E-commerce supply chain anti-shortage decision-making system based on multi-source data fusion

Abstract

The invention relates to the technical field of logistics supply, and discloses a multi-source data fusion-based e-commerce supply chain anti-break decision-making system, which comprises a heterogeneous graph network construction and risk prediction module, a multi-source data fusion-based e-commerce supply chain anti-break decision-making module, a multi-source data fusion-based e-commerce supply chain anti-break decision-making module and a multi-source data fusion-based e-commerce supply chain anti-break decision-making module, a time sequence sensitive graph attention network is used to predict a goods shortage risk propagation path; the elastic recovery path generation module is used for modeling supply chain recovery into a sequence decision problem on a graph based on the constructed heterogeneous graph network, and generating a plurality of elastic recovery paths through a variational inference method; according to the method, the heterogeneous graph network is constructed and the time-sequence-sensitive graph attention network is realized, so that the system can accurately capture the cascade effect and the risk propagation path in the complex supply network, the prediction accuracy of the cargo shortage risk propagation path is improved, and false report and missing report are greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of logistics supply, and more specifically, to an out-of-stock prevention decision-making system for e-commerce supply chains based on multi-source data fusion. Background Art

[0002] The out-of-stock problem in e-commerce supply chains seriously affects user experience and platform revenue. Existing technologies have limitations in solving such problems: existing out-of-stock warning systems mainly rely on a single data source (such as historical sales data or inventory data) for simple statistical analysis and prediction, and cannot comprehensively capture the complex information of each link in the supply chain, resulting in limited prediction accuracy. Existing methods ignore the mutual influence and implicit association between multi-source data and cannot effectively cope with the complex and changeable e-commerce environment; moreover, traditional supply chain risk analysis methods usually adopt linear models or simple network analysis and cannot effectively capture the cascade effect and risk propagation path in complex supply networks. When a problem occurs in a certain link of the supply chain, it is difficult to accurately predict the impact range and degree on the overall supply network; at the same time, existing supply chain recovery methods are mostly based on deterministic models and lack sufficient modeling of uncertainty, resulting in insufficient robustness of recovery strategies in a highly uncertain environment. Traditional methods cannot effectively consider various possible scenarios and risk factors and are difficult to adapt to the dynamic changes of e-commerce supply chains. Summary of the Invention

[0003] The present invention provides an out-of-stock prevention decision-making system for e-commerce supply chains based on multi-source data fusion to solve the technical problem of out-of-stock prevention in related e-commerce supply chains.

[0004] The present invention provides an out-of-stock prevention decision-making system for e-commerce supply chains based on multi-source data fusion, including: Heterogeneous graph network construction and risk prediction module: construct a heterogeneous graph network representing e-commerce supply chain relationships based on multi-source data, and use a time-series sensitive graph attention network to predict the out-of-stock risk propagation path; Elastic recovery path generation module: based on the constructed heterogeneous graph network, model the supply chain recovery as a sequential decision-making problem on the graph, and generate multiple elastic recovery paths through variational inference methods; Strategy optimization and uncertainty modeling module: construct a deep probabilistic programming model for multiple recovery paths to model their uncertainties, and use a multi-agent Monte Carlo tree search algorithm to optimize the recovery strategy; Data fusion and strategy execution module: based on the optimized recovery strategy, achieve real-time dynamic fusion of multi-source data, adaptively adjust the data source weights, and transform the abstract strategy into specific executable supply chain adjustment instructions; Closed-loop monitoring and optimization module: construct a closed-loop feedback algorithm for the executed adjustment instructions, continuously monitor the execution effect, and optimize the strategy.

[0005] As a further optimization solution of the present invention, the steps of constructing the heterogeneous graph network include: Collect multi-source data including sales data, inventory data, supplier data, logistics data, and market trend data in the e-commerce supply chain; Preprocess the collected multi-source data, including data cleaning, missing value processing, outlier detection, and data standardization; Construct a heterogeneous graph network representing the e-commerce supply chain relationship, where the nodes include commodity nodes, warehouse nodes, supplier nodes, and logistics nodes, and the edges include supply relationships, inventory relationships, and logistics relationships.

[0006] As a further optimization solution of the present invention, the time-sensitive graph attention network measures the association strength between different nodes by calculating attention coefficients, where the time characteristics of the edges are considered, including time encoding, periodic pattern encoding, and time series relationship encoding, to capture the characteristics of the supply chain relationship changing over time.

[0007] As a further optimization solution of the present invention, the variational inference method generates an elastic recovery path through the following steps: Define the posterior distribution of the recovery path, and introduce a variational distribution as an approximate substitute for the posterior distribution; Optimize the parameters by maximizing the variational lower bound; Integrate the graph network structure into the variational inference model, and use a graph neural network to process the structural information of the supply chain network; Based on Bayesian decision theory, implement a multi-step inference algorithm to evaluate the costs and effects of different recovery paths.

[0008] As a further optimization solution of the present invention, define a set of random variables representing the supply chain state and the conditional dependence relationship between variables; Parameterize the conditional probability distribution using a neural network; Apply the stochastic variational inference algorithm to estimate the posterior distribution.

[0009] As a further optimization solution of the present invention, the multi-agent Monte Carlo tree search algorithm includes: Model the supply chain recovery problem as a multi-agent system, where each agent represents a decision-making entity in the supply chain; Define the system state and the joint action space; Execute the Monte Carlo tree search process, including four steps: selection, expansion, simulation, and backpropagation; Implement a collaborative decision-making method among agents to ensure that the decisions of each agent form a consistent global recovery strategy.

[0010] As a further optimization solution of the present invention, the strategy optimization objective function introducing risk perception includes a reward term and a risk aversion term. Among them, the reward term measures the expected return of the strategy, and the risk aversion term controls the risk level by calculating the difference between the current strategy and the reference strategy.

[0011] As a further optimization solution of the present invention, the steps of real-time dynamic fusion of multi-source data include: Construct a unified data access layer to support access to various data sources; Conduct real-time quality assessment on the accessed multi-source data, including integrity, timeliness, consistency, and accuracy assessment; Adaptive adjustment of data source weights according to the out-of-stock risk scenario and data quality; Based on the allocated weights, achieve feature-level fusion, decision-level fusion, and model-level fusion.

[0012] As a further optimization solution of the present invention, the steps of converting the abstract strategy into specific executable supply chain adjustment instructions include: Parse the optimized recovery strategy into a specific set of execution instructions, including procurement instructions, inventory adjustment instructions, logistics instructions, and pricing instructions; Based on the parsed instruction set, generate a detailed execution plan, including task decomposition, task sequencing, resource allocation, and time planning; Conduct constraint checking on the generated execution plan to ensure its feasibility; Construct a standardized interface with each system of the supply chain to achieve automated execution of instructions.

[0013] A storage medium includes a memory and one or more processors. Executable code is stored in the memory. When the one or more processors execute the executable code, it is used to implement the above-mentioned e-commerce supply chain out-of-stock prevention decision-making system based on multi-source data fusion.

[0014] The beneficial effects of the present invention are as follows: By constructing a heterogeneous graph network and implementing a time-sensitive graph attention network, the system can accurately capture the cascade effect and risk propagation path in the complex supply network, improve the prediction accuracy of the out-of-stock risk propagation path, greatly reduce false alarms and missed reports. At the same time, based on the method of graph structure variational inference and deep probabilistic programming, the system can generate robust recovery strategies in a high-uncertainty environment, improve the robustness of the recovery strategy of the system in a high-uncertainty environment, and can cope with various complex scenarios and perturbation situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a module diagram of the e-commerce supply chain out-of-stock prevention decision-making system based on multi-source data fusion of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0017] In at least one embodiment of the present invention, a decision-making system for preventing out-of-stock in an e-commerce supply chain based on multi-source data fusion is disclosed, as Figure 1 shown, including: Heterogeneous graph network construction and risk prediction module: Construct a heterogeneous graph network representing the e-commerce supply chain relationship based on multi-source data, and use a time-series sensitive graph attention network to predict the out-of-stock risk propagation path; Specifically, it includes the following steps: Step 1.1: Multi-source data collection and preprocessing; Specifically, collect multi-source data in the e-commerce supply chain, including but not limited to: Commodity sales data: including information such as commodity ID, sales volume, sales time, price, etc.; Inventory data: including information such as commodity ID, inventory quantity, inventory location, minimum safety inventory, etc.; Supplier data: including information such as supplier ID, supply capacity, historical supply records, supply cycle, etc.; Logistics data: including information such as logistics path, transportation time, logistics cost, etc.; Market trend data: including information such as seasonal factors, market competition situation, consumer behavior, etc.

[0018] Furthermore, preprocess the collected multi-source data, including data cleaning, missing value processing, outlier detection, data standardization, etc., to form a standardized data set for subsequent analysis.

[0019] Step 1.2: Heterogeneous graph network construction; Specifically, based on the preprocessed multi-source data, construct a heterogeneous graph network representing the e-commerce supply chain relationship: Specifically, represents the constructed heterogeneous graph network; represents the node set; Among them, represents the node set in the entire graph network; , , respectively represent the , , th nodes; is the total number of nodes; represents the edge set; Among them, represents the edge set in the entire graph network; represents an edge from node to node ; represents the starting node of the edge; represents the ending node of the edge; represents the relationship type of the edge; represents the node set; represents the relationship type set; represents the node attribute set; Among them, represents the node attribute set in the entire graph network; represents the attribute set of node ; represents a certain node; represents the node set; represents the relationship type set; Among them, represents the relationship type set in the entire graph network; , , respectively represent the , , th relationship types, is the total number of relationship types; Furthermore, map various types of data into the graph network: commodity data is mapped to commodity nodes and their attributes; warehouse data is mapped to warehouse nodes and their attributes; supplier data is mapped to supplier nodes and their attributes; logistics data is mapped to logistics nodes and their attributes as well as logistics relationship edges; supply chain relationship data is mapped to different types of edges and their attributes.

[0020] Step 1.3, Implementation of the time-sensitive graph attention network; Specifically, construct a time-sensitive graph attention network for learning the important relationships between nodes, and the implementation method is as follows: First, define the calculation formula of the graph attention layer: Among them, represents the feature representation of node in the th layer; represents a non - linear activation function; represents summing over all neighbor nodes of node , that is, performing weighted aggregation on the features of all neighbor nodes of node ; represents the index of neighbor nodes; represents the set of neighbor nodes of node ; represents the attention coefficient of node to node ; represents the weight matrix of the -th layer; ; represents the feature representation of node j in the l - th layer.

[0021] The attention coefficient is calculated as follows: where represents the attention coefficient of node to node ; represents the exponential function; represents the leaky rectified linear activation function; represents the attention vector; represents vector transpose; represents the weight matrix of the -th layer; represents the feature representation of node ; represents the feature representation of node ; represents the feature representation of node ; represents the vector concatenation operation; represents the temporal feature of edge ; represents the temporal feature of edge ; represents summing over all neighbor nodes of node ; represents the index of neighbor nodes; represents the set of neighbor nodes of node ; It should be noted that by introducing the temporal feature , this application can effectively capture the dynamic changes of supply chain relationships over time, which is crucial for accurately predicting the propagation path of out - of - stock risks. ;

[0022] According to an embodiment of the present application, the time-sensitive graph attention network specifically includes an input layer, multiple graph attention layers, and an output layer. The input layer receives a node feature matrix and an adjacency matrix, where the node feature matrix contains the attribute information of various types of nodes (such as goods, warehouses, suppliers, etc.), and the adjacency matrix contains the connection relationships between nodes. The graph attention layers are responsible for calculating the attention coefficients between nodes and updating the node representations, and the output layer generates the final node representations for subsequent risk prediction tasks.

[0023] In a specific implementation, for the characteristics of the e-commerce supply chain scenario, the time series features are constructed in the following way: Time encoding: For the time information of each edge (such as the establishment time of the supply relationship, the most recent interaction time, etc.), apply sine-cosine positional encoding:

[0024] where, represents the positional encoding function; represents the time point; represents the dimension index; represents the sine function; represents the encoding dimension; represents the even-dimensional position; represents the odd-dimensional position; represents the cosine function; represents the scaling factor.

[0025] Periodic pattern encoding: Capture the periodic patterns in the supply chain (such as weekly, monthly, seasonal), through Fourier feature transformation: where, represents the Fourier transform function; represents the time point; represents the sine function; represents the cosine function; represents pi; represents the th period length; represents the th period length; Temporal relationship encoding: Capture the temporal relationships of interactions between nodes, such as the time since the last interaction, the interaction frequency, etc.

[0026] The time series features constructed in this way can effectively capture the time dynamics in the supply chain and improve the accuracy of out-of-stock risk propagation prediction.

[0027] In the specific application scenarios of preventing out - of - stock in the e - commerce supply chain, the time - series sensitive graph attention network can be used in the following typical situations: Out - of - stock risk prediction for seasonal goods: For goods with strong seasonal characteristics (such as holiday supplies, seasonal clothing, etc.), capture historical seasonal patterns through time - series features to identify potential out - of - stock risks in advance. For example, before the Spring Festival, it is identified that some suppliers of New Year goods may not be able to meet the surging demand, leading to the spread of out - of - stock risks to related goods.

[0028] Impact analysis of promotional activities: Before large - scale promotional activities on e - commerce platforms (such as Double Eleven), analyze the time - series patterns in historical promotional data to predict which goods may experience inventory shortages due to promotional activities and identify how such shortages spread along the supply chain.

[0029] Supplier stability assessment: Evaluate the stability of suppliers by analyzing the time - series characteristics of historical supply (such as whether there are periodic delays, quality fluctuations, etc.), and predict potential supply interruption risks and their impacts on downstream goods.

[0030] Step 1.4, Prediction of out - of - stock risk propagation path; Specifically, based on the constructed time - series sensitive graph attention network, realize the prediction of the out - of - stock risk propagation path: Initial risk identification: Identify the initial out - of - stock risk points by analyzing factors such as inventory levels, sales trends, and supplier status.

[0031] Propagation path learning: Use self - supervised learning methods to learn the propagation patterns of out - of - stock risks based on historical out - of - stock event data. The specific implementation is as follows: Construct training samples: Among them, represents the graph structure at time , represents the risk distribution at time , represents the risk distribution at time .

[0032] Training objective: Minimize the difference between the predicted risk and the actual risk: Among them, represents the loss function; represents the square of the two - norm; represents the prediction function with parameter ; represents the graph structure at time ; represents the risk distribution at time . Represents the time risk distribution; Represents the time interval; Represents the model parameters.

[0033] Train the model parameters using historical data .

[0034] High-risk propagator substructure identification: Apply the subgraph mining algorithm to identify the substructures in the graph network that are most likely to have cascading out-of-stock risks. The specific implementation is as follows: Define the risk propagation subgraph scoring function: where represents the subgraph scoring function; represents the subgraph; represents the subgraph the sum of the risk probabilities of all nodes in; represents the th node in the subgraph; represents the node risk probability of; represents the balance parameter; represents the subgraph the sum of the risk propagation probabilities of all edges in; represents the edge risk propagation probability of; represents an edge from node to node , representing the supply chain relationship between two nodes; Use the heuristic search algorithm to find the set of subgraphs with the highest scores.

[0035] Risk propagation path generation: Based on the identified high-risk substructures and the learned propagation patterns, generate the most likely out-of-stock risk propagation paths and calculate the risk probabilities of each path.

[0036] The output results include: The initial out-of-stock risk point list and their risk levels; The risk propagation path graph, including the propagation direction, propagation probability, and time estimate; The identification results of key risk nodes (nodes that may cause large-scale cascading risks).

[0037] It should be noted that through the heterogeneous graph network and the time-sensitive graph attention network, the cascading effects and risk propagation paths in the complex supply chain can be effectively captured, which has significant advantages compared with traditional methods.

[0038] Elastic Recovery Path Generation Module: Based on the constructed heterogeneous graph network, the supply chain recovery is modeled as a sequential decision-making problem on the graph, and multiple elastic recovery paths are generated through variational inference methods; Specifically, it includes the following steps: Step 2.1, Recovery Path Modeling; Model the supply chain recovery as a sequential decision-making problem on the graph as follows: Define the state space : Represents the set of states of each node in the supply chain network, including inventory levels, production capacity, logistics status, etc.; Define the action space : Represents the set of possible recovery operations, such as increasing orders, adjusting inventory, changing suppliers, adjusting logistics paths, etc.; Define the transition function : Represents the probability of transitioning to state after executing action ; Define the reward function : Represents the benefit of executing action in state , considering factors such as recovery efficiency, cost, and time comprehensively.

[0039] Step 2.2, Variational Inference Model Construction; Construct a path generation model based on variational inference to generate robust recovery paths in a highly uncertain environment; the specific implementation is as follows: Posterior distribution modeling: Define the posterior distribution of the recovery path. Directly calculating the posterior distribution is usually difficult to handle, so a variational method is used for approximation.

[0040] Variational lower bound construction: Introduce a variational distribution as an approximation of the posterior distribution , and optimize the parameters by maximizing the variational lower bound (ELBO): where represents the variational lower bound loss function; represents the parameters of the generative model; represents the parameters of the inference model; represents the expectation operation; represents the variational distribution with parameter ; represents the latent recovery path variable; represents the observed supply chain state; represents the logarithmic function; represents the variational distribution with parameter The distribution of the generative model; denotes the KL divergence; denotes the prior distribution; Specifically: denotes the reconstruction term, encouraging the model to reconstruct the observed data based on the latent variable ; ; denotes the KL divergence term, ensuring that the variational distribution does not deviate too far from the prior distribution .

[0041] Graph structure integration: Integrate the graph network structure into the variational inference model and use graph neural networks to process the structural information of the supply chain network: Encoder network: Where: denotes the variational distribution with parameters ; denotes the latent recovery path variable; denotes the observed supply chain state; denotes the normal distribution; denotes the mean function; denotes the variance function; denotes the observed data corresponding graph structure.

[0042] The decoder network maps the latent variable back to the supply chain state space, and its expression is: Where, denotes the distribution of the generative model with parameters ; denotes the observed supply chain state; denotes the latent recovery path variable; denotes the decoder function with parameters ; denotes the graph structure of the supply chain network.

[0043] It should be noted that by integrating the graph structure information into the variational inference process, the present application can more accurately capture the complex dependencies in the supply chain network, thereby generating a recovery path that is more in line with the actual situation.

[0044] According to an embodiment of the present application, the specific implementation of the graph-structured variational inference model includes the following network structures: Graph structure encoder: Use a graph convolutional network ( Graph Convolutional Network, the graph structure information is processed by a graph convolutional network (GCN) or a graph attention network (GAT) to generate node representations and graph representations .

[0045] Node embedding layer: Maps node features to the latent space; Multi-layer graph convolution: Aggregates neighbor node information to update node representations; Graph pooling layer: Generates a global representation of the entire graph; Variational encoder: Generates distribution parameters of latent variables based on the graph representation; Mean network: where, represents the mean function with parameter ; represents the input data; represents the multi-layer perceptron network used to calculate the mean; represents the global representation of the graph; Variance network: where, represents the logarithmic function; represents the variance function with parameter ; represents the input data; represents the multi-layer perceptron network used to calculate the variance; represents the global representation of the graph.

[0046] Sampling operation: where, represents the sampled latent variable; represents the mapping result of the mean function with parameter to the input ; represents the mapping result of the standard deviation function with parameter to the input ; represents the element-wise multiplication operator; represents the random noise sampled from the standard normal distribution; represents the standard normal distribution with mean 0 and covariance matrix being the identity matrix; Decoder network: Decodes the latent variable into the recovery path; State decoding layer: Maps to the initial state representation; Sequence generation layer: Use a Recurrent Neural Network (RNN) or a Transformer decoder to generate a sequence of recovery paths; Output layer: Generate specific recovery operations for each time step; In the scenario of preventing out - of - stock in the e - commerce supply chain, application examples of the graph - structured variational inference model include: Recovery from out - of - stock of seasonal goods: For goods highly affected by seasons (such as holiday supplies, clothing for specific seasons, etc.), when the out - of - stock risk is detected, the system first represents the current supply chain state, including inventory levels, supplier capacities, logistics status, etc., through a heterogeneous graph network. Then, the graph - structured variational inference model generates multiple possible recovery paths and considers the uncertainty of seasonal factors. Finally, a robust recovery strategy is generated, such as adjusting safety inventory in advance, adding alternative suppliers, optimizing logistics paths, etc.

[0047] Multi - region collaborative recovery: When there is an out - of - stock risk in a warehouse in a certain region, the system uses the graph - structured variational inference model to generate a recovery path involving the collaboration of multiple regional warehouses. The model considers factors such as inventory levels, logistics costs, and time constraints in each region and generates an optimal cross - regional inventory allocation plan without affecting the normal supply in other regions.

[0048] Response to supplier interruption: When the main supplier experiences production interruption, the system uses the graph - structured variational inference model. Based on the supply chain network structure and historical data, it generates a recovery strategy including operations such as activating alternative suppliers, adjusting logistics paths, and re - allocating production orders, minimizing the impact of supply interruption on product supply.

[0049] Step 2.3: Implementation of the multi - step inference algorithm; Based on Bayesian decision theory, implement a multi - step inference algorithm to evaluate the costs and effects of different recovery paths: Path sampling: Sample multiple possible recovery paths from the variational distribution ; ; where represents the variational distribution function with parameter ; represents the latent recovery path variable; represents the observed supply chain state; , , respectively represent the , , th sampled recovery paths; represents the total number of sampled paths; Path evaluation: Evaluate each sampled path and calculate its expected utility: where: represents the expected utility function; represents the th sampled recovery path; represents the expectation operator; represents the probability of transitioning to the next state after executing the joint action in the current state; ; represents the immediate reward obtained by executing the joint action in the current state ; represents the discount factor; represents the value function of the next state .

[0050] Bayesian risk analysis: Consider the decision-making risk and calculate the Bayesian risk of each path: where: represents the Bayesian risk function; represents the th sampled recovery path; represents the expectation operator; represents the probability distribution of transitioning to the next state after executing the joint action in the current state; ; represents the loss function; represents the next state of the system; represents the current state of the system; represents the joint action where represents the loss function of executing path in state .

[0051] Multi-objective trade-off: Comprehensively consider multiple objectives such as recovery cost, time, and risk, and calculate the comprehensive score: where: represents the comprehensive score function; represents the th sampled recovery path; , , , represent the weight coefficients of the utility term, risk term, cost term, and time term respectively; The expected utility function representing the path; The Bayesian risk function representing the path; The cost function representing the path; The recovery time function representing the path; In addition, through the Bayesian decision theory framework, the multi-step inference algorithm can make more robust decisions considering uncertainty, which is particularly important for out-of-stock recovery in the e-commerce supply chain.

[0052] Step 2.4, Generation of the optimal resilient recovery path; Based on the multi-step inference results, generate the optimal resilient recovery path: Path ranking: Sort all sampled paths according to the comprehensive score, and select the top K paths with the highest scores as candidates; Robustness analysis: Conduct robustness analysis on the candidate paths to test their performance under different perturbation conditions: Supplier delay perturbation: Simulate the situation of supplier delivery delay; Demand fluctuation perturbation: Simulate the situation of sudden changes in market demand; Logistics interruption perturbation: Simulate the situation of logistics channel interruption; Path optimization: Based on the robustness analysis results, fine-tune and optimize the candidate paths to enhance their ability to cope with uncertainty; Determination of the optimal path: Considering the score and robustness comprehensively, determine the final optimal resilient recovery path; It should be understood that the core innovation of this step is to model the supply chain recovery problem as a sequential decision problem on a graph and generate a highly robust recovery path through the variational inference method. Compared with traditional methods, the method provided in this application can better handle the out-of-stock recovery problem in a highly uncertain environment.

[0053] The output results include: The optimal recovery path, including specific operation steps, execution time, and expected effects; Risk assessment and countermeasure suggestions for path execution; Resource requirements and cost estimates for path implementation; Strategy optimization and uncertainty modeling module: Construct a deep probabilistic programming model for multiple recovery paths to model their uncertainty, and use the multi-agent Monte Carlo tree search algorithm to optimize the recovery strategy; This module uses the deep probabilistic programming and multi-agent Monte Carlo tree search methods to further optimize the generated recovery strategy and improve its robustness in a highly uncertain environment. The specific steps are as follows: Step 3.1, Construction of the deep probabilistic programming model; Construct a deep probabilistic programming model to accurately model the uncertainty of the recovery strategy as follows: Definition of probabilistic program: Construct a probabilistic program representing the recovery strategy which includes the following components: Set of random variables: representing the supply chain state variables; where is the set of random variables; , , are the th, th, th supply chain state random variables respectively; is the total number of random variables; Conditional dependence relationship representing the probabilistic dependence graph between variables; Conditional probability distribution: ; where represents the conditional probability distribution function; represents the th supply chain state random variable; represents the set of parent nodes of variable ; Neural network parameterization: Use neural network to parameterize the conditional probability distribution: where represents the conditional probability distribution function; the th supply chain state random variable; represents the set of parent nodes of variable ; represents the neural network function with parameter ; represents the parameters of the neural network; Stochastic variational inference: Apply the stochastic variational inference algorithm to estimate the posterior distribution: where represents the variational objective function; represents the variational parameters; represents the expectation operator; represents the variational distribution with parameter ; represents the path variable; represents the observed data; represents the logarithmic function; represents the probability distribution; Denotes the KL divergence, which is used to measure the difference between two probability distributions; Optimize the variational parameters by stochastic gradient ascent : Wherein, Denotes the variational parameter of the Denotes the variational parameter of the Denotes the learning rate of the Denotes taking the gradient with respect to the variational parameter λ; table Denotes the variational objective function; Among them, the gradient estimation uses the reparameterization trick: Wherein, Denotes taking the gradient with respect to the parameter ; Denotes the expectation operator; Denotes the variational distribution with parameter ; Denotes the path variable; Denotes the observed data; In variational inference, it denotes the evaluation function to be optimized, which is used to calculate the performance of the model under the given path variable; Denotes the number of samples; Denotes the sum operation from to , that is, adding the results of samples; Denotes the sampling index; Denotes the generation function with parameter ; Denotes the th random noise sample; Denotes the th random noise sample follows the noise distribution ; Denotes the th generated path sample; It should be noted that the deep probabilistic programming model can more accurately model the uncertainty in the supply chain recovery process by combining the advantages of deep learning and probabilistic programming, thereby generating more robust recovery strategies.

[0054] According to an embodiment of the present application, the specific structure of the deep probabilistic programming model includes: Probability variable definition layer: Define the key random variables of the supply chain state, including: Inventory level variable: Represents the inventory status of each node; Supply capacity variable: Represents the production capacity of the supplier; Logistics status variable: Represents the status of the logistics channel; Demand variable: Represents the market demand for the product; Risk variable: Represents the risk status of each link.

[0055] Probabilistic dependence network: Defines the conditional dependence relationships between variables and constructs a directed graph: Direct dependence: For example, the inventory level depends on upstream supply and downstream demand; Indirect dependence: For example, the substitution effect between different products; Temporal dependence: For example, the impact of the current state on the future state.

[0056] Neural network parameterization module: Uses different types of neural networks to parameterize different types of conditional probability distributions: Continuous variables: Uses a mixture of Gaussian networks; Discrete variables: Uses a categorical distribution network; Temporal variables: Uses a recurrent neural network.

[0057] In the specific application of anti-stockout decision-making in e-commerce supply chains, the deep probabilistic programming model can be used in the following scenarios: Modeling demand uncertainty: In cases such as promotional activities or new product launches, the demand for products is highly uncertain. The deep probabilistic programming model learns the demand patterns from historical data, constructs the probability distribution of the demand variable, and establishes dependence relationships with other variables in the supply chain to generate inventory strategies that can handle multiple demand scenarios.

[0058] Supplier risk assessment: For products that rely on multiple suppliers, the deep probabilistic programming model evaluates the risk levels of different supply options by modeling the probability distributions of variables such as supplier reliability and production capacity, and generates robust multi-supplier collaboration strategies.

[0059] Inventory-cost balance decision: In inventory management, excessive inventory increases costs, and insufficient inventory increases the risk of stockouts. The deep probabilistic programming model generates optimal inventory strategies that can balance costs and stockout risks under multiple possible scenarios by modeling the joint probability distribution of variables such as inventory, cost, and demand.

[0060] Optionally, the deep probabilistic programming model can also integrate external knowledge (such as seasonal rules, market expert experience, etc.) as prior information to further improve the accuracy and interpretability of the model. In some embodiments, the model can continuously update the probability distribution from new data through an active learning framework to achieve continuous optimization of the model.

[0061] Step 3.2, Implementation of the Multi-Agent Monte Carlo Tree Search Algorithm; Based on the deep probabilistic programming model, implement the multi-agent Monte Carlo tree search algorithm, and explore a better strategy space by sampling and evaluating multiple possible recovery paths: Definition of multi-agent: Model the supply chain recovery problem as a multi-agent system , where each agent represents a decision-making entity in the supply chain (such as the procurement department, inventory management department, logistics department, etc.); Among them, 、 、 respectively represent the 、 、 -th agent; represents the total number of agents; Definition of state space: The system state represents the current state of the supply chain network, including the state information of each node Joint action space: The joint action , where represents the action of agent ; Among them, represents the joint action; 、 、 respectively represent the 、 、 -th agent's action; represents the total number of agents; represents the -th agent's action; represents the joint action space; represents the -th agent's action space; represents the agent index; Monte Carlo tree search process: Selection: Starting from the root node, select a child node according to the UCT (Upper Confidence Bound for Trees) formula: Among them, represents the UCT function, which is used to select nodes in the tree search; represents the current state; represents the joint action; represents the valuation of the state-action pair; represents the exploration constant, which is used to balance exploration and exploitation; The number of visits to the state ; The number of visits to the state-action pair; Denotes the natural logarithm function; Expansion: Expand new nodes according to the transition probability; Simulation: Starting from the expanded nodes, use a random policy or a heuristic policy for simulation until a termination state is reached; Backpropagation: Propagate the simulation results back to all nodes in the tree to update the node statistics; Multi-agent cooperation method: Implement a cooperative decision-making method among agents to ensure that the decisions of each agent form a consistent global recovery strategy; Information sharing: Construct an information sharing protocol among agents to allow agents to share key state information; Cooperative decision-making: Construct a consensus-based decision-making algorithm to coordinate the action selection of each agent; Conflict resolution: Implement a conflict detection and resolution algorithm to handle possible decision conflicts among agents; In addition, according to the embodiments of the present application, the advantage of the multi-agent Monte Carlo tree search algorithm is that it can more comprehensively evaluate the possible decision space through the cooperative exploration of multiple agents and find the global optimal or near-optimal recovery strategy.

[0062] According to an embodiment of the present application, the detailed implementation of the multi-agent Monte Carlo tree search algorithm includes the following specific components: Tree structure design: Node representation: Each node contains state information , cumulative reward , visit count and a reference to the parent node; Edge representation: The edge represents an action , connecting the parent node and the child node, and storing action-related information; Tree depth control: Set a maximum depth limit to prevent the search space from exploding; Decision time allocation: Fixed iteration allocation: Allocate a fixed number of MCTS iterations to each decision point; Adaptive allocation: Dynamically allocate computing resources according to the state complexity and uncertainty; Real-time constraint: Optimize the search quality within a limited time; Multi-agent cooperation mechanism: Hierarchical decision-making framework: Divide the decision-making into two levels: the policy layer and the execution layer; Shared value network: All agents share the same value evaluation network but maintain their own search trees; Message passing mechanism: Define a standardized message format to enable information exchange between agents; Conflict resolution strategy: Use a priority-based conflict resolution solution or auction-based resource allocation; In the specific application of anti-stockout decision-making in the e-commerce supply chain, the multi-agent Monte Carlo tree search algorithm can be used in the following scenarios: Multi-warehouse collaborative replenishment: When a certain product faces the risk of out-of-stock simultaneously in multiple warehouses, the system needs to coordinate the replenishment behaviors of multiple warehouses.

[0063] By modeling each warehouse as an agent, multi-agent MCTS considers factors such as the inventory status, service area, and replenishment cost of each warehouse, and generates a collaborative replenishment plan to avoid blind competition or resource waste among warehouses. For example, the system may decide to prioritize replenishing the warehouses in the core area while adjusting the order allocation of other warehouses to achieve overall optimization.

[0064] Supplier-logistics-warehouse linkage decision-making: In the face of potential out-of-stock risks, it is necessary to coordinate multiple links such as supplier production, logistics transportation, and warehouse management. Multi-agent MCTS models supplier management, logistics scheduling, and warehouse management as different agents, and through collaborative search, generates a linkage recovery strategy considering multiple constraints such as production capacity, transportation timeliness, and inventory capacity. For example, before a holiday promotion, the system may generate a collaborative strategy including advance production, batch transportation, and temporary expansion of the warehouse.

[0065] Inventory balance of multi-category products: For multi-category products that are interrelated (such as main products and accessories), when a certain category faces the risk of out-of-stock, it is necessary to comprehensively consider the correlation between categories. Multi-agent MCTS models the inventory management of different categories as different agents, and through collaborative search, generates a recovery strategy that balances the demands of each category to avoid the out-of-stock of a certain category affecting the sales of other related categories.

[0066] Optionally, in some embodiments, multi-agent MCTS can be combined with deep reinforcement learning, using a policy network to guide the tree search process to improve search efficiency; using a value network to replace random simulation to improve evaluation accuracy. In other embodiments, a distributed implementation method can be adopted to allocate search tasks to multiple computing nodes to improve the ability to handle complex supply chain networks.

[0067] Step 3.3, Implementation of risk-aware policy optimization; Introduce a risk-aware policy optimization objective to balance recovery efficiency and risk exposure: Definition of the risk-aware objective function: Among them, represents the objective function for optimizing the risk perception strategy; represents the current strategy; represents the expectation operator; represents the current state; represents the state distribution under the current strategy; represents the joint action; represents the reward function; represents the reference strategy; is a trade-off parameter that controls the degree of risk aversion; represents the KL divergence, which measures the deviation between the current strategy and the reference strategy; Risk metric indicators: Define multiple risk metric indicators to comprehensively evaluate the risk characteristics of the strategy: Expected risk: ; Among them, represents the expectation operator; represents the risk loss function; represents the current state; represents the joint action; Conditional Value at Risk (CVaR): Among them, represents the conditional value at risk; represents the risk probability threshold; represents the risk loss function; represents the expectation operator; represents the value at risk; Maximum risk: Among them, it represents represents the maximum operator; represents the current state; represents the state distribution under the current strategy; represents the risk loss function; represents the current strategy; Policy optimization algorithm: Based on the risk perception objective function, apply the policy gradient method to optimize the policy parameters: Among them, represents the policy parameter at the th iteration; represents the policy parameter at the th iteration; represents the learning rate coefficient; represents the parameter Find the gradient; Represent the policy of the objective function The gradient estimate is: where, represents the gradient with respect to the policy parameters Find the gradient; Represent the policy of the objective function; represents the summation operator from to ; represents the number of samples; represents the parameterized policy function; represents the joint action of the th sample; represents the current state of the th sample; represents the sample index; represents the reward function of the state-action pair; represents the risk weight coefficient; represents the KL divergence; represents the reference policy; It should be understood that the risk-aware policy optimization method, by explicitly considering risk factors in the objective function, enables the generated policy to achieve a better balance between efficiency and risk, which is particularly important for out-of-stock recovery decisions in the e-commerce supply chain.

[0068] According to an embodiment of the present application, the specific implementation of risk-aware policy optimization includes the following core components: Risk measurement framework: Multi-dimensional risk representation: Decompose risks into multiple dimensions, such as time risk, cost risk, quality risk, etc.; Risk dependency modeling: Capture the interdependencies between different risk dimensions; Risk propagation model: Simulate the propagation process of risks in the supply chain network.

[0069] Reference policy construction: Expert policy extraction: Extract expert policies from historical successful cases as references; Conservative policy generation: Construct a conservative benchmark policy to ensure basic security; Hybrid reference policy: Dynamically combine multiple reference policies according to different situations.

[0070] Risk-aware optimizer: Constraint handling: Convert risk thresholds into optimization constraints; Multi-objective balance: Seek the best balance point among efficiency, cost, and risk; Adaptive weight adjustment: Dynamically adjust the risk weight according to environmental feedback 。

[0071] In the specific application of anti-stockout decision-making in the e-commerce supply chain, the strategy optimization of risk perception can be used in the following scenarios: Risk control of key commodities: For core or high-value commodities, the risk of stockout may bring serious reputation and economic losses. The strategy optimization of risk perception generates a more conservative but lower-risk recovery strategy by setting higher risk aversion parameters , such as maintaining a higher safety stock, establishing multiple alternative supply channels, etc., to ensure the supply stability of such commodities.

[0072] Strategy adjustment for seasonal commodities: For commodities with strong seasonality, both the risk of stockout and the risk of overstock need to be considered. The strategy optimization of risk perception dynamically adjusts the risk parameters, pays more attention to stockout risk control during the peak sales season, and pays more attention to overstock risk control during the off-season, so as to achieve the annual risk balance.

[0073] Risk management for new product launches: New product launches face a high risk of inaccurate demand forecasting. The strategy optimization of risk perception generates a supply strategy that balances innovation and safety by comparing the new product strategy with the conservative strategy of similar historical products, such as adopting a combination of small-batch trial sales and quick replenishment to control the risks brought by uncertainties.

[0074] In some embodiments, the strategy optimization of risk perception can also be combined with online learning technology to adjust the risk weight according to real-time market feedback and adapt to the changing market environment. Optionally, the system can also build a risk warning mechanism to automatically trigger an audit process or switch to a more conservative strategy when the strategy risk exceeds a preset threshold.

[0075] Step 3.4, Generate optimized recovery strategy; Generate an optimized recovery strategy by integrating the results of deep probabilistic programming and multi-agent Monte Carlo tree search: Strategy integration: Integrate the generated elastic recovery paths with the strategies explored by multi-agent Monte Carlo tree search to combine the advantages of both; Risk adjustment: Adjust the strategy according to the optimization results of risk perception to ensure that while maintaining efficient recovery, the risk exposure is controlled; Strategy verification: Verify the performance of the optimized strategy in various simulation scenarios to ensure that it can perform well in all situations; Generate the final optimized recovery strategy based on the verification results, including a detailed implementation plan and emergency measures.

[0076] Therefore, through the method combining deep probabilistic programming and multi-agent Monte Carlo tree search in this step, a more robust and efficient recovery strategy can be generated while fully considering uncertainty and risk factors.

[0077] The output results include: An optimized recovery strategy with higher robustness and lower risk; Uncertainty estimation and risk assessment during the strategy execution process; Prediction of the strategy performance under various possible scenarios.

[0078] Data fusion and strategy execution module: Based on the optimized recovery strategy, it realizes the real-time dynamic fusion of multi-source data, adaptively adjusts the data source weights, and transforms the abstract strategy into specific executable supply chain adjustment instructions; Specifically, it includes the following steps: Step 4.1, Real-time dynamic fusion of multi-source data; Realize the real-time dynamic fusion of multi-source data in the e-commerce supply chain, and adaptively adjust the data source weights and fusion strategies for different out-of-stock risk scenarios: Data source access: Build a unified data access layer to support the access of various data sources: Transaction data sources: Sales orders, refund orders, evaluation data, etc.; Inventory data sources: Commodity inventory levels, inventory locations, inventory status, etc.; Supplier data sources: Supplier production capacity, supply cycle, quality rating, etc.; Logistics data sources: Transportation status, logistics timeliness, transport capacity, etc.; External data sources: Market trends, competitor activities, seasonal factors, etc.

[0079] Data quality assessment: Conduct real-time quality assessment on the accessed multi-source data: Integrity assessment: Detect data missing situations and make reasonable supplements; Timeliness assessment: Evaluate the time delay of the data and give priority to using the latest data; Consistency assessment: Detect and resolve conflicts between different data sources; Accuracy assessment: Verify and evaluate the data accuracy through historical data; Adaptive weight allocation: According to the out-of-stock risk scenario and data quality, adaptively adjust the data source weights: Among them, represents the weight of the data source ; represents the data source quality score; represents the data source Relevance score for the current risk scenario; Indicates From To Sum all items; Indicates the summation variable, an integer from To ; Indicates the total number of data sources; Indicates the data source Quality score; Indicates the data source Relevance score for the current risk scenario; Indicates the index of the data source; Multimodal data fusion, based on the assigned weights, realizes the fusion of multimodal data: Feature-level fusion: Weightedly fuse the features of different data sources; Decision-level fusion: Make decisions based on different data sources respectively, and then perform weighted integration; Model-level fusion: Train different models with different data sources, and then perform model integration.

[0080] It should be understood that the adaptive data fusion method provided by this application can dynamically adjust the importance of data sources according to different out-of-stock risk scenarios, so as to provide more accurate decision support information.

[0081] Step 4.2, Construction of out-of-stock recovery execution engine; Construct an out-of-stock recovery execution engine to transform the abstract strategy into specific executable supply chain adjustment instructions: Instruction parsing: Parse the generated optimized recovery strategy into a specific instruction set: Procurement instructions: including order quantity, supplier selection, delivery time, etc.; Inventory adjustment instructions: including inventory transfer, safety inventory adjustment, etc.; Logistics instructions: including logistics path optimization, transportation mode selection, etc.; Pricing instructions: including promotion strategies, purchase limit strategies, etc.

[0082] Execution plan generation, based on the parsed instruction set, generate a detailed execution plan: Task decomposition: Decompose complex instructions into atomic-level tasks; Task sorting: Sort tasks according to dependencies and priorities; Resource allocation: Allocate the required resources for each task; Time planning: Develop a schedule for task execution.

[0083] Constraint checking, check the generated execution plan for constraints to ensure its feasibility: Resource constraints: Check whether the available resource limits are exceeded; Time constraints: Check whether the time requirements are met; Business rule constraints: Check whether the business rules and policies are complied with; Conflict checking: Check whether there are conflicts between instructions.

[0084] Instruction execution interface, constructing standardized interfaces with various supply chain systems to achieve automated instruction execution: Procurement system interface: Automatically generate and send purchase orders; Inventory management system interface: Automatically adjust inventory allocation and transfer; Logistics system interface: Automatically optimize and adjust logistics routes; Pricing system interface: Automatically adjust product prices and promotion strategies.

[0085] In addition, the out-of-stock recovery execution engine can transform abstract recovery strategies into specific executable operation instructions, achieving seamless connection from decision-making to execution.

[0086] Closed-loop monitoring and optimization module: Construct a closed-loop feedback algorithm for the executed adjustment instructions, continuously monitor the execution effect and optimize the strategy; Specifically, it includes the following steps: Step 5.1, Implementation of the closed-loop feedback algorithm; Construct a closed-loop feedback algorithm, continuously monitor the execution effect and optimize the strategy: Execution effect monitoring: Real-time monitor the execution effect of the recovery strategy: Key indicator monitoring: Out-of-stock rate, inventory level, customer satisfaction, etc.; Deviation analysis: Compare the deviation between the actual execution effect and the expected effect; Anomaly detection: Real-time detect abnormal situations during the execution process.

[0087] Real-time adjustment, based on the monitoring results, real-time adjust the execution plan: Parameter fine-tuning: Fine-tune the strategy parameters according to the execution effect; Re-planning: Re-plan the execution plan when major deviations occur; Emergency handling: Activate the emergency plan when anomalies are detected.

[0088] Strategy evaluation and optimization: Regularly evaluate the strategy effect and optimize it: Effect evaluation: Comprehensively evaluate the execution effect of the strategy; Experience accumulation: Add the execution data and results to the historical database; Strategy optimization: Optimize the strategy generation model based on the accumulated execution experience.

[0089] It should be noted that the closed-loop feedback algorithm can continuously improve the performance and adaptability of the out-of-stock prevention decision-making system by continuously monitoring the execution effect and making real-time adjustments.

[0090] Step 5.2, System integration and collaboration; Integrate each component into a unified system to achieve collaborative work: Component integration: Integrate each component in each module into a unified system architecture; Interface standardization: Define standardized interfaces between components to ensure smooth transfer of data and control flow.

[0091] Collaborative workflow: Define the workflow of the system to ensure collaborative work of each component: Risk monitoring process: Continuously monitor out-of-stock risks; Strategy generation process: Generate recovery strategies when risks are detected; Policy execution process: Transform the policy into actions and execute them; Feedback optimization process: Optimize the policy based on the execution effect.

[0092] Therefore, through system integration and collaboration, the method provided by this application can form a complete closed-loop out-of-stock prevention decision-making system, providing comprehensive out-of-stock risk management capabilities from risk prediction, recovery strategy generation to execution monitoring and continuous optimization.

[0093] Through the above data, it can be verified that the out-of-stock prevention decision-making method for e-commerce supply chains based on multi-source data fusion provided by this application has significant technical effects in practical applications and can effectively improve the resilience and operational efficiency of e-commerce platform supply chains.

[0094] The above describes the embodiments of the present invention, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. An anti-out-of-stock decision-making system for e-commerce supply chains based on multi-source data fusion, characterized in that, Including: Heterogeneous graph network construction and risk prediction module: Construct a heterogeneous graph network representing e-commerce supply chain relationships based on multi-source data, and use a time-sensitive graph attention network to predict the propagation path of out-of-stock risks; Elastic recovery path generation module: Based on the constructed heterogeneous graph network, model the supply chain recovery as a sequential decision-making problem on the graph, and generate multiple elastic recovery paths through variational inference methods; Policy optimization and uncertainty modeling module: Construct a deep probabilistic programming model for multiple recovery paths to model their uncertainties, and use a multi-agent Monte Carlo tree search algorithm to optimize the recovery strategy; Data fusion and policy execution module: Based on the optimized recovery strategy, achieve real-time dynamic fusion of multi-source data, adaptively adjust the data source weights, and transform the abstract policy into specific executable supply chain adjustment instructions; Closed-loop monitoring and optimization module: Construct a closed-loop feedback algorithm for the executed adjustment instructions, continuously monitor the execution effect and optimize the strategy.

2. The e-commerce supply chain out-of-stock prevention decision-making system based on multi-source data fusion according to claim 1, characterized in that, The steps of constructing the heterogeneous graph network include: Collect multi-source data in the e-commerce supply chain, including sales data, inventory data, supplier data, logistics data, and market trend data; Preprocess the collected multi-source data, including data cleaning, missing value handling, outlier detection, and data standardization; Construct a heterogeneous graph network representing e-commerce supply chain relationships, where the nodes include commodity nodes, warehouse nodes, supplier nodes, and logistics nodes, and the edges include supply relationships, inventory relationships, and logistics relationships.

3. The e-commerce supply chain out-of-stock prevention decision-making system based on multi-source data fusion according to claim 1, wherein The time-sensitive graph attention network measures the association strength between different nodes by calculating attention coefficients, where the temporal characteristics of the edges are considered, including time encoding, periodic pattern encoding, and temporal relationship encoding, to capture the characteristics of supply chain relationships changing over time.

4. The e-commerce supply chain out-of-stock prevention decision-making system based on multi-source data fusion according to claim 1, characterized in that, The variational inference method generates elastic recovery paths through the following steps: Define the posterior distribution of the recovery path, and introduce a variational distribution as an approximate substitute for the posterior distribution; Optimize the parameters by maximizing the variational lower bound; Integrate the graph network structure into the variational inference model, and use a graph neural network to process the structural information of the supply chain network; Based on Bayesian decision theory, implement a multi-step inference algorithm to evaluate the costs and effects of different recovery paths.

5. The decision-making system for preventing out-of-stock in an e-commerce supply chain based on multi-source data fusion according to claim 1, wherein Define a set of random variables representing the supply chain state and the conditional dependence relationships between variables; Parameterize the conditional probability distribution using a neural network; Apply the stochastic variational inference algorithm to estimate the posterior distribution.

6. The e-commerce supply chain out-of-stock prevention decision-making system based on multi-source data fusion according to claim 1, characterized in that, The multi-agent Monte Carlo tree search algorithm includes: Model the supply chain recovery problem as a multi-agent system, where each agent represents a decision-making entity in the supply chain; Define the system state and the joint action space; Execute the Monte Carlo tree search process, including four steps: selection, expansion, simulation, and backpropagation; Implement a collaborative decision-making method among agents to ensure that the decisions of each agent form a consistent global recovery strategy.

7. The e-commerce supply chain out-of-stock prevention decision-making system based on multi-source data fusion according to claim 1, characterized in that, Introduce a risk-aware policy optimization objective function that includes a reward term and a risk aversion term. Among them, the reward term measures the expected benefit of the policy, and the risk aversion term controls the risk level by calculating the difference between the current policy and the reference policy.

8. The decision-making system for preventing out-of-stock in an e-commerce supply chain based on multi-source data fusion according to claim 1, wherein The steps of real-time dynamic fusion of multi-source data include: Construct a unified data access layer to support the access of various data sources; Perform real-time quality assessment on the accessed multi-source data, including integrity, timeliness, consistency, and accuracy assessment; Adaptive adjustment of data source weights according to out-of-stock risk scenarios and data quality; Based on the assigned weights, achieve feature-level fusion, decision-level fusion, and model-level fusion.

9. The decision-making system for preventing out-of-stock in an e-commerce supply chain based on multi-source data fusion according to claim 1, characterized in that The steps of converting the abstract strategy into specific executable supply chain adjustment instructions include: Parse the optimized recovery strategy into a specific set of execution instructions, including procurement instructions, inventory adjustment instructions, logistics instructions, and pricing instructions; Based on the parsed instruction set, generate a detailed execution plan, including task decomposition, task sequencing, resource allocation, and time planning; Conduct constraint checking on the generated execution plan to ensure its feasibility; Build a standardized interface with each system in the supply chain to achieve automated execution of instructions.

10. A storage medium, characterized in that, Including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement an e-commerce supply chain out-of-stock prevention decision-making system according to any one of claims 1-9.

Images