Supply chain esg risk transmission simulation method, system, device and medium

Abstract

The application relates to a supply chain ESG risk transmission simulation method, system, device and medium. The method comprises the following steps: acquiring text event data and a dynamic supply chain network graph of a supply chain; based on a preset text encoder, extracting semantic features of the text event data to obtain an event feature vector; based on the event feature vector, analyzing the maintenance probability of the dynamic supply relationship between each supply chain entity in the dynamic supply chain network graph, and obtaining supply relationship update data based on the maintenance probability; based on the supply relationship update data, updating the dynamic supply chain network graph to obtain an updated dynamic supply chain network graph; based on an event type component, performing risk propagation simulation on the updated dynamic supply chain network graph to obtain a risk propagation simulation result, and based on the risk propagation simulation result and the supply relationship update data, obtaining a risk early warning report. The method can improve the timeliness and authenticity of supply chain risk early warning.

Description

Technical Field

[0001] This invention belongs to the field of supply chain risk management, and in particular relates to a method, system, equipment and medium for simulating the transmission of ESG risks in the supply chain. Background Technology

[0002] With the development of supply chain risk management technologies, complex network analysis methods such as dynamic graph neural networks have emerged. These technologies can effectively characterize the temporal changes in the supply chain network topology and node attributes, leading to the current mainstream dynamic supply chain risk simulation method based on "time snapshots." This method treats supply chain entities as nodes and business relationships as edges, assessing risk by integrating multi-source data.

[0003] Existing processing methods typically involve sampling the supply chain network periodically or at fixed time intervals to construct a series of static graph snapshots. Specifically, the system collects data from enterprise resource planning systems, third-party databases, and news and public opinion sources according to a preset cycle (such as daily or weekly), and regenerates the nodes and edges of the entire network. The attributes of each node (such as risk score) are updated based on the data within that cycle. Subsequently, graph embedding or propagation algorithms are used to simulate the near-uniform diffusion process of risk from one snapshot to the next, thereby assessing the static distribution and evolution trend of risk in the network.

[0004] However, current time-snapshot methods have fundamental limitations. Their biggest problem lies in the fact that the pace of risk perception and transmission simulation is confined by a pre-set time window, making it unable to respond to discrete, sudden, and asynchronous ESG risk events in the real world. This results in significant perception delays after critical events occur, and the model cannot characterize the complex dynamics of risk transmission, such as heterogeneous delays, nonlinear accumulation, and threshold breaches caused by different event types and network paths. Consequently, the timeliness and accuracy of supply chain risk warnings are insufficient. Summary of the Invention

[0005] Therefore, it is necessary to provide a supply chain ESG risk transmission simulation method, system, equipment, and medium that can improve the timeliness and accuracy of supply chain risk early warning in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a method for simulating the transmission of ESG risks in a supply chain, including:

[0007] Acquire text event data and dynamic supply chain network diagrams from the supply chain; the dynamic supply chain network diagram is used to characterize the current risk status of each supply chain entity and the dynamic supply relationships between supply chain entities.

[0008] Based on a pre-defined text encoder, semantic features of text event data are extracted to obtain an event feature vector; the event feature vector includes an event subject component and an event type component.

[0009] Based on event feature vectors, the maintenance probability of dynamic supply relationships among supply chain entities in a dynamic supply chain network diagram is analyzed, and supply relationship update data is obtained based on the maintenance probability.

[0010] Based on the updated supply relationship data, the dynamic supply chain network diagram is updated to obtain the updated dynamic supply chain network diagram.

[0011] Based on event type components, risk propagation simulation is performed on the updated dynamic supply chain network diagram to obtain risk propagation simulation results. Based on the risk propagation simulation results and supply relationship update data, a risk warning report is obtained.

[0012] Furthermore, the dynamic supply chain network graph includes graph nodes, graph node feature vectors, directed edges, and directed edge attribute vectors. Graph nodes represent supply chain entities, graph node feature vectors represent the risk status of supply chain entities, connected edges represent the supply relationships between supply chain entities, and connected edge attribute vectors represent the supply information of the supply relationships. Based on event feature vectors, the maintenance probability of dynamic supply relationships between supply chain entities in the dynamic supply chain network graph is analyzed, and supply relationship update data is obtained based on the maintenance probability, including:

[0013] Define the supply chain entity corresponding to the main component of the event in the event feature vector as the event source supply chain entity; and define the graph node corresponding to the event source supply chain entity in the dynamic supply chain network diagram as the source graph node.

[0014] Based on the dynamic supply chain network graph, with the source graph node as the center, the neighbor nodes are traversed, the neighbor nodes are defined as the association graph nodes, and the directed connection edges between the association graph nodes and the source graph nodes are defined as the influence connection edges corresponding to the association graph nodes.

[0015] For each node in the graph, the event type component, the feature vector of the graph node, and the attribute vector of the connecting edge corresponding to the node are input into a preset survival probability prediction model to obtain the maintenance probability of the connecting edge corresponding to the node. The survival probability prediction model is constructed based on a multilayer perceptron.

[0016] Select the directed edges whose maintenance probability is less than the preset lower limit threshold to form the deletion edge set; and determine the deletion edge set as the supply relationship update data.

[0017] Furthermore, based on the updated supply relationship data, the dynamic supply chain network diagram is updated to obtain the updated dynamic supply chain network diagram, including:

[0018] Based on the updated supply relationship data, the directed connection edges and their corresponding connection attribute vectors included in the deleted edge set of the dynamic supply chain network graph are removed to obtain the adjusted supply chain network graph.

[0019] For each graph node in the adjusted supply chain network diagram, extract the graph nodes connected by directed edges to form the neighborhood graph node set of the graph node;

[0020] For each graph node in the supply chain network diagram, based on the graph node feature vector, directed connection edge, and connection edge attribute vector, the attention coefficient between the graph node and each graph node in the neighborhood graph node set is calculated to obtain the attention coefficient set of the graph node.

[0021] For each graph node in the supply chain network diagram, based on the attention coefficient, the graph node feature vectors of each graph node in the neighborhood graph node set are weighted and aggregated to obtain the updated graph node feature vector.

[0022] Based on the updated feature vectors of each graph node, the feature vectors of the graph nodes in the adjusted supply chain network graph are updated to obtain the updated dynamic supply chain network graph.

[0023] Furthermore, the graph node feature vectors include node risk scores and node resilience scores. Based on the event type component, risk propagation simulation is performed on the updated dynamic supply chain network graph to obtain the risk propagation simulation results, including:

[0024] For each graph node in the updated dynamic supply chain network graph, starting from the graph node, traverse the neighboring nodes to obtain the set of upstream neighboring graph nodes of the graph node; and extract the connection attribute vectors corresponding to the directed connection edges between the neighboring nodes and the graph node to form the set of directed connection edge attributes of the graph node.

[0025] For each graph node in the updated dynamic supply chain network graph, the risk propagation delay parameter set of the graph node is obtained based on the event type component, the directed connection edge attribute set, and the preset risk delay derivation model.

[0026] For each graph node in the updated dynamic supply chain network graph, the node risk score in the graph node feature vector is defined as the initial risk prediction value. Based on the upstream neighbor graph node set, the risk propagation delay parameter set, and the updated graph node feature vector, the S1 to S3 steps are repeated until the preset maximum number of iterations is reached to obtain the risk prediction value sequence of the graph node.

[0027] The risk propagation simulation results are composed of the risk prediction value sequence of each graph node;

[0028] Steps S1 to S3 are as follows:

[0029] S1, based on the risk propagation delay parameter set and graph node feature vector, aggregates the risk prediction values ​​of graph nodes in the upstream neighbor graph node set of the graph node to obtain the comprehensive risk input value of the graph node;

[0030] S2, based on the node risk score and node resilience score of the graph nodes, the dynamic risk threshold is calculated;

[0031] S3 compares the comprehensive risk input value with the dynamic risk threshold to obtain the comparison result, and updates the risk prediction value of the graph node based on the comparison result through a preset state update function to obtain the updated risk prediction value of the graph node.

[0032] Furthermore, for each graph node in the updated dynamic supply chain network graph, based on event type components, directed connection edge attribute sets, and a pre-defined risk delay derivation model, the risk propagation delay parameter set for the graph node is obtained, including:

[0033] For each edge attribute vector in the directed edge attribute set, the edge attribute vector and event type component are input into a preset risk propagation prediction model to obtain the risk propagation delay parameter corresponding to the edge attribute vector. The risk propagation delay parameter is used to characterize the degree of risk propagation delay from the neighbor node at one end of the directed edge corresponding to the edge attribute vector to the graph node.

[0034] Based on the risk propagation delay parameters corresponding to the attribute vectors of each connected edge, the risk propagation delay parameter set of the graph nodes is obtained;

[0035] The risk transmission prediction model was obtained through the following method:

[0036] Obtain a pre-defined historical risk propagation dataset; the historical risk propagation dataset includes historical event type components, historical supply information data, and historical risk transmission delay time parameters of historical risk propagation records; and use the historical event type components and historical supply information data as input feature vectors, and the historical risk transmission delay labels corresponding to the input feature vectors as training labels, divide the historical risk propagation dataset according to a pre-defined dataset division ratio to obtain a delay model training set and a delay model test set;

[0037] An initial delay prediction model is constructed based on the gradient boosting tree algorithm; and based on the delay model training set, the input feature vector is input into the initial delay prediction model to obtain the delay prediction value corresponding to the input feature vector.

[0038] The model prediction loss is calculated based on the delayed prediction value corresponding to the input feature vector and the historical risk transmission delay label corresponding to the input feature vector.

[0039] Based on the model prediction loss, the initial delay prediction model is iteratively trained until the preset iteration termination condition is met, thus obtaining the preliminary delay prediction model; and the preliminary delay prediction model is tested through the model test set until the preliminary delay prediction model meets the preset accuracy requirements, thus obtaining the trained risk transmission prediction model.

[0040] Secondly, this application also provides a supply chain ESG risk transmission simulation system, including:

[0041] The data acquisition module is used to acquire text event data and dynamic supply chain network diagrams of the supply chain; the dynamic supply chain network diagram is used to represent the current risk status of each supply chain entity and the dynamic supply relationship between supply chain entities.

[0042] The text feature extraction module is used to extract semantic features from text event data based on a preset text encoder to obtain an event feature vector; the event feature vector includes an event subject component and an event type component;

[0043] The preliminary relationship analysis module is used to analyze the maintenance probability of dynamic supply relationships between supply chain entities in a dynamic supply chain network diagram based on event feature vectors, and to obtain updated supply relationship data based on the maintenance probability.

[0044] The dynamic graph update module is used to update the dynamic supply chain network diagram based on supply relationship data, and obtain the updated dynamic supply chain network diagram.

[0045] The current risk analysis module is used to simulate risk propagation on the updated dynamic supply chain network diagram based on event type components, obtain risk propagation simulation results, and generate risk warning reports based on the risk propagation simulation results and supply relationship update data.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the supply chain ESG risk transmission simulation methods described in the first aspect of this application.

[0047] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the supply chain ESG risk transmission simulation methods described in the first aspect of this application.

[0048] The aforementioned supply chain ESG risk transmission simulation method, system, equipment, and medium acquire text event data and a dynamic supply chain network diagram. The dynamic supply chain network diagram represents the current risk status of each supply chain entity and the dynamic supply relationships between them. Based on a pre-set text encoder, semantic features of the text event data are extracted to obtain event feature vectors. These feature vectors include event subject components and event type components. Based on the event feature vectors, the maintenance probability of the dynamic supply relationships between each supply chain entity in the dynamic supply chain network diagram is analyzed, and supply relationship update data is obtained based on the maintenance probability. The dynamic supply chain network diagram is updated based on the updated supply relationship data, resulting in an updated dynamic supply chain network diagram. Based on the event type components, risk propagation simulation is performed on the updated dynamic supply chain network diagram to obtain the risk propagation simulation results. Based on the risk propagation simulation results and the updated supply relationship data, a risk warning report is generated. This approach changes the traditional simulation mode based on fixed-period snapshots, achieving a leap from periodic checkups to event-triggered real-time diagnosis, thus improving the timeliness and accuracy of supply chain ESG risk warnings. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart illustrating a supply chain ESG risk transmission simulation method provided in one embodiment of this application;

[0051] Figure 2 A schematic diagram of a supply chain ESG risk transmission simulation system provided in one embodiment of this application;

[0052] Figure 3 This is a schematic diagram of the structure of a computer device for simulating the transmission of supply chain ESG risks, provided as an embodiment of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] In one embodiment, such as Figure 1As shown, a supply chain ESG risk transmission simulation is provided. This embodiment illustrates the application of this method to a simulation terminal. It is understood that this method can also be applied to a server, and further to a system including both a simulation terminal and a server, and is implemented through the interaction between the simulation terminal and the server. In this embodiment, the method includes the following steps S101-S105, wherein:

[0055] S101, Obtain text event data and dynamic supply chain network diagram of the supply chain; the dynamic supply chain network diagram is used to characterize the current risk status of each supply chain entity and the dynamic supply relationship between supply chain entities.

[0056] Specifically, the simulated terminal acquires text event data and a dynamic supply chain network diagram of the supply chain. Text event data refers to ESG (Environmental, Social, and Governance) event reports related to supply chain entities, described in natural language text form and obtained in real-time from external data sources. The data format can be a complete news article, a summary of a regulatory announcement, or a social media post, such as: "It is reported that supplier A has been heavily fined and ordered to rectify its violations by the local environmental protection department within a specified period." Text event data can be crawled or pushed from news aggregation platforms, regulatory agency websites, or corporate public opinion monitoring systems through a pre-defined API interface. The dynamic supply chain network diagram is a directed graph structure used to represent the real-time state of the entire supply chain network; its mathematical form can be... ,in It is a set of graph nodes, each graph node It uniquely represents a specific supply chain entity (such as "supplier A" or "manufacturer B"). It is a set of directed connecting edges, each edge Characterization from supply chain entities To supply chain entities The existing supply, cooperation, or control relationships indicate the main direction of material or capital transmission. It is a node feature matrix, where any element Graph nodes The feature vectors of the graph nodes are used to represent the corresponding supply chain entities. The current, comprehensive risk status can specifically include environmental risk scores, social risk scores, and governance risk scores, as well as the mapping relationships between graph nodes and the names of the supply chain entities they represent. The dynamic supply chain network graph is constructed based on historical business data such as enterprise ERP and supplier directories during initialization, and is updated and persistently stored in each subsequent simulation, loaded as an initial network state snapshot for that simulation.

[0057] S102, based on a preset text encoder, extracts semantic features from text event data to obtain an event feature vector; the event feature vector includes an event subject component and an event type component.

[0058] Specifically, the preset text encoder is a pre-trained deep neural network model, whose architecture can be based on Transformer models (such as BERT or RoBERTa). The preset text encoder can be obtained through self-supervised pre-training on a large amount of general corpus and / or domain-specific corpus (such as business or financial news). During processing, the simulated terminal inputs the currently received text event data into the preset text encoder, which outputs a high-dimensional semantic feature vector. Subsequently, the simulated terminal identifies and determines the main components of the event through a lightweight information extraction module (which can be a named entity recognition and classification head based on a pre-trained model). The main components of the event This represents the names of supply chain entities directly associated with the text event data (such as "Supplier A"). Secondly, the event type components are obtained through an event classification module (which can be a classification head based on a pre-trained model). The value range of this event type component is a preset set of ESG risk types, which can be in the form of... The main components of the event and event type components The feature vector of an event can be represented as a structured tuple. .

[0059] S103, based on event feature vectors, analyzes the maintenance probability of dynamic supply relationships between supply chain entities in the dynamic supply chain network diagram, and obtains supply relationship update data based on the maintenance probability.

[0060] Specifically, the analog terminal uses event feature vectors and dynamic supply chain network diagram This quantifies the impact of the currently acquired event on the stability of existing supply relationships, resulting in updated supply relationship data. This updated data is used to indicate the dynamic supply chain network under the influence of this event. Which supply relationships (i.e., directed edges) are severely threatened in terms of stability and should be considered potentially broken? The maintenance probability is a scalar value between 0 and 1, used to quantify the likelihood that a specific supply relationship will be maintained after an event occurs; a lower probability indicates a higher risk of breakage. Supply relationship update data can specifically take the form of a set of directed edge identifiers.

[0061] S104. Update the dynamic supply chain network diagram based on the updated supply relationship data to obtain the updated dynamic supply chain network diagram.

[0062] Specifically, the simulated terminal updates data based on supply relationships and executes a dynamic supply chain network diagram. The real-time adjustments yielded an updated dynamic supply chain network graph. Specifically, the updated dynamic supply chain network graph reflects the new state of the network topology after the current event occurs. The adjustment process may include deleting corresponding directed edges in the dynamic supply chain network graph based on updated supply relationship data, and adjusting the graph node feature vectors of each graph node based on the adjusted edge relationships. The graph is updated to obtain the updated feature vectors of the graph nodes. The updated dynamic supply chain network graph can be used as the dynamic supply chain network graph obtained by S101 in the next event simulation analysis.

[0063] S105, based on event type components, performs risk propagation simulation on the updated dynamic supply chain network diagram, obtains risk propagation simulation results, and obtains a risk warning report based on the risk propagation simulation results and supply relationship update data.

[0064] Specifically, the analog terminal uses the event type component in the event feature vector. The system uses an updated dynamic supply chain network diagram to perform dynamic simulations of risk transmission, predicting the trajectory of risk over a future period. This yields risk propagation simulation results, which are then integrated with updated supply relationship data to generate a risk warning report. Specifically, the risk propagation simulation results characterize the risk level of each node in the updated dynamic supply chain network diagram within a pre-defined future time window. The simulation process relies on event type components. Because of the event type components The speed of risk propagation and the pattern of its impact differ. Risk propagation simulation results can take the form of a time series dataset. For each graph node (supply chain entity) in the network... Each corresponds to a risk state value sequence, which records the nodes. From the start of the simulation (Typically, the time of the event) to multiple discrete future moments. Risk prediction value . It is a scalar used to quantize the node at... The overall risk level at any given moment or its risk score across the various ESG sub-dimensions.

[0065] This embodiment provides a supply chain ESG risk transmission simulation method. It receives real-time text event data, automatically parses its semantics, and locates the source of influence, obtaining an event feature vector including event subject and event type components. Then, based on the event semantics, it dynamically assesses the maintenance probability of each supply relationship in the supply chain network, obtaining updated supply relationship data. This updated data is then used to update the dynamic supply chain network diagram, resulting in an updated dynamic supply chain network diagram. Subsequently, on the updated dynamic supply chain network diagram, a nonlinear time-series risk propagation simulation is performed, combining event type-driven heterogeneous propagation delay and node dynamic threshold mechanisms, generating accurate risk warning reports. This method changes the traditional simulation mode based on fixed-period snapshots, achieving a leap from periodic checks to event-triggered real-time diagnosis, improving the timeliness and accuracy of supply chain ESG risk warnings.

[0066] In one embodiment, the dynamic supply chain network graph includes graph nodes, graph node feature vectors, directed edges, and directed edge attribute vectors. Graph nodes represent supply chain entities, graph node feature vectors represent the risk status of supply chain entities, connected edges represent supply relationships between supply chain entities, and connected edge attribute vectors represent supply information of supply relationships. Based on event feature vectors, the maintenance probability of dynamic supply relationships between supply chain entities in the dynamic supply chain network graph is analyzed, and supply relationship update data is obtained based on the maintenance probability, including:

[0067] S201, define the supply chain entity corresponding to the main component of the event in the event feature vector as the event source supply chain entity; and define the graph node corresponding to the event source supply chain entity in the dynamic supply chain network graph as the source graph node.

[0068] Specifically, dynamic supply chain network diagram It also includes the edge attribute matrix any of the elements Directed connection edge The connection edge attribute vector is used to represent the corresponding supply relationship. The static attribute information includes, but is not limited to, cooperation duration, geographical distance, mode of transport identifier, and risk transmission efficiency value. The risk transmission efficiency value can be obtained by linearly transforming the cooperation duration, geographical distance, and mode of transport identifier using a preset weight vector, and is used to characterize the efficiency of risk transmission. Its value range is [insert range here]. The simulated terminal obtains information from the event feature vector. Extract the main component of the event The main component of the event The represented supply chain entity is defined as the source supply chain entity of the event, and is based on the main components of that event. In dynamic supply chain network diagram Query within, based on each graph node Graph node feature vectors The mapping relationship between the names of supply chain entities represented by the nodes in the middle graph and the nodes in the graph matches the main components of the current event. In dynamic supply chain network diagram The corresponding unique graph node and the graph nodes Defined as a source graph node.

[0069] S202, based on the dynamic supply chain network graph, takes the source graph node as the center, traverses the neighbor nodes, defines the neighbor nodes as association graph nodes, and defines the directed connection edge between the association graph node and the source graph node as the influence connection edge corresponding to the association graph node.

[0070] Specifically, the simulated terminal uses source graph nodes. Centered on the dynamic supply chain network diagram The directed edges in the array. Specifically, the traversal is performed in two directions: first, searching for all edges connected by... Directed connecting edges originating from (i.e.) These edges represent supply relationships originating from the source entity; secondly, it involves finding all edges that represent supply relationships originating from the source entity. Directed connecting edges with endpoints (i.e.) These edges represent supply relationships pointing to the source entity. For each directed connection edge found during traversal, the graph node at the other end of the edge... This is defined as a node in the relational graph. It also connects to the source graph node. Related graph nodes This specific directed connection edge Defined as a node in the association graph The corresponding impact on the connecting edges.

[0071] S203, for each node in the associated graph, the event type component, the feature vector of the graph node, and the connection attribute vector of the influencing connection edge corresponding to the associated graph node are input into the preset survival probability prediction model to obtain the maintenance probability of the influencing connection edge corresponding to the associated graph node; the survival probability prediction model is constructed based on a multilayer perceptron.

[0072] Specifically, for each given node in the association graph The simulated terminal will divide the event type components. Nodes in the relational graph Graph node feature vectors Nodes in the relational graph Corresponding influence on the connecting edge Connecting edge attribute vector The input is fed into a preset survival probability prediction model to obtain the influencing connection edges. Maintenance probability A lower probability indicates a more fragile supply relationship under the current event's impact. The preset survival probability prediction model is a machine learning model used to evaluate relationship stability, built upon a multilayer perceptron. This model comprises an input layer, at least one hidden layer, and an output layer. The number of neurons in the input layer equals the sum of the event type component embedding dimension, the graph node feature vector dimension, and the connection edge attribute vector dimension. Event type components... First, it is converted into a dense vector through an embedding layer. The input features are the concatenated vectors: The hidden layers use non-linear activation functions (such as ReLU) for feature transformation. The output layer uses a sigmoid activation function to map the final output value to the (0,1) interval; this value is the maintenance probability. The pre-defined survival probability prediction model is obtained through supervised training using historical data. The training data includes a large number of historical samples of "event type - associated node state - edge attribute," along with corresponding true result labels (i.e., whether the edge broke after the event). The training objective is to improve the model's prediction accuracy. Get as close to reality as possible.

[0073] S204, select the influential directed connection edges whose maintenance probability is less than the preset lower limit threshold of maintenance probability, form a set of deleted edges; and determine the set of deleted edges as the supply relationship update data.

[0074] Specifically, the preset lower bound threshold for the maintenance probability is a predefined constant used to delineate the boundary between maintaining and potentially breaking the relationship, and can be denoted as: Its setting can be determined based on historical data analysis (such as observing that most fractures occur when the probability is below a certain value) or domain expert experience; the default setting is 0.2. The simulated terminal sequentially traverses each influential connection edge evaluated by S203. For each affected connecting edge Maintain its probability With the lower limit threshold of the maintenance probability Compare. If Then the affected connecting edge Unique identifier information (which can be derived from source graph nodes) and associated graph nodes Identifiers Each edge is added to a temporary set. After traversing all edges, the final temporary set is the set of edges to be deleted. This set of deleted edges is then used as the supply relationship update data.

[0075] This embodiment provides a supply chain ESG risk transmission simulation method. It accurately locates the source supply chain entities from event characteristics and intelligently traverses their direct network neighbors to identify potentially impacted supply relationships, obtaining association graph nodes and corresponding impact edges. Then, using a pre-trained, multi-layer perceptron-based survival probability prediction model, it comprehensively considers multi-dimensional information such as event type, stakeholder status, and relationship attributes to perform a refined and quantitative assessment of the maintenance probability of each target relationship, obtaining the maintenance probability of the impact edges corresponding to the association graph nodes. Finally, low-probability relationships are filtered out based on preset thresholds to obtain updated supply relationship data. This method automates and data-driven inference from discrete events to network topology change decisions, demonstrating the invention of event-driven topology adaptation. It enables the risk simulation system to respond to sudden events in real time and predict potential network structure changes, enhancing the foresight and realism of risk warnings.

[0076] In one embodiment, the dynamic supply chain network diagram is updated based on supply relationship update data to obtain an updated dynamic supply chain network diagram, including:

[0077] S301. Based on the updated supply relationship data, delete the directed connecting edges and their corresponding connecting edge attribute vectors from the deleted edge set in the dynamic supply chain network graph to obtain the adjusted supply chain network graph.

[0078] Specifically, the simulated terminal traverses the impact connection edges included in the supply relationship update data. Unique identifier information for each affected connection edge Unique Identification Information The simulated terminal first starts from the set of directed connected edges. Remove the edge from the middle. Secondly, from the edge attribute matrix Remove the connection edge attribute vector corresponding to the edge in the middle. After all edges to be deleted have been processed, a new graph data structure is obtained, denoted as the Adjusted Supply Chain Network Graph. Among them, the node set and node feature matrix The set of directed connecting edges remains unchanged. ( (This refers to the set of edges to be deleted), edge attribute matrix. The connection edge attribute vector of the directed connection edge corresponding to the deleted edge was also removed.

[0079] S302, for each graph node in the adjusted supply chain network diagram, extract the graph nodes connected by directed edges to form a set of neighboring graph nodes of the graph node.

[0080] Specifically, regarding adjusting the supply chain network diagram Each graph node The simulated terminal performs a partial traversal operation. Specifically, the simulated terminal uses graph nodes... For query center, in the adjusted set of directed edges In the middle, find all that satisfy the condition that the starting point or ending point is Directed connecting edges (i.e.) or edge ). Connect these to the graph nodes Directly connected graph nodes The set of nodes is called the graph nodes. The set of nodes in the neighborhood graph can be denoted as... .

[0081] S303, for each graph node in the adjusted supply chain network graph, based on the graph node feature vector, directed connection edge and connection edge attribute vector, calculate the attention coefficient between the graph node and each graph node in the neighborhood graph node set, and obtain the attention coefficient set of the graph node.

[0082] Specifically, the simulated terminal uses the core operations of Graph Attention Network (GAT) to quantify the differences in importance of different graph nodes to the central graph node within the neighborhood, thus obtaining a set of attention coefficients for the graph nodes. This is particularly relevant for adjusting the supply chain network graph. Each graph node First, pair each node , Through a learnable weight vector and a shared linear transformation matrix Calculate an unnormalized attention score The specific calculation formula is as follows: .in, Graph nodes Graph node feature vectors and Graph nodes Graph node feature vectors; It is to transfer the feature vectors of graph nodes from Dimensional projection to A shared weight matrix of dimensions; This represents a vector concatenation operation; Directed connection edge The connection edge attribute vector; It is a learnable attention vector; LeakyReLU is a non-linear activation function whose negative slope is typically set to a small constant (e.g., 0.2). Matrix sum vector These are the model parameters, which can be obtained by training a graph neural network on historical data. Then, for easier comparison and weighting, the attention scores of all neighbors of a node need to be standardized. Normalization is performed using the softmax function to obtain the final attention coefficients. : . It is a scalar between 0 and 1, and satisfies For graph nodes All corresponding graph nodes Calculated The set of nodes constitutes the nodes of the graph. The set of attention coefficients.

[0083] S304. For each graph node in the supply chain network graph, based on the attention coefficient, the graph node feature vectors of each graph node in the neighborhood graph node set are weighted and aggregated to obtain the updated graph node feature vector.

[0084] Specifically, regarding adjusting the supply chain network diagram Each graph node The simulated terminal uses the attention coefficient set calculated by S303 to fuse the graph node feature vectors of neighboring graph nodes, generating an updated graph node feature vector for that graph node. The specific calculation formula is: ,in It is the shared weight matrix described in S303. It is the current graph node The feature vectors of the graph nodes. It can be calculated based on S303. It is a preset non-linear activation function, such as ELU (Exponential Linear Unit) or ReLU.

[0085] S305, based on the updated graph node feature vectors of each graph node, update the graph node feature vectors in the adjusted supply chain network graph to obtain the updated dynamic supply chain network graph.

[0086] Specifically, the simulation terminal will update the graph node feature vectors of each graph node calculated by S304. Adjusting the supply chain network diagram Middle node feature matrix Replace the corresponding elements, that is, for all ,make The final updated dynamic supply chain network diagram is obtained. .

[0087] This embodiment provides a supply chain ESG risk transmission simulation method. First, it prunes the network topology based on supply relationship update data to obtain an adjusted supply chain network graph. Then, it reconstructs the local neighbor environment of each graph node under the new topology. Subsequently, it dynamically calculates the attention coefficients between a node and its neighbors using a graph attention mechanism based on edge attributes. Finally, it weights and aggregates the features of neighboring nodes based on these coefficients and synchronously updates the graph node feature vectors of all graph nodes in the entire network, resulting in an updated dynamic supply chain network graph. This method achieves efficient, non-uniform, and adaptive rebalancing and fusion of risk state information along the remaining connections after event-driven topology changes. This allows the new state of a node to simultaneously absorb the direct impact of the event (through changes in the state of the source node) and the indirect transmission impact of the network, laying an accurate initial state foundation for subsequent precise time-series risk propagation simulations.

[0088] In one embodiment, the graph node feature vector includes node risk score and node resilience score. Based on the event type component, risk propagation simulation is performed on the updated dynamic supply chain network graph to obtain the risk propagation simulation results, including:

[0089] S401. For each graph node in the updated dynamic supply chain network graph, traverse the neighboring nodes with the graph node as the endpoint to obtain the upstream neighbor graph node set of the graph node; and extract the connection edge attribute vectors corresponding to the directed connection edges between the neighboring nodes and the graph node to form the directed connection edge attribute set of the graph node.

[0090] Specifically, regarding the updated dynamic supply chain network diagram Each graph node in The simulated terminal performs a directed traversal operation. (Based on graph nodes...) As the endpoint, in the adjusted set of directed edges In the middle, find all nodes in the graph. A directed connection edge with endpoint is of the form: The edges. All graph nodes with such starting points. The collected data forms a set of graph nodes. The set of upstream neighbor nodes in the graph can be denoted as... At the same time, for each such directed connection edge... Extract the corresponding connection edge attribute vector. The set of all such connection edge attribute vectors is called a graph node. The set of attributes of directed connected edges can be denoted as... .

[0091] S402, for each graph node in the updated dynamic supply chain network graph, based on the event type component, the set of directed connection edge attributes and the preset risk delay derivation model, the risk propagation delay parameter set of the graph node is obtained.

[0092] Specifically, regarding the updated dynamic supply chain network diagram Each graph node in The simulated terminal will divide the event type components. The nodes in this diagram The set of attributes of directed connection edges Each connecting edge attribute vector The input is fed into the preset risk delay derivation model to obtain the connection edge attribute vector. Corresponding risk propagation delay parameters Directed connection edge attribute set The risk propagation delay parameters corresponding to the attribute vectors of each connecting edge in the graph form the graph nodes. The risk propagation delay parameter set can be denoted as: The preset risk delay derivation model is a predictive model that maps event types and edge attributes to time delays. It can predict the speed of risk propagation based on different types of risks and different business relationships; for example, environmental penalty risks propagate slowly, while financial scandal risks propagate quickly. Risk propagation delay parameters. Characterizes the components of the current event type. Under risk events, the risk originates from the nodes in the graph. The corresponding supply chain entities along the edge Transmission to graph nodes Approximate time required (e.g., in days).

[0093] S403, for each graph node in the updated dynamic supply chain network graph, the node risk score in the graph node feature vector is defined as the initial risk prediction value. Based on the upstream neighbor graph node set, the risk propagation delay parameter set and the updated graph node feature vector, the S1 to S3 steps are repeated until the preset maximum number of iterations is reached to obtain the risk prediction value sequence of the graph node.

[0094] Specifically, any graph node Graph node feature vectors It also includes node risk scoring. and node resilience score Node risk scoring This is a scalar value between 0 and 1, used to characterize the overall risk level of the supply chain entity at the start of the simulation; a higher score indicates higher risk. Node Resilience Score It is a scalar value between 0 and 1, used to characterize the inherent ability of a supply chain entity to absorb, buffer, or withstand external risk shocks. A higher score indicates greater resilience and less susceptibility to being breached by risk events. Node Risk Score and node resilience score This can be obtained by weighting or extracting features from the environmental risk score, social risk score, and governance risk score in the feature vector of the graph nodes. This is applicable to the updated dynamic supply chain network graph. Each graph node in The simulated terminal uses its graph node feature vectors Node risk scoring The initial risk forecast value can be denoted as: The simulation process proceeds in discrete time steps, with a fixed time step size set. (For example, 1 day, whose unit of length is consistent with the unit of length of the risk propagation delay parameter). Preset "number of large cycles". The time span of the simulation was determined. The loop starts from... Beginning, until Among them, the first Cycle corresponding to simulation time In each cycle In the graph, for nodes Perform three sub-steps, S1, S2, and S3, to calculate the graph nodes. New risk forecasts for the current round After each round of iteration, the simulation terminal records the results of that round. When all is completed After the round loop is completed, the graph nodes are obtained. A length of sequence This sequence is the graph node. The risk prediction value sequence.

[0095] S404, based on the risk prediction value sequence of each graph node, constitutes the risk propagation simulation result.

[0096] Specifically, the simulation terminal integrates the risk prediction value sequences of each graph node to obtain the risk propagation simulation results.

[0097] Steps S1 to S3 are as follows:

[0098] S1, based on the risk propagation delay parameter set and graph node feature vector, aggregates the risk prediction values ​​of graph nodes in the upstream neighbor graph node set of the graph node to obtain the comprehensive risk input value of the graph node.

[0099] Specifically, the simulated terminal calculates the risk prediction value received by a graph node from its upstream neighbor graph nodes in the current round. The specific inputs include the current graph node... Risk propagation delay parameter set Current cycle round Upstream Neighbor Graph Node Set and upstream neighbor graph node set Risk prediction values ​​of each node in the graph in historical rounds , Specifically, for each graph node The analog terminal first determines the delay parameter. Calculate the historical round index of its impact. Assume the time step is... Then delay The corresponding round offset is (Round down). When calculating the current cycle iteration. Time graph nodes When inputting the comprehensive risk value, it is necessary to select graph nodes. In the Risk prediction value of the wheel .like Then take its initial risk prediction value. Secondly, the attribute vectors of the corresponding connecting edges are also extracted. The risk transmission efficiency value in the graph. Finally, the graph nodes... In the Comprehensive risk input value of the wheel The result is obtained through weighted summation: .

[0100] S2, based on the node risk score and node resilience score of the graph nodes, calculates the dynamic risk threshold.

[0101] Specifically, the simulation terminal calculates the risk tolerance threshold of the graph nodes in the current round. The specific inputs include graph nodes. Graph node feature vectors Mid-node risk scoring and node resilience score Dynamic risk threshold It is a simulation round A scalar that may change, representing a graph node. The maximum risk input that can be tolerated in the current state without triggering a risk transition. The specific calculation formula is as follows: .in, It is a fundamental threshold constant shared by all graph nodes; It is a positive coefficient used to represent the nodal resilience score. Each additional unit contributes to the threshold improvement; It is a node resilience score; It is a preset risk fatigue coefficient, which is a positive number; This indicates the time from the start of the simulation to the previous round ( Up to ), graph nodes The sum of all received comprehensive risk input values ​​represents the cumulative level of risk exposure, which lowers the node's tolerance threshold (i.e., the "fatigue effect"). Therefore, the threshold... Determined by the base value, inherent resilience, and accumulated risk fatigue, it is dynamic and simulates the real-world situation where the node's ability to withstand continuous risk pressure gradually decreases.

[0102] S3 compares the comprehensive risk input value with the dynamic risk threshold to obtain the comparison result, and updates the risk prediction value of the graph node based on the comparison result through a preset state update function to obtain the updated risk prediction value of the graph node.

[0103] Specifically, the simulation terminal compares the current round's overall risk input value with the dynamic risk threshold to obtain a comparison result. The specific input is the current round's overall risk input value. and dynamic risk threshold and graph nodes Previous round of risk forecast values The preset state update function can take the following form: The preset state update function is used to simulate two evolution modes of risk: among which, It is a decay factor used to simulate how graph nodes digest or mitigate existing risks; It is the risk absorption coefficient, which represents the proportion by which a node transforms external risk input into an increase in its own state. This is the burst inertia coefficient, used to control the speed at which a state transitions to its upper limit. If This indicates that the current external input risk has not yet exceeded the node's tolerance limit, and the risk accumulates or decays within the node in a relatively gradual manner. If This indicates that the external risk shock is too high, and the node risk state may undergo a nonlinear leap. The preset maximum risk value can be set according to the actual work situation.

[0104] This embodiment provides a supply chain ESG risk transmission simulation method. It locates the upstream risk sources and path attributes for each node and uses an event-type-driven delay model to predict the heterogeneous delay of risk transmission along each path. Then, it constructs a cyclical simulation framework starting from the initial state of each node and progressing in discrete time steps. Within each time step, it accurately calculates the risk input aggregated from each node's delayed upstream neighbors and combines it with its dynamically decaying tolerance threshold. A nonlinear state update function determines whether the risk level gradually accumulates or explodes, ultimately obtaining the complete risk trajectory of all nodes within a future time window. This process meticulously depicts the complex dynamic behavior of risk transmission in the supply chain network over time and topology, including transmission delay, threshold effects, and nonlinear jumps. It achieves a leap from static snapshot analysis to dynamic process extrapolation, significantly improving the realism and predictability of risk warnings.

[0105] In one embodiment, for each graph node in the updated dynamic supply chain network graph, based on event type components, directed connection edge attribute sets, and a preset risk delay derivation model, a risk propagation delay parameter set for the graph node is obtained, including:

[0106] S501, for each connection edge attribute vector in the directed connection edge attribute set, input the connection edge attribute vector and event type component into the preset risk propagation prediction model to obtain the risk propagation delay parameter corresponding to the connection edge attribute vector; the risk propagation delay parameter is used to characterize the degree of risk propagation delay from the neighbor node at one end of the directed connection edge corresponding to the connection edge attribute vector to the graph node.

[0107] Specifically, the simulated terminal interacts with graph nodes. The set of attributes of directed connection edges Each connection edge attribute vector in Perform a prediction operation. The simulation terminal will connect the edge attribute vectors. and the event type component in the event feature vector The input is fed into a pre-defined risk transmission prediction model to obtain the connection edge attribute vector. Corresponding risk propagation delay parameters The preset risk transmission prediction model is a pre-trained machine learning model whose function is to learn and establish a mapping relationship from "event type-side attribute" features to "risk transmission time". Risk propagation delay parameter. The current event type components were quantified. Under the influence of risk events, the risk originates from upstream neighboring nodes. Depart, through this route The specific supply relationships (directed connections) represented are propagated to downstream graph nodes. The required time length. For example, ... This could indicate the risk of an "environmental penalty" incident, and it would take approximately 5.2 days to obtain approval from the supplier. Transmitted to manufacturers .

[0108] S502, based on the risk propagation delay parameters corresponding to the attribute vectors of each connected edge, obtain the risk propagation delay parameter set of the graph node.

[0109] Specifically, the simulated terminal completes the process of checking all upstream neighbor edges (i.e. After predicting the delay parameters of all directed edges (corresponding to the attribute vectors of all connected edges in the simulation), the results are summarized. Specifically, the simulation terminal performs step S501 for each upstream neighbor node. Calculated single delay parameter They are organized according to their corresponding upstream neighbor node identifiers to form a set: This set is the graph node. The risk propagation delay parameter set.

[0110] The risk transmission prediction model was obtained through the following method:

[0111] S5011, Obtain the preset historical risk propagation dataset; The historical risk propagation dataset includes historical event type components, historical supply information data, and historical risk transmission delay time parameters of historical risk propagation records; Using the historical event type components and historical supply information data as input feature vectors, and the historical risk transmission delay labels corresponding to the input feature vectors as training labels, the historical risk propagation dataset is divided according to the preset dataset division ratio to obtain the delay model training set and the delay model test set.

[0112] Specifically, the simulated terminal acquires a pre-set historical risk propagation dataset. This dataset is prepared in advance during the model training phase and contains a large number of records collected, cleaned, and labeled from historical cases. Each historical risk propagation record contains three core pieces of information: 1) Historical event type classification. 1) Indicates the type of ESG risk event that has occurred historically, and its value range is the same as the value range of the event type component described in S101; 2) Historical supply information data The data format is consistent with the connection edge attribute vector in the dynamic supply chain network diagram, used to characterize the attributes of the specific supply chain relationships involved in the historical event (such as the proportion of procurement amount at that time, geographical distance, and historical risk transmission efficiency value, etc.); 3) Historical risk transmission delay time parameter This is a scalar quantity, obtained through post-hoc analysis, representing the actual time it takes for risk to propagate from the source entity of the event along the aforementioned specific relationship to another entity (e.g., the number of days between the issuance of a penalty notice and a decrease in customer orders). The simulation terminal will... and Combined into an input feature vector At the same time, the corresponding The training labels for this input feature vector are then used. Subsequently, the simulated terminal randomly divides the entire historical risk propagation dataset into two parts according to a preset dataset partitioning ratio: this results in the delay model training set. and delay model test set For example, the preset dataset split ratio can be set according to the actual work, and the default setting is 8:2 for the delayed model training set and the delayed model test set.

[0113] S5012 constructs an initial delay prediction model based on the gradient boosting tree algorithm; and based on the delay model training set, the input feature vector is input into the initial delay prediction model to obtain the delay prediction value corresponding to the input feature vector.

[0114] Specifically, the simulated terminal uses the Gradient Boosting Decision Tree (GBDT) algorithm as the learning framework to initialize the latency prediction model. GBDT is an additive model, and the final model is a collection of multiple decision trees: ,in It is the first A decision tree, It is its weight. During the training process, the simulated terminal is based on Perform iterations. In each iteration... The simulated terminal will input the feature vector of the training set. Input the currently constructed model combination In this process, the current input feature vector is obtained. Corresponding delayed prediction value .

[0115] S5013, based on the delayed prediction value corresponding to the input feature vector and the historical risk transmission delay label corresponding to the input feature vector, calculate the model prediction loss.

[0116] Specifically, for the training set One of the input feature vectors The simulated terminal has obtained its predicted delay value using the current model. And obtain its corresponding historical risk transmission delay label. The simulated terminal uses a loss function. To measure the difference between these two values. For regression tasks, the loss function... It can be the mean squared error (MSE). The loss for a single input feature vector is... So, the model prediction loss over the entire training set or a single batch. That is, the average loss of all samples: ,in This is the number of samples used in the calculation. This is the loss value. This reflects the average prediction error of the current model across the entire training data and is the target driving iterative optimization of the model (the goal is to make...). minimize).

[0117] S5014. Based on the model prediction loss, the initial delay prediction model is iteratively trained until the preset iteration termination condition is met, thus obtaining the preliminary delay prediction model; and the preliminary delay prediction model is tested through the model test set until the preliminary delay prediction model meets the preset accuracy requirements, thus obtaining the trained risk transmission prediction model.

[0118] Specifically, the simulated terminal predicts the loss using a calculated model. To guide this process, the model is iteratively optimized. Within the gradient boosting tree framework, optimization is achieved by continuously adding new decision trees. This is achieved by fitting the residual predicted by the current model to each new tree. This process continues until a preset iteration termination condition is met, resulting in a preliminary delayed prediction model. This preliminary delayed prediction model is then tested on a model test set until it meets a preset accuracy requirement, resulting in a trained risk propagation prediction model. For example, the preset iteration termination condition could be reaching a preset maximum number of iterations (i.e., the maximum number of trees). The settings can also be customized according to the actual work requirements. Optionally, the preset accuracy requirement can be an accuracy rate of 90%, or it can be adjusted according to the actual work requirements.

[0119] This embodiment provides a supply chain ESG risk transmission simulation method. It constructs a specialized dataset for training a delay prediction model by extracting event types, relationship attributes, and observed transmission delay data from historical risk event cases. Then, it employs a powerful ensemble learning algorithm, gradient boosting trees, to iteratively build multiple decision trees. Each tree focuses on learning the residuals of the previous stage's predictions, gradually approximating complex nonlinear mapping relationships. During training, the mean squared error loss function quantifies prediction bias and drives model optimization until the iteration termination condition is met, resulting in a preliminary model. Finally, a rigorous generalization performance evaluation of the model is conducted using an independent test set to ensure its accuracy meets the required standards before it is solidified as a usable prediction tool. This complete, data-driven model building process ensures the scientific validity and reliability of risk propagation delay parameters, providing key time dynamic parameters for simulating the heterogeneity and nonlinear transmission of risks along the supply chain network. This is the core technological foundation for improving the realism of the entire simulation system.

[0120] The aforementioned supply chain ESG risk transmission simulation method acquires text event data and a dynamic supply chain network diagram. The dynamic supply chain network diagram represents the current risk status of each supply chain entity and the dynamic supply relationships between them. Based on a pre-set text encoder, semantic features of the text event data are extracted to obtain event feature vectors. These feature vectors include event subject components and event type components. Based on the event feature vectors, the maintenance probability of the dynamic supply relationships between each supply chain entity in the dynamic supply chain network diagram is analyzed, and supply relationship update data is obtained based on the maintenance probability. The dynamic supply chain network diagram is updated based on the updated supply relationship data, resulting in an updated dynamic supply chain network diagram. Based on the event type components, risk propagation simulation is performed on the updated dynamic supply chain network diagram to obtain the risk propagation simulation results. Finally, based on the risk propagation simulation results and the updated supply relationship data, a risk warning report is generated. This method changes the traditional simulation model based on fixed-period snapshots, achieving a leap from periodic checkups to event-triggered real-time diagnosis, thus improving the timeliness and accuracy of supply chain ESG risk warnings.

[0121] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0122] Based on the same inventive concept, this application also provides a supply chain ESG risk transmission simulation system for implementing the supply chain ESG risk transmission simulation method described above. The solution provided by this system is similar to the implementation scheme described in the above method; therefore, the specific limitations of one or more supply chain ESG risk transmission simulation system embodiments provided below can be found in the limitations of the supply chain ESG risk transmission simulation method described above, and will not be repeated here.

[0123] In one exemplary embodiment, such as Figure 2 As shown, a supply chain ESG risk transmission simulation system 200 is provided, including:

[0124] The data acquisition module 201 is used to acquire text event data and dynamic supply chain network diagrams of the supply chain; the dynamic supply chain network diagram is used to represent the current risk status of each supply chain entity and the dynamic supply relationship between supply chain entities.

[0125] The text feature extraction module 202 is used to extract semantic features of text event data based on a preset text encoder to obtain an event feature vector; the event feature vector includes an event subject component and an event type component;

[0126] The preliminary relationship analysis module 203 is used to analyze the maintenance probability of dynamic supply relationships between supply chain entities in the dynamic supply chain network diagram based on event feature vectors, and to obtain supply relationship update data based on the maintenance probability.

[0127] The dynamic graph update module 204 is used to update the dynamic supply chain network diagram based on the supply relationship update data, and obtain the updated dynamic supply chain network diagram.

[0128] The current risk analysis module 205 is used to simulate risk propagation on the updated dynamic supply chain network diagram based on event type components, obtain risk propagation simulation results, and generate risk warning reports based on the risk propagation simulation results and supply relationship update data.

[0129] Furthermore, the dynamic supply chain network graph includes graph nodes, graph node feature vectors, directed edges, and directed edge attribute vectors. Graph nodes represent supply chain entities, graph node feature vectors represent the risk status of supply chain entities, connected edges represent the supply relationships between supply chain entities, and connected edge attribute vectors represent the supply information of the supply relationships. The preliminary relationship analysis module can also be used for:

[0130] Define the supply chain entity corresponding to the main component of the event in the event feature vector as the event source supply chain entity; and define the graph node corresponding to the event source supply chain entity in the dynamic supply chain network diagram as the source graph node.

[0131] Based on the dynamic supply chain network graph, with the source graph node as the center, the neighbor nodes are traversed, the neighbor nodes are defined as the association graph nodes, and the directed connection edges between the association graph nodes and the source graph nodes are defined as the influence connection edges corresponding to the association graph nodes.

[0132] For each node in the graph, the event type component, the feature vector of the graph node, and the attribute vector of the connecting edge corresponding to the node are input into a preset survival probability prediction model to obtain the maintenance probability of the connecting edge corresponding to the node. The survival probability prediction model is constructed based on a multilayer perceptron.

[0133] Select the directed edges whose maintenance probability is less than the preset lower limit threshold to form the deletion edge set; and determine the deletion edge set as the supply relationship update data.

[0134] Furthermore, the dynamic graph update module can also be used for:

[0135] Based on the updated supply relationship data, the directed connection edges and their corresponding connection attribute vectors included in the deleted edge set of the dynamic supply chain network graph are removed to obtain the adjusted supply chain network graph.

[0136] For each graph node in the adjusted supply chain network diagram, extract the graph nodes connected by directed edges to form the neighborhood graph node set of the graph node;

[0137] For each graph node in the supply chain network diagram, based on the graph node feature vector, directed connection edge, and connection edge attribute vector, the attention coefficient between the graph node and each graph node in the neighborhood graph node set is calculated to obtain the attention coefficient set of the graph node.

[0138] For each graph node in the supply chain network diagram, based on the attention coefficient, the graph node feature vectors of each graph node in the neighborhood graph node set are weighted and aggregated to obtain the updated graph node feature vector.

[0139] Based on the updated feature vectors of each graph node, the feature vectors of the graph nodes in the adjusted supply chain network graph are updated to obtain the updated dynamic supply chain network graph.

[0140] Furthermore, the graph node feature vectors include node risk scores and node resilience scores. The current risk analysis module includes:

[0141] The upstream attribute determination unit is used to determine the upstream neighbor graph node set for each graph node in the updated dynamic supply chain network graph, starting from the graph node as the endpoint and traversing the neighbor nodes; and extracting the connection edge attribute vectors corresponding to the directed connection edges between the neighbor nodes and the graph node to form the directed connection edge attribute set of the graph node.

[0142] The delay parameter determination unit is used to obtain the risk propagation delay parameter set of each graph node in the updated dynamic supply chain network graph based on the event type component, the directed connection edge attribute set, and the preset risk delay derivation model.

[0143] The loop analysis unit is used to define the node risk score in the feature vector of each graph node in the updated dynamic supply chain network graph as the initial risk prediction value. Based on the upstream neighbor graph node set, the risk propagation delay parameter set and the updated graph node feature vector, the loop S1 to S3 steps are repeated until the preset maximum number of loops is reached to obtain the risk prediction value sequence of the graph node.

[0144] The simulation results are composed of units used to form risk propagation simulation results based on the risk prediction value sequence of each graph node;

[0145] Steps S1 to S3 are as follows:

[0146] S1, based on the risk propagation delay parameter set and graph node feature vector, aggregates the risk prediction values ​​of graph nodes in the upstream neighbor graph node set of the graph node to obtain the comprehensive risk input value of the graph node;

[0147] S2, based on the node risk score and node resilience score of the graph nodes, the dynamic risk threshold is calculated;

[0148] S3 compares the comprehensive risk input value with the dynamic risk threshold to obtain the comparison result, and updates the risk prediction value of the graph node based on the comparison result through a preset state update function to obtain the updated risk prediction value of the graph node.

[0149] Furthermore, the delay parameter determination unit can also be used for:

[0150] For each edge attribute vector in the directed edge attribute set, the edge attribute vector and event type component are input into a preset risk propagation prediction model to obtain the risk propagation delay parameter corresponding to the edge attribute vector. The risk propagation delay parameter is used to characterize the degree of risk propagation delay from the neighbor node at one end of the directed edge corresponding to the edge attribute vector to the graph node.

[0151] Based on the risk propagation delay parameters corresponding to the attribute vectors of each connected edge, the risk propagation delay parameter set of the graph nodes is obtained;

[0152] The risk transmission prediction model was obtained through the following method:

[0153] Obtain a pre-defined historical risk propagation dataset; the historical risk propagation dataset includes historical event type components, historical supply information data, and historical risk transmission delay time parameters of historical risk propagation records; and use the historical event type components and historical supply information data as input feature vectors, and the historical risk transmission delay labels corresponding to the input feature vectors as training labels, divide the historical risk propagation dataset according to a pre-defined dataset division ratio to obtain a delay model training set and a delay model test set;

[0154] An initial delay prediction model is constructed based on the gradient boosting tree algorithm; and based on the delay model training set, the input feature vector is input into the initial delay prediction model to obtain the delay prediction value corresponding to the input feature vector.

[0155] The model prediction loss is calculated based on the delayed prediction value corresponding to the input feature vector and the historical risk transmission delay label corresponding to the input feature vector.

[0156] Based on the model prediction loss, the initial delay prediction model is iteratively trained until the preset iteration termination condition is met, thus obtaining the preliminary delay prediction model; and the preliminary delay prediction model is tested through the model test set until the preliminary delay prediction model meets the preset accuracy requirements, thus obtaining the trained risk transmission prediction model.

[0157] In one embodiment, such as Figure 3 A computer device is provided, comprising:

[0158] At least one processor 301, and a memory 302 communicatively connected to at least one of the processors 301: the memory stores application code that can be executed by at least one of the processors to enable at least one of the processors to perform the supply chain ESG risk transmission simulation method as described above.

[0159] Computer equipment may also include: sensor 303.

[0160] The processor 301, memory 302 and sensor 303 can be connected via a bus or other means, with the bus being an example in the figure.

[0161] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0162] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0163] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for simulating the transmission of ESG risks in a supply chain, characterized in that, The method includes: Acquire text event data and dynamic supply chain network diagrams of the supply chain; the dynamic supply chain network diagrams are used to characterize the current risk status of each supply chain entity and the dynamic supply relationships between the supply chain entities. Based on a preset text encoder, semantic features of the text event data are extracted to obtain an event feature vector; the event feature vector includes an event subject component and an event type component; Based on the event feature vector, the maintenance probability of the dynamic supply relationship between each supply chain entity in the dynamic supply chain network diagram is analyzed, and supply relationship update data is obtained based on the maintenance probability. Based on the updated supply relationship data, the dynamic supply chain network diagram is updated to obtain the updated dynamic supply chain network diagram. Based on the event type components, a risk propagation simulation is performed on the updated dynamic supply chain network diagram to obtain the risk propagation simulation results. Based on the risk propagation simulation results and the supply relationship update data, a risk warning report is obtained.

2. The method according to claim 1, characterized in that, The dynamic supply chain network graph includes graph nodes, graph node feature vectors of the graph nodes, directed edges, and edge attribute vectors of the directed edges. The graph nodes represent the supply chain entities, the graph node feature vectors represent the risk status of the supply chain entities, the edges represent the supply relationships between the supply chain entities, and the edge attribute vectors represent the supply information of the supply relationships. Based on the event feature vectors, the maintenance probability of the dynamic supply relationships between the supply chain entities in the dynamic supply chain network graph is analyzed, and supply relationship update data is obtained based on the maintenance probability, including: The supply chain entity corresponding to the main component of the event in the event feature vector is defined as the event source supply chain entity; and the graph node corresponding to the event source supply chain entity in the dynamic supply chain network diagram is defined as the source graph node. Based on the dynamic supply chain network diagram, with the source graph node as the center, the neighbor nodes are traversed, the neighbor nodes are defined as association graph nodes, and the directed connection edge between the association graph node and the source graph node is defined as the influence connection edge corresponding to the association graph node. For each of the aforementioned graph nodes, the event type component, the graph node feature vector, and the connection edge attribute vector of the influencing connection edge corresponding to the graph node are input into a preset survival probability prediction model to obtain the maintenance probability of the influencing connection edge corresponding to the graph node; the survival probability prediction model is constructed based on a multilayer perceptron. Select the influential directed connection edges whose maintenance probability is less than the preset lower limit threshold of maintenance probability to form a set of deleted edges; and determine the set of deleted edges as the supply relationship update data.

3. The method according to claim 2, characterized in that, The step of updating the dynamic supply chain network diagram based on the supply relationship update data to obtain the updated dynamic supply chain network diagram includes: Based on the updated supply relationship data, the directed connection edges and the corresponding connection attribute vectors of the deleted edges in the dynamic supply chain network graph are deleted to obtain the adjusted supply chain network graph. For each graph node in the adjusted supply chain network graph, extract the graph nodes connected to the graph node through the directed connection edge, and form a neighborhood graph node set of the graph node; For each graph node in the adjusted supply chain network graph, based on the graph node feature vector, the directed connection edge, and the connection edge attribute vector, the attention coefficient between the graph node and each graph node in the neighborhood graph node set is calculated to obtain the attention coefficient set of the graph node; For each graph node in the adjusted supply chain network graph, based on the attention coefficient, the graph node feature vectors of each graph node in the neighborhood graph node set are weighted and aggregated to obtain the updated graph node feature vector of the graph node. Based on the updated graph node feature vectors of each graph node, the graph node feature vectors in the adjusted supply chain network graph are updated to obtain the updated dynamic supply chain network graph.

4. The method according to claim 3, characterized in that, The graph node feature vector includes node risk score and node resilience score. Based on the event type component, the updated dynamic supply chain network graph is subjected to risk propagation simulation to obtain the risk propagation simulation results, including: For each graph node in the updated dynamic supply chain network graph, starting from the graph node, traverse the neighboring nodes to obtain the upstream neighbor graph node set of the graph node; and extract the connection edge attribute vectors corresponding to the directed connection edges between the neighboring nodes and the graph node to form the directed connection edge attribute set of the graph node. For each graph node in the updated dynamic supply chain network graph, the risk propagation delay parameter set of the graph node is obtained based on the event type component, the directed connection edge attribute set, and the preset risk delay derivation model. For each graph node in the updated dynamic supply chain network graph, the node risk score in the graph node feature vector is defined as the initial risk prediction value. Based on the upstream neighbor graph node set, the risk propagation delay parameter set, and the updated graph node feature vector, steps S1 to S3 are repeated until the preset maximum number of iterations is reached to obtain the risk prediction value sequence of the graph node. The risk propagation simulation results are composed of the risk prediction value sequence of each of the graph nodes; Steps S1 to S3 are as follows: S1, based on the risk propagation delay parameter set and the graph node feature vector, aggregate the risk prediction values ​​of the graph nodes in the upstream neighbor graph node set of the graph node to obtain the comprehensive risk input value of the graph node; S2, Based on the node risk score and node resilience score of the graph node, calculate the dynamic risk threshold; S3, compare the comprehensive risk input value with the dynamic risk threshold to obtain a comparison result, and update the risk prediction value of the graph node based on the comparison result through a preset state update function to obtain the updated risk prediction value of the graph node.

5. The method according to claim 4, characterized in that, For each graph node in the updated dynamic supply chain network graph, based on the event type component, the directed connection edge attribute set, and a preset risk delay derivation model, a risk propagation delay parameter set for the graph node is obtained, including: For each of the directed edge attribute vectors in the set of directed edge attributes, the connected edge attribute vector and the event type component are input into a preset risk propagation prediction model to obtain the risk propagation delay parameter corresponding to the connected edge attribute vector; the risk propagation delay parameter is used to characterize the degree of risk propagation delay from the neighbor node at one end of the directed edge corresponding to the connected edge attribute vector to the graph node. Based on the risk propagation delay parameters corresponding to the attribute vectors of each of the aforementioned connecting edges, the risk propagation delay parameter set of the graph nodes is obtained; The risk transmission prediction model is obtained through the following method: Obtain a preset historical risk propagation dataset; the historical risk propagation dataset includes historical event type components, historical supply information data, and historical risk transmission delay time parameters of historical risk propagation records; and use the historical event type components and the historical supply information data as input feature vectors, and use the historical risk transmission delay labels corresponding to the input feature vectors as training labels, divide the historical risk propagation dataset according to a preset dataset division ratio to obtain a delay model training set and a delay model test set; An initial delay prediction model is constructed based on the gradient boosting tree algorithm; and based on the training set of the delay model, the input feature vector is input into the initial delay prediction model to obtain the delay prediction value corresponding to the input feature vector; Based on the predicted delay value corresponding to the input feature vector and the historical risk transmission delay label corresponding to the input feature vector, the model prediction loss is calculated; Based on the model prediction loss, the initial delay prediction model is iteratively trained until a preset iteration termination condition is met to obtain a preliminary delay prediction model; and the preliminary delay prediction model is tested using the model test set until the preliminary delay prediction model meets a preset accuracy requirement to obtain the trained risk transmission prediction model.

6. A supply chain ESG risk transmission simulation system, characterized in that, The system includes: The data acquisition module is used to acquire text event data and dynamic supply chain network diagrams of the supply chain; the dynamic supply chain network diagram is used to characterize the current risk status of each supply chain entity and the dynamic supply relationship between the supply chain entities. The text feature extraction module is used to extract the semantic features of the text event data based on a preset text encoder to obtain an event feature vector; the event feature vector includes an event subject component and an event type component; The preliminary relationship analysis module is used to analyze the maintenance probability of the dynamic supply relationship between each supply chain entity in the dynamic supply chain network diagram based on the event feature vector, and to obtain supply relationship update data based on the maintenance probability. The dynamic graph update module is used to update the dynamic supply chain network graph based on the supply relationship update data, so as to obtain the updated dynamic supply chain network graph. The current risk analysis module is used to simulate risk propagation on the updated dynamic supply chain network diagram based on the event type components, obtain risk propagation simulation results, and obtain a risk warning report based on the risk propagation simulation results and the supply relationship update data.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images