from 12.5% ​​to 8.33%. This effectively reduces stockout losses and inventory backlog costs, improves supply chain stability and user experience, demonstrating the significant advantages and practical application value of this invention in accurately matching demand, ensuring supply continuity, and optimizing overall inventory management.

[0242] This invention addresses the challenges faced by enterprises with multiple inventory points, particularly those experiencing significant demand fluctuations during holidays, promotions, and regional demand volatility. It addresses the complexity of inventory management and the need for flexible adjustments to safety stock levels rather than reliance on experience. By incorporating real-world supply chain business fluctuations and utilizing historical upstream and downstream data to predict demand and supply capacity, the invention identifies the phase of supply and demand changes and generates a unified mapping strategy for accelerating supply and demand trends. It improves multi-agent systems by adding an environmental awareness layer and supply constraints, generating safety stock strategies to supplement forecasting deficiencies and address sudden fluctuations. This leads to the construction of an intelligent supply chain inventory decision-making method, which is then optimized. The model is rapidly executed using a computer programming language, forming a closed-loop optimization process from prediction and decision-making to execution. This invention improves service rates and reduces stockout rates while better adapting to real-world business fluctuations and the supply and demand characteristics of different nodes through intelligent dynamic decision-making. It overcomes the shortcomings of traditional inventory management, which relies on fixed rules, lacks flexibility, and is difficult to coordinate and optimize, providing theoretical and practical basis for supply chain management decisions. This method can reduce stockout rates and improve service levels, offering a new perspective and approach for intelligent supply chain inventory management research and practice.

[0243] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent supply chain inventory optimization method, characterized in that: The method includes the following steps: Step S1: Collect the actual inventory and order lead time of the current node within the enterprise inventory system information period, the actual demand received by the current node from the downstream node, the stock shortage owed by the current node to the downstream node, the amount in transit from the upstream node to the current node, and the amount supplied by the upstream node to the current node. Step S2: Update the actual inventory of the current node based on the amount in transit from the upstream node to the current node; Step S3: Based on the updated actual inventory of the current node, allocate the inventory proportionally according to the actual demand received by the current node from the downstream nodes. Step S4: Use a long short-term memory neural network to predict the actual demand received by the current node from downstream nodes and the supply from upstream nodes to the current node. Step S5: Based on the forecast results, determine the phase of supply and demand changes, and adopt a unified mapping strategy for accelerated supply and demand trends and upstream supply constraints. Specifically, based on the changes in demand and supply in adjacent cycles, the phase of demand change rhythm and the phase of supply change rhythm are calculated respectively, and the phase difference of supply and demand change rhythm is obtained. The unified mapping strategy for the accelerating supply and demand trend is represented by the following formula: ； in, This represents the unified mapping strategy for the accelerating supply and demand situation at the current node j in the t-th period. This represents the phase difference in the supply and demand change rhythm of the current node at the t-th cycle, used to characterize the relationship between the speed of change of demand and supply over time. Step S6: Combining the unified mapping strategy of accelerating supply and demand and upstream supply constraints, improve the multi-agent system, add an environmental perception layer to generate a dynamic safety stock strategy, and execute the final ordering decision; the specific steps are as follows: Step S61: Deploy an agent for each node in the current node set J. Each intelligent agent The predicted demand of the current node j is the predicted demand amount received by the downstream node k in the (t+1)th period. This information is shared among all nodes in the current node set J, forming shared prediction information. Step S62: Obtain the input state of the multi-agent at the t-th cycle; Step S63: Set up a hidden layer for multiple agents, and add an environmental perception layer based on the unified mapping strategy of supply and demand acceleration, as shown in the following formula: ； ； in, This represents the hidden layer of the multi-agent system during the t-th cycle; This represents the activation function. ; Represents the weights of the hidden layer neural network in a multi-agent system; This represents the input state of the multi-agent system in the t-th cycle; This represents the bias term in a multi-agent hidden layer neural network. The environment perception layer of the multi-agent neural network is represented by the weights and biases of the hidden layer neural network. The historical demand of the previous t-1 cycles is used as input, and the optimal safety stock coefficient obtained by backtracking calculation is used as the supervision signal. The weights and biases of the hidden layer neural network are automatically updated along the gradient direction by iterating 100 times. Step S64: Obtain the output state of the multi-agent at the t-th cycle; Step S65: Obtain the safety stock of the current node j through the output state of the multi-agent; Step S66: Before making an ordering decision, further adjust the target inventory based on the actual supply in the current cycle, and recalculate the target inventory of the current node j based on the safety stock of the current node j. Step S67: Calculate the order quantity from the current node j to the upstream node i based on the target inventory quantity of the current node j. Step S68: Since the supply quantity from upstream node i to current node j is limited by the supply capacity of upstream node i, the order quantity from current node j to upstream node i needs to be finally revised.

2. The intelligent supply chain inventory optimization method according to claim 1, characterized in that, Step S1 involves collecting the actual inventory level and order lead time of the current node within the enterprise inventory system information period, the actual demand received by the current node from downstream nodes, the stockout owed by the current node to downstream nodes, the amount in transit from upstream nodes to the current node, and the amount supplied from upstream nodes to the current node; the formula is as follows: ； ； ； ； ； ； Where j represents the current node, i represents the upstream node, and k represents the downstream node. This indicates the actual demand received by current node j from downstream node k in the t-th period; J represents the set of all current nodes, corresponding to the existence of multiple inventory points for the enterprise; This represents the set of downstream nodes served by the current node j; Indicates the total collection period; This represents the actual inventory level of node j in the t-th period; This represents the amount of stock shortage that current node j owes to downstream node k in the t-th period; This represents the amount of time that is in transit from upstream node i to current node j in the t-th period; Let represent the amount of goods supplied by upstream node i to current node j in the t-th period; I represents the set of upstream nodes that can supply current node j. This represents the order lead time of the current node j in the t-th period.

3. The intelligent supply chain inventory optimization method according to claim 2, characterized in that, In step S2, the actual inventory of the current node is updated based on the amount in transit from the upstream node to the current node; the specific steps are as follows: Step S21, receive incoming goods in transit. The lead time from upstream node i to current node j is... In the t-th period, the current node j receives the data from the node in the t-th period. The amount of goods sent by upstream node i in each cycle; Step S22, update the in-transit inventory. After the current node j receives the goods, the in-transit inventory from the upstream node i to the current node j decreases; the formula is as follows: ； in, This represents the amount of time that is in transit from upstream node i to current node j during the (t-1)th period; Indicates the first The order quantity from the current node j to the upstream node i in each cycle; Step S23: Update the actual inventory. After the current node j receives the goods, the actual inventory increases; the formula is as follows: ； in, This represents the actual inventory of node j in the (t-1)th period.

4. The intelligent supply chain inventory optimization method according to claim 3, characterized in that, In step S3, the inventory of the current node is allocated proportionally according to the actual demand received by the current node from downstream nodes, based on the updated actual inventory level of the current node. The specific steps are as follows: Step S31, Receive demand: In the t-th period, the current node j receives the actual demand from the downstream node k. ; Step S32: Determine whether the updated actual inventory of the current node meets the actual demand received by the current node from the downstream node; the specific steps are as follows: when When the current node's actual inventory meets the actual demand received by the current node from downstream nodes, update the current node's shipments to downstream nodes, the current node's actual inventory, and the current node's stockout to downstream nodes; the formula is as follows: ； ； ； in, This represents the amount of goods shipped from the current node j to the downstream node k during the t-th period. when When this occurs, it indicates that the current node's actual inventory cannot meet the actual demand received by the current node from downstream nodes. For downstream nodes, the inventory is allocated proportionally based on the actual demand received by the current node from downstream nodes; the formula is as follows: ； in, This represents the amount of stock shortage that current node j owes to downstream node k in the (t-1)th period; Update the current node's outstanding inventory to downstream nodes using the following formula: ； Update the current node's actual inventory to zero using the following formula: 。 5. The intelligent supply chain inventory optimization method according to claim 4, characterized in that, Step S4 uses a long short-term memory neural network to predict the actual demand received by the current node from downstream nodes and the supply from upstream nodes to the current node; the specific steps are as follows: Step S41: Obtain the actual demand input sequence and the supply quantity input sequence based on a sliding window. The sliding window length is w periods. Extract the actual demand quantity received by the current node j from the downstream node over the past w periods to obtain the actual demand input sequence, as shown in the following formula: ； in, Indicates the first During each cycle, the current node j receives the actual demand from the downstream node k. Indicates the first During each cycle, the current node j receives the actual demand from the downstream node k, where w represents the sliding window length. The supply quantity input sequence is obtained by extracting the supply quantity from upstream node i to current node j over the past w periods, as shown in the following formula: ； in, They represent in The amount of goods supplied from upstream node i to current node j in each cycle, The amount of goods supplied from upstream node i to current node j in each cycle; Step S42: Based on the obtained actual demand input sequence, use a long short-term memory neural network to predict the predicted demand received by the current node j from the downstream node k in the (t+1)th period, as shown in the following formula: ； in, This indicates the predicted demand received by the current node j from the downstream node k in the (t+1)th period. This represents a long short-term memory neural network; Step S43: Based on the obtained supply quantity input sequence, use a long short-term memory neural network to predict the predicted supply quantity from upstream node i to current node j in the (t+1)th period; the formula is as follows: ； in, This represents the predicted supply volume from upstream node i to current node j in the (t+1)th period.

6. The intelligent supply chain inventory optimization method according to claim 5, characterized in that, In step S5, the phase of supply and demand changes is determined based on the forecast results, and a unified mapping strategy for the accelerated trend of supply and demand and upstream supply constraints are adopted. The specific steps are as follows: Step S51: Based on the changes in demand and supply within adjacent periods, calculate the phase of the demand change rhythm and the phase of the supply change rhythm, and obtain the phase difference between the supply and demand change rhythms, as shown in the following formula: ； in, This represents the phase difference in the supply and demand rhythm at the current node at the nth cycle, used to characterize the relationship between the rate of change of demand and supply over time; when When the rate of change in demand exceeds the rate of change in supply, there is a potential risk of shortages, indicating a strong demand phase; when When the rate of increase in demand matches the rate of increase in supply, supply is in a stable state, which is the equilibrium phase; when... When the rate of change in demand is less than the rate of change in supply, a supply redundancy occurs, which is a strong supply phase. Represents the arctangent function. This indicates an extremely small positive number that avoids a denominator of zero, and its value is 0.00001; The above three strategies can be represented using a unified mapping strategy of accelerating supply and demand, as shown in the following formula: ； in, This represents the unified mapping strategy for the accelerating supply and demand situation at the current node j in the t-th period; Step S52: Based on the actual demand received by the current node j from the downstream node k within the collected enterprise inventory system information period, calculate the historical average demand and historical demand standard deviation; the formula is as follows: ； ； in, This indicates that at the t-th period, the current node j receives the historical average demand from the downstream node k. This indicates that at the t-th period, the current node j receives the historical demand standard deviation from the downstream node k. Step S53: Calculate the initial safety stock for the current node j based on the calculated historical demand standard deviation, using the following formula: ； in, This represents the initial safety stock of the current node j; This represents the safety stock coefficient of the current node j. According to the traditional ordering strategy, the safety stock coefficient is set to 1.

65. Step S54: Calculate the initial target inventory of current node j based on its initial safety stock, using the following formula: ； in, This represents the initial target inventory level of the current node j in the t-th period; Step S55: Calculate the initial order quantity from current node j to upstream node i based on the initial target inventory of current node j, using the following formula: ； in, This represents the initial order quantity from the current node j to the upstream node i in the t-th period, and max represents taking the maximum value; Step S56: Since the supply quantity from upstream node i to current node j is limited by the supply capacity of upstream node i, the initial order quantity from current node j to upstream node i needs to be feasibility-adjusted, as shown in the following formula: ； in, This represents the initial feasible order quantity from the current node j to the upstream node i in the t-th period, and min indicates taking the minimum value. This represents the order quantity from the current node j to the upstream node i at the t-th period.

7. The intelligent supply chain inventory optimization method according to claim 6, characterized in that, In step S6, the unified mapping strategy of accelerating supply and demand and upstream supply constraints are combined to improve the multi-agent system, add an environmental perception layer to generate a dynamic safety stock strategy, and execute the final ordering decision; the specific steps are as follows: Step S61: Deploy an agent for each node in the current node set J. The multi-agent A is defined as: ； Each agent The predicted demand of the current node j is the predicted demand amount received by the downstream node k in the (t+1)th period. This information is shared among all nodes in the current node set J to form shared prediction information, as shown in the following formula: ； in, Indicates origin from multiple agents Shared prediction information This represents all nodes in the current node set except for the current node j; This indicates the predicted demand received by node m from downstream node k in the (t+1)th period. Step S62: Obtain the input state of the multi-agent at the t-th cycle, as shown in the following formula: ； in, This represents the input state of the multi-agent system in the t-th cycle; Step S63: Set up a hidden layer for multiple agents, and add an environmental perception layer based on the unified mapping strategy of supply and demand acceleration, as shown in the following formula: ； ； in, This represents the hidden layer of the multi-agent system during the t-th cycle; This represents the activation function. ; Represents the weights of the hidden layer neural network in a multi-agent system; This represents the bias term in a multi-agent hidden layer neural network. The environment perception layer of the multi-agent neural network is represented by the weights and biases of the hidden layer neural network. The historical demand of the previous t-1 cycles is used as input, and the optimal safety stock coefficient obtained by backtracking calculation is used as the supervision signal. The weights and biases of the hidden layer neural network are automatically updated along the gradient direction by iterating 100 times. Step S64: Obtain the output state of the multi-agent system in the t-th cycle, as shown in the following formula: ； in, This represents the output state of the multi-agent system in the t-th cycle; This indicates that the output will not exceed the realistically feasible boundary during the current cycle. Represents the weights of the output layer neural network of a multi-agent system; The bias term of the output layer neural network of the agent is represented by the weights and bias terms of the multi-agent output layer neural network. The historical demand of the previous t-1 cycles is used as input, and the optimal safety stock coefficient obtained by backtracking calculation is used as the supervision signal. The weights and bias terms of the multi-agent output layer neural network are automatically updated along the gradient direction by iterating 100 times. Step S65: Obtain the safety stock of the current node j through the output state of the multi-agent system, using the following formula: ； in, This represents the safety stock of node j in the t-th period; Step S66: Before making an ordering decision, further adjust the target inventory level based on the actual supply in the current period. Recalculate the target inventory level for the current node j based on the safety stock level at the current node j, using the following formula: ； in, This represents the target inventory level of node j in the t-th period; Step S67: Calculate the order quantity from current node j to upstream node i based on the target inventory quantity of current node j, using the following formula: ； in, This represents the order quantity from the current node j to the upstream node i in the t-th period, and max represents taking the maximum value; Step S68: Since the supply quantity from upstream node i to current node j is limited by the supply capacity of upstream node i, the order quantity from current node j to upstream node i needs to be finally adjusted, as shown in the following formula: ； in, This represents the final order quantity from the current node j to the upstream node i at the t-th period.

8. An intelligent supply chain inventory optimization system, applied to the intelligent supply chain inventory optimization method described in claim 7, characterized in that, The system includes: The data acquisition module is used to obtain actual demand, actual inventory, out-of-stock items, and items in transit. The update module is used to update the actual status of the inventory; The allocation module is used to match actual demand with actual inventory levels. The forecasting module is used to predict the actual demand and supply capacity for the next cycle; The phase module is used to generate a unified mapping strategy for the accelerating supply and demand situation; The perception module is used to improve the explicit perception of the accelerating supply and demand situation by multiple agents; The ordering module is used to generate order quantities based on forecast results; The decision-making module is used to execute ordering decisions.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements an intelligent supply chain inventory optimization method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements an intelligent supply chain inventory optimization method as described in any one of claims 1 to 7.

Images