A monitoring method and system for industry chain breakpoint management

Abstract

The present application relates to the technical field of industry chain monitoring and management, in particular to a monitoring method and system for industry chain breakpoint management; the method comprises: collecting production, logistics, finance, policy and environmental dynamic information through multi-source data acquisition, generating industry chain comprehensive features through standardization, dimension embedding representation and cross-dimension fusion, and introducing historical trend dependence to enhance forward-looking prediction; and constructing a dynamic coupling network, using dynamic edge weight to quantify the risk propagation strength between nodes, realizing cross-level breakpoint propagation analysis; and calculating risk indicators by fusing node features and network structure, combining adaptive threshold to filter breakpoint candidate nodes, using a nonlinear propagation function to predict multi-step evolution trend and map it to multi-level warning level. According to the risk level, a management strategy is generated, and the effect is evaluated in real time and the parameters are dynamically adjusted through a closed-loop optimization mechanism. The present application realizes closed-loop management of risk identification, prediction and adaptive management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industry chain monitoring and management, and particularly relates to a monitoring method and system for industry chain breakpoint management. BACKGROUND

[0002] Under the background of continuous improvement of global industry chain complexity and diversification of sudden disturbance factors, breakpoint risk prevention and control faces common technical challenges such as dynamic fusion of multi-source heterogeneous data, cross-level quantitative dependence and forward-looking early warning.

[0003] A kind of industry chain risk identification method and system based on timing chart disclosed in Chinese patent application with publication number CN117094558A, the method comprises: obtaining industry chain data and constructing risk factor graph with time series information, obtaining risk factor time series graph;Obtain industry chain risk data;Obtain industry chain state sequence and industry chain state transition graph, and extract risk factor time series graph according to industry chain state sequence and industry chain state transition graph, obtain risk factor time series sequence data;Industry chain risk data and risk factor time series sequence data are respectively embedded statically and dynamically, and static embedding vector and dynamic embedding vector are obtained respectively;The dynamic embedding vector and risk factor time series graph are jointly embedded information, and the result vector set of joint information embedding is obtained;The static embedding vector and the result vector set of joint information embedding are subjected to information fusion based on attention mechanism, and the fusion result is obtained;The final identification and prediction of industry chain risk are carried out.

[0004] The existing solution needs to consider the collaborative analysis capability of micro-enterprise behavior, meso-linkage and macro-regional influence, and simultaneously meet the real-time response demand of compound risk factors such as financial fluctuations, geopolitical policy adjustments and natural disasters, so as to drive the development of industry chain management towards the direction of real-time sensing, self-adaptive modeling and intelligent decision-making closed loop. SUMMARY

[0005] The present application aims at the problems in the background art and provides a monitoring method and system for industry chain breakpoint management.

[0006] The technical scheme of the present application is a monitoring method for industry chain breakpoint management, comprising the following specific implementation steps:

[0007] S1, obtain production, logistics, finance, policy and environmental dynamic information related to the industry chain through multi-source data acquisition;Standardize and fill in the missing values of the multi-source data, and construct fractal embedding representation for each type of data;Cross-dimension fusion is performed on the embedding vector to generate a comprehensive embedding representation of the industry chain;Introduce historical trend dependence for dynamic feature enhancement to obtain enhanced comprehensive features;

[0008] S2, based on the enhanced comprehensive features and the personalized attributes of each node in the industry chain, a node feature vector of each node is generated by fusion; an instant dependence sensitivity between nodes is calculated based on the node feature vector, and a dynamic adjacency matrix is constructed by combining a historical dependence strength index; a three-layer dynamic coupling network is constructed based on the dynamic adjacency matrix and the node set, and a coupling weight between nodes across layers is quantified through a cross-layer scaling factor;

[0009] S3, based on the dynamic coupling network, a breakpoint risk index of each node is calculated by fusing the node's own state, upstream dependence sensitivity and downstream conduction potential; an adaptive threshold is dynamically calculated according to the distribution of the node risk index of the whole network, and a breakpoint candidate node is screened; a multi-step risk evolution prediction is performed on the breakpoint candidate node through a nonlinear risk propagation function, and a predicted risk value is obtained; the predicted risk value is mapped to a multi-level early warning level;

[0010] S4, based on the early warning level and the risk evolution trend, a multi-objective governance strategy is generated, and the strategy execution effect is evaluated and the parameters are dynamically adjusted in real time through a closed-loop optimization mechanism.

[0011] Preferably, the fractal embedding representation is obtained by mapping the observation data of each type of data to a low-dimensional feature space through an embedding mapping function to obtain an embedding vector.

[0012] The embedding mapping function is a learnable parameter function, including a weight matrix and a bias term.

[0013] Preferably, the fusion process of cross-dimension fusion is specifically:

[0014] Align the embedding vectors from different dimensions according to time synchronization;

[0015] An attention weight matrix is constructed to evaluate the importance of each dimension to the overall industry chain state;

[0016] The weighted dimension vectors are combined, and a unified comprehensive embedding representation is generated through nonlinear transformation and feature mapping.

[0017] Preferably, the historical trend dependence is introduced for dynamic feature enhancement, specifically:

[0018] A historical time window is intercepted from the comprehensive embedding sequence;

[0019] The embedding vectors in the window are sequentially modeled by a recurrent neural network to capture time dependence and dynamic evolution features;

[0020] A context vector containing historical trend information is generated by adjusting the weights of different time steps.

[0021] Preferably, the three-layer dynamic coupling network includes:

[0022] The nodes of the link layer are supply, production, logistics, and sales links;

[0023] The nodes of the enterprise layer are specific enterprise entities;

[0024] The nodes of the regional layer are at least one of geographical regions and industrial clusters;

[0025] The nodes of each layer are connected by cross-layer edges, and the edge weights are calculated by multiplying the coupling factor by the cross-layer scaling factor.

[0026] Preferably, the breakpoint risk indicator of each node is calculated, specifically:

[0027] The node's own state, upstream dependency sensitivity, and downstream transmission potential are comprehensively considered;

[0028] The risk values transmitted by upstream nodes and the transmission potential of downstream nodes are weighted and aggregated by the node risk comprehensive function;

[0029] The risk is dynamically adjusted in combination with the coupling strength between nodes and historical dependency information.

[0030] Preferably, a self-adaptive threshold is dynamically calculated, specifically:

[0031] The self-adaptive threshold is dynamically calculated based on the mean and standard deviation of the risk values of all nodes in the network;

[0032] The filtering sensitivity is controlled by the threshold adjustment coefficient, that is:

[0033] ;

[0034] Where θ t represents the breakpoint risk threshold at time t; represents the mean of all node risk values at time t; represents the standard deviation of all node risk values at time t; represents the threshold adjustment coefficient, which controls the filtering sensitivity.

[0035] Preferably, the non-linear risk propagation function is based on a dynamic coupling network, which integrates the node feature vector and the risk values of upstream and downstream nodes, and predicts the node risk at the next time step through a non-linear mapping mechanism.

[0036] Preferably, the predicted risk value is mapped to a multi-level early warning level, specifically:

[0037] The continuous risk value is converted into a discrete low, medium, high, and extremely high four-level early warning level through a risk level mapping function;

[0038] The risk level mapping function dynamically adjusts the threshold interval according to the change trend, peak size, and fluctuation characteristics of the predicted risk value within a time window, and in combination with the overall risk distribution of the network.

[0039] The technical scheme of the present application: a monitoring system for industry chain breakpoint governance, which is used to execute the above-mentioned monitoring method for industry chain breakpoint governance, comprising:

[0040] A multi-source data acquisition and preprocessing module is used to acquire multi-source dynamic data of the industry chain in real time, and to preprocess, standardize and feature fuse the data to generate comprehensive feature vectors of the nodes;

[0041] An industry chain dynamic coupling network construction module is used to dynamically generate a multi-layer coupling network and establish the coupling strength between nodes based on the comprehensive features of the nodes and the dependency relationships;

[0042] A breakpoint risk identification and dynamic evolution prediction module is used to perform risk quantification, candidate node screening and multi-step risk evolution prediction using the network structure and node features, and to generate multi-level warning levels;

[0043] An adaptive governance strategy generation and closed-loop optimization module is used to intelligently generate multi-objective governance strategies based on warning information, and to dynamically adjust the strategy parameters through a closed-loop optimization algorithm for adaptive iteration of the governance strategies.

[0044] Compared with the prior art, the above technical scheme of the present application has the following beneficial technical effects:

[0045] The present application designs a monitoring method and system for industry chain breakpoint governance, which significantly improves the accuracy and timeliness of industry chain breakpoint monitoring through multi-source heterogeneous data fusion and multi-layer dynamic network modeling. Specifically, embedded mapping and cross-dimensional fusion functions are used to realize the unified representation of multi-source data such as production, logistics and finance, solving the feature fragmentation problem caused by data heterogeneity in traditional methods and enhancing the robustness and information integrity of the monitoring system; a three-layer dynamic coupling network (link layer / enterprise layer / region layer) is constructed in combination with a cross-layer scaling factor to accurately quantify the dependency sensitivity between nodes and the cross-layer risk propagation path, breaking through the limitations of single-layer supply chain models in expressing complex industry chain topologies; a multi-step evolution prediction is realized based on a nonlinear risk propagation function, high-risk nodes are identified in advance through dynamic threshold screening and risk level mapping, and the forward-looking and positioning accuracy of breakpoint warning are significantly improved; relying on a closed-loop optimization mechanism, multi-objective governance strategies such as resource adjustment and path switching are dynamically generated, and the execution effect is adjusted in real time to optimize the parameters, effectively overcoming the lag of traditional manual decision-making, reducing the intervention cost while improving the resilience of the industry chain; the present application realizes intelligent closed-loop of the whole process of "monitoring-prediction-governance", and provides a quantifiable, adaptive and high-response systematic solution for the safety control of the industry chain. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1A method flow chart of a monitoring method for industry chain breakpoint management is proposed in the present application.

[0047] Figure 2 A system architecture diagram of a monitoring system for industry chain breakpoint management is proposed in the present application. DETAILED DESCRIPTION

[0048] Embodiment one, as shown in the present application, a monitoring method for industry chain breakpoint management, comprising the following specific implementation steps: Figure 1

[0049] S1, through multi-source data acquisition, time series embedding, dynamic fusion and feature enhancement, forming a comprehensive representation of industry chain operation, laying a foundation for breakpoint risk identification and management, the specific implementation process is as follows:

[0050] S11, the multi-source data related to the operation of the industry chain is dynamically collected, and the data sources include but are not limited to: production data (yield, raw material inventory, equipment operation efficiency), logistics data (transportation delay, channel congestion index, warehouse turnover rate), financial data (credit level, liquidity, financing constraint), policy data (regulatory adjustment, trade barriers), environmental data (natural disaster index, energy supply fluctuation);

[0051] In order to ensure the comparability of data from different sources, standardization processing and missing value filling method are adopted to obtain the collected multi-source data D(t): ;

[0052] Among them, D i (t) is the observation data of data source i; N is the total number of data sources;

[0053] S12, for each type of data D i (t), construct fractal embedding representation: ;

[0054] Among them, E i (t) represents the embedding vector of the observation data of data source i at time t; W i represents the parameter set of embedding function, including weight matrix, bias term; represents the embedding mapping function, which is used to map the observation data to a low-dimensional feature space;

[0055] It should be noted that the embedding mapping function ​is a method for converting multi-dimensional heterogeneous data into low-dimensional feature representation, which can preserve the structural characteristics and timing rules of the data itself, while mapping different types of data into the same feature space for unified processing and analysis; in the industry chain monitoring, the embedding mapping function can convert multi-source data such as production, logistics, finance, policy and environment into vector representation with computability and comparability, so that the system can capture the unique information of each type of data and realize cross-dimensional fusion analysis, thereby providing accurate and reliable input basis for breakpoint risk identification and dynamic evolution prediction;

[0056] Through the above dimension-embedding operation, an embedded vector set is obtained at time t : ;

[0057] S13, after obtaining the embedded vectors of each dimension, further fusion is carried out to generate the time sequence representation X of the whole industry chain t :

[0058] ;

[0059] Wherein, X t represents the comprehensive embedded representation at time t; represents a cross-dimensional fusion function; represents a set of fusion layer parameters;

[0060] It should be noted that the cross-dimensional fusion function firstly aligns the embedded vectors from different dimensions according to time synchronization to ensure the integrity of information of each dimension at each time point, then constructs an attention weight matrix to evaluate the importance of each dimension to the overall industry chain state at the current time point, so as to adaptively allocate weights to reflect its relative influence, then combines the weighted dimension vectors to generate a unified comprehensive representation through nonlinear transformation and feature mapping; the fusion function can dynamically capture the mutual correlation and timing change of different dimensions in this process, so that the final comprehensive representation not only preserves the independent features of each dimension, but also reflects the overall trend and key driving factors across dimensions, providing high-fidelity and multi-level input features for subsequent breakpoint risk identification;

[0061] S14, introduce a dynamic feature enhancement mechanism, add historical trend dependence on the basis of comprehensive embedding to improve the forward-looking of breakpoint prediction: ;

[0062] Wherein, represents the enhanced comprehensive feature; represents context extraction parameters; represents the comprehensive embedding of the past k time points; represents a historical window context function;

[0063] It should be noted that the history window context function First, a certain length of historical time window is extracted from the collected comprehensive embedding sequence. These time windows can cover the recent changes in the operating status of the industrial chain and potential fluctuation patterns. Then, the embedding vectors within the window are sequentially modeled. Recurrent neural networks (RNNs) are used to capture time dependencies and dynamic evolution features. During the modeling process, the weights of different time steps are adjusted so that the model can focus on historical moment information that has a greater impact on the risk of future breakpoints. At the same time, comprehensive representations are extracted through hidden states or feature aggregation methods. The final generated context vector contains both historical trend information and can dynamically reflect the current state and future change potential.

[0064] S2. The enhanced comprehensive features output in step S1. This is transformed into a structured, analyzable supply chain network. The network structure reveals the dependencies between enterprises or links and the potential propagation paths of breakpoints, providing a foundation for subsequent breakpoint identification and governance strategy generation. The specific implementation process is as follows:

[0065] S21. Integrate global information with local attributes to form computable node features, enabling each node to dynamically adjust its feature representation according to the state of the industry chain. Specifically:

[0066] Define the set of nodes in the industry chain as V = {v1, v2, ..., v} i ,…,v n};

[0067] Among them, v i Let be the i-th node, and each node corresponds to an enterprise, production process, or key industry segment; n is the total number of nodes.

[0068] The enhanced comprehensive features output in step S1 Personalized attributes of each node (Including but not limited to enterprise size, process type, geographical location, and historical breakpoint frequency) are fused to obtain the node feature vector H. i (t):

[0069] ;

[0070] in, The feature fusion function is represented in this embodiment, which uses weighted concatenation to fuse enhanced comprehensive features and node individual attributes into a computable feature vector;

[0071] S22. By using dynamic edge weights, the intensity of potential risk propagation between nodes is quantified, upstream and downstream sensitivity is captured, and a quantitative basis for predicting breakpoint risks is provided. Specifically:

[0072] Based on node features H i (t) Calculate the coupling strength and dependence sensitivity between nodes, and construct a dynamic adjacency matrix A t : ;

[0073] Wherein, represents the instantaneous dependence sensitivity of node i to node j, with a value range of [0, 1]; W c represents the coupling parameter matrix between nodes, representing the influence weight between node feature vectors; represents the weight coefficient of balancing instantaneous features and historical dependence; A t represents the dynamic adjacency matrix, i.e. the dependence or coupling strength of node i to node j at time t; S ij represents the historical dependence strength index between node i and node j, such as long-term transaction frequency, supply-demand dependence or historical cooperation data;

[0074] S23, using multi-layer network, capturing micro-enterprise behavior, link dependence and macro regional influence, realizing cross-level breakpoint risk propagation analysis, that is, dividing the industrial chain into three layers of link layer, enterprise layer and regional layer network:

[0075] Link layer: nodes are supply, production, logistics and sales links, reflecting the whole chain process state;

[0076] Enterprise layer: nodes are specific enterprise entities, reflecting the micro operation state;

[0077] Regional layer: nodes are geographical regions or industrial clusters, reflecting the macro environmental influence;

[0078] Each layer of nodes is connected through cross-layer edges, and the edge weight uses the coupling factor of small step S22 multiplied by the cross-layer scaling factor : ;

[0079] Wherein, represents the coupling weight of cross-layer nodes i and j; represents the cross-layer scaling factor, reflecting the propagation ratio between different levels, which is set and optimized according to the actual experience or historical data of the industrial chain in the embodiment;

[0080] S24, the industrial chain network constructed is dynamically updated with the running state of the industrial chain, and can also respond to emergencies locally, ensuring that the breakpoint risk prediction model can reflect the actual state of the industrial chain in real time, specifically:

[0081] At each time step t, according to the node features H i (t) and the coupling factor Update network structure: G t = (V, At );

[0082] When an abnormal event or sudden disturbance is detected, the local edge weights are adjusted in real time:

[0083] ;

[0084] Among them, G t The dynamic network of the industrial chain at time t is represented, including the set of nodes and the dynamic edge weight matrix; This represents the local disturbance adjustment value, used for dynamic correction of edge weights when a sudden event occurs. It is calculated based on the event intensity, node sensitivity, and historical anomaly data.

[0085] S3. The dynamic coupling network G of the industrial chain constructed based on step S2 t =(V,A t This involves identifying potential breakpoints and links, predicting the evolution of breakpoint risks along the industrial chain, and providing decision support for governance strategies. The specific implementation process is as follows:

[0086] S31. By combining node characteristics with network structure, risk indicators are calculated, considering not only the node's own state but also quantifying upstream transmission risks and downstream potential impacts, to conduct a multi-dimensional comprehensive risk assessment, specifically:

[0087] For each network node v i Calculate the breakpoint risk index R i (t), taking into account the node's own state, upstream dependency sensitivity, and downstream transmission potential:

[0088] ;

[0089] Among them, R i (t) represents node v i The breakpoint risk index at time t reflects the probability of a breakpoint occurring at that node and its potential impact on the entire industry chain; U(i) represents the set of upstream neighbors of node i, that is, all nodes that have a supply relationship or dependency relationship with node i; D(i) represents the set of downstream neighbors of node i, that is, all nodes that depend on node i. This represents the coupling weight of node j∈U(i) to node i, i.e., the upstream risk transmission capability; R represents the coupling weight of node i to node k∈D(i), i.e., the potential for the risk of a node to propagate downstream; j (t-1) represents the risk value of upstream node j in the previous time step, which is used to dynamically recursively estimate the node risk and is obtained by quantifying the node risk in the previous time step. Represents the node risk synthesis function;

[0090] It should be noted that the node risk comprehensive function First, starting from the feature vector of each node, which integrates the attribute information of the node itself and the enhanced comprehensive features in step S1, the multi-dimensional attributes are unified to the same representation space through nonlinear mapping, while preserving the individual differences and the whole chain state information of the node, then the historical risk value of the node and the risk transmission of its upstream nodes are considered, through weighted aggregation or correlation calculation based on attention mechanism, the function can reflect the node's own vulnerability, upstream risk pressure and downstream potential impact at the same time, combined with the coupling strength and historical dependence information between nodes, the risk is dynamically adjusted, and the finally generated function can quantify the breakpoint risk of each node in the multi-dimensional and multi-level industrial chain network, and provide reliable input for subsequent candidate node screening and dynamic evolution prediction;

[0091] S32, through a dynamic threshold screening method, adapt to different industrial chain states and highlight the risk concentration area, specifically:

[0092] According to the node risk index R i (t) sets an adaptive threshold θ t (the threshold θ t Based on the dynamic adjustment of the whole network risk distribution, the potential breakpoint nodes are screened:

[0093] ; ;

[0094] Among them, B t represents the set of breakpoint candidate nodes at time t, that is, the nodes whose risk values are higher than the threshold; θ t represents the breakpoint risk threshold at time t, which is used to screen candidate nodes; represents the mean of all node risk values at time t; represents the standard deviation of all node risk values at time t; represents the threshold adjustment coefficient, which controls the screening sensitivity;

[0095] S33, simulate the dynamic process of breakpoint risk propagation from nodes to upstream and downstream and across layers, and predict the evolution trend of industrial chain breakpoints, that is: use the dynamic multi-layer coupled network G t establish a breakpoint risk propagation model: ;

[0096] Among them, represents the predicted risk value of node i at the next time step; represents a nonlinear risk propagation function for predicting the risk of a node at the next time step;

[0097] It should be noted that the nonlinear risk propagation function Based on the industry chain dynamic coupling network constructed in step S2, the feature vector of each node is combined with the risk values of its upstream and downstream nodes, and an initial risk response representation is formed through multi-dimensional feature fusion, which reflects the node's own state, historical risk transmission and upstream and downstream coupling strength; and a nonlinear mapping mechanism is introduced, which can use graph neural networks or attention mechanisms to nonlinearly weight different influence weights and coupling relationships between nodes through learnable parameters, so that the risk transmission in the network can reflect the sensitivity and propagation strength of different nodes; then through training or iterative optimization, the function can capture the non-uniformity, suddenness and cross-level influence of risk propagation, ensuring that the function output reasonably reflects the risk evolution trend under node local disturbance or whole chain abnormal situation, and finally forming a nonlinear risk propagation function that can be used for multi-step breakpoint prediction and dynamic early warning, realizing the complete mapping of industry chain risk from micro nodes to macro chains;

[0098] S34, through multi-step prediction and grade division, the position and severity of the possible breakpoint are predicted in advance, and a quantitative basis is provided for governance strategy, specifically:

[0099] The candidate breakpoint node B t and its upstream and downstream neighbors are subjected to multi-step risk evolution prediction to generate short-term and medium-term risk evolution curves;

[0100] The prediction results are mapped to a multi-level early warning level L i (t)∈{low, medium, high, very high}:

[0101] ;

[0102] Where L i (t) represents the early warning level (low, medium, high, very high) of node i, which represents the breakpoint risk severity, and the predicted risk is obtained through a mapping function; represents the risk level mapping function; τ represents the prediction time window length, i.e. the time span of multi-step evolution prediction;

[0103] It should be noted that the risk level mapping function is a nonlinear function that converts continuous risk values into discrete level labels, and its core idea is to comprehensively judge the risk state of the node according to the change trend, peak size and fluctuation characteristics of the predicted risk value within the time window, and map it to low, medium, high or very high level; This function is usually realized by setting dynamic threshold intervals, different intervals correspond to different risk levels, and these intervals are not fixed, but are dynamically adjusted in combination with the overall risk distribution, historical mean and standard deviation of the network, etc. statistical characteristics, so as to ensure that the mapping result can truly reflect the changes in the system operating environment;

[0104] S35. The predicted breakpoint risks and evolution trends are presented through multi-dimensional visualization, including node risk heatmaps, link risk propagation diagrams, and cross-layer network evolution dynamic diagrams.

[0105] S4. Based on the breakpoint risk prediction and multi-level early warning information generated in step S3, formulate targeted governance strategies for potential breakpoints in the industrial chain, and achieve dynamic adjustment and adaptive iteration of the strategies through closed-loop optimization. Through the cycle of risk-strategy-feedback-optimization, the breakpoint governance process of complex industrial chains is systematized and made intelligent, specifically as follows:

[0106] Based on the risk level and evolution trend of candidate breakpoint nodes, multi-objective governance strategies are automatically generated, including but not limited to node resource adjustment, inventory optimization, alternative path switching in the supply chain, and collaborative intervention by key enterprises. The strategy generation process fully considers the node risk priority, governance cost, intervention timeliness, and the chain reaction of the strategy on the entire network to avoid new risk spillovers that may be caused by single-point governance.

[0107] Through a closed-loop optimization mechanism, the effectiveness of governance strategies is evaluated in real time: continuously monitor changes in node risks and execution feedback, and dynamically adjust strategy parameters, such as intervention intensity, execution order, and resource allocation ratio, to maximize risk mitigation and supply chain stability. The optimization process uses adaptive algorithms, which can combine predicted risk evolution curves and network coupling characteristics to achieve multi-objective trade-offs, while avoiding reliance on traditional fixed rules and human experience.

[0108] Example 2, as Figure 2 As shown, the present invention proposes a monitoring system for supply chain disruption governance, which is used to execute a monitoring method for supply chain disruption governance proposed in Embodiment 1, including: a multi-source data acquisition and preprocessing module, a supply chain dynamic coupling network construction module, a disruption risk identification and dynamic evolution prediction module, and an adaptive governance strategy generation and closed-loop optimization module.

[0109] The multi-source data acquisition and preprocessing module is responsible for acquiring multi-source dynamic data from all links of the industry chain in real time, and performing preprocessing, noise reduction, missing value filling and standardization on the acquired data. It also generates a comprehensive feature vector for each node through feature extraction, cross-source fusion and dynamic embedding technology, ensuring that the node characteristics and the overall industry chain status are represented efficiently and accurately, providing high-quality input for subsequent network construction.

[0110] The industrial chain dynamic coupling network construction module, based on the comprehensive characteristics of nodes and the upstream and downstream dependencies of the supply chain, dynamically generates a multi-layer coupling network, establishes the edge weights and coupling strength between nodes, and considers cross-level influence and time changes to form a structured network that can reflect the operating status of the industrial chain and the interaction of nodes, providing a network foundation for identifying breakpoint risks.

[0111] The breakpoint risk identification and dynamic evolution prediction module quantifies the risk of each node and screens candidate breakpoints by using network structure and node features, and predicts the multi-step evolution trend of the risk through nonlinear propagation modeling to generate multi-level early warning levels. The module simultaneously considers upstream transmission risk, downstream impact and historical risk dynamics to realize closed-loop analysis from node risk quantification to whole-chain evolution prediction.

[0112] The adaptive governance strategy generation and closed-loop optimization module intelligently generates multi-objective governance strategies, including resource allocation, inventory optimization, supply link switching and key enterprise collaborative intervention, based on the predicted breakpoint risk and early warning information. Through closed-loop optimization algorithm, the strategy parameters are dynamically adjusted to realize adaptive iteration of the governance strategy, continuously optimize the breakpoint relief effect and industry chain stability, and feedback the strategy execution effect for the next round of strategy adjustment and optimization decision.

[0113] The embodiments of the present application are described in detail above in combination with the drawings, but the present application is not limited thereto, and various changes can be made within the knowledge range possessed by those skilled in the art without departing from the purpose of the present application.

Claims

1. A monitoring method for addressing supply chain disruptions, characterized in that, The specific implementation steps include the following: S1. Collect dynamic information on production, logistics, finance, policy and environment related to the industrial chain through multi-source data collection; standardize the multi-source data and fill in missing values, and construct a fractal embedding representation for each type of data; fuse the embedding vectors across dimensions to generate a comprehensive embedding representation of the industrial chain; introduce historical trend dependence to perform dynamic feature enhancement to obtain the enhanced comprehensive features; In this context, fractal embedding means that the observed data of each class of data is mapped to a low-dimensional feature space through an embedding mapping function to obtain an embedding vector. The embedding mapping function is a learnable parameter function that includes a weight matrix and a bias term; The cross-dimensional fusion process is specifically as follows: Align embedding vectors from different dimensions according to time synchronization; Construct an attention weight matrix to assess the importance of each dimension to the overall industry chain status; The weighted vectors of each dimension are combined, and a unified comprehensive embedding representation is generated through nonlinear transformation and feature mapping. Among these, the introduction of historical trend dependence for dynamic feature enhancement specifically includes: Extract historical time windows from the integrated embedded sequence; Embedded vectors within the window are sequentially modeled using a recurrent neural network to capture temporal dependencies and dynamic evolutionary features; By adjusting the weights of different time steps, a context vector containing historical trend information is generated; S2. Based on the enhanced comprehensive features and the personalized attributes of each node in the industry chain, a node feature vector for each node is generated by fusion; the instantaneous dependency sensitivity between nodes is calculated based on the node feature vector, and a dynamic adjacency matrix is ​​constructed by combining the historical dependency strength index; based on the dynamic adjacency matrix and the node set, a three-layer dynamic coupling network is constructed, and the coupling weight between cross-level nodes is quantified by the cross-layer scaling factor. S3. Based on a dynamic coupling network, the system integrates the node's own state, upstream dependency sensitivity, and downstream transmission potential to calculate the breakpoint risk index for each node; it dynamically calculates an adaptive threshold based on the distribution of risk indices across the entire network to screen candidate breakpoint nodes; it performs multi-step risk evolution prediction on the candidate breakpoint nodes using a nonlinear risk propagation function to obtain the predicted risk value; and it maps the predicted risk value to a multi-level early warning system. Specifically, the breakpoint risk index for each node is calculated as follows: Taking into account the node's own status, upstream dependency sensitivity, and downstream transmission potential; The risk value transmitted by upstream nodes and the transmission potential of downstream nodes are weighted and aggregated using a node risk comprehensive function; The risk is dynamically adjusted by combining the coupling strength between nodes and historical dependency information; S4. Based on the early warning level and risk evolution trend, generate multi-objective governance strategies, and use a closed-loop optimization mechanism to evaluate the effectiveness of strategy implementation and dynamically adjust parameters in real time.

2. The monitoring method for addressing supply chain disruptions according to claim 1, characterized in that, Three-layer dynamically coupled networks include: The nodes in the process layer are supply, production, logistics, and sales. The nodes in the enterprise layer are specific enterprise entities; The nodes in the regional layer can be at least one of geographical regions or industrial clusters; Each layer of nodes is connected by cross-layer edges, and the edge weight is calculated by multiplying the coupling factor by the cross-layer scaling factor.

3. The monitoring method for addressing supply chain disruptions according to claim 2, characterized in that, The adaptive threshold is calculated dynamically, specifically as follows: The adaptive threshold is dynamically calculated based on the mean and standard deviation of the risk values ​​of all nodes in the network. The sensitivity of the screening is controlled by adjusting the threshold coefficient, that is: ； Where, θ t The risk threshold for the breakpoint at time t; This represents the average risk value of all nodes at time t. This represents the standard deviation of the risk values ​​of all nodes at time t; This represents the threshold adjustment coefficient, which controls the screening sensitivity.

4. The monitoring method for addressing supply chain disruptions according to claim 3, characterized in that, The nonlinear risk propagation function is based on a dynamically coupled network. It integrates the node feature vectors with the risk values ​​of upstream and downstream nodes and predicts the node risk at the next time step through a nonlinear mapping mechanism.

5. The monitoring method for addressing supply chain disruptions according to claim 4, characterized in that, The predicted risk values ​​are mapped to multiple warning levels, specifically as follows: The continuous risk value is transformed into a discrete four-level early warning level: low, medium, high, and extremely high, through a risk level mapping function. The risk level mapping function dynamically adjusts the threshold range based on the trend, peak size, and fluctuation characteristics of the predicted risk value within the time window, combined with the overall risk distribution of the network.

6. A monitoring system for managing supply chain disruptions, used to execute the monitoring method for managing supply chain disruptions as described in any one of claims 1 to 5, characterized in that, include: The multi-source data acquisition and preprocessing module is used to acquire multi-source dynamic data of the industrial chain in real time, and to preprocess, standardize and fuse the data to generate a comprehensive feature vector of the node. The industrial chain dynamic coupling network construction module is used to dynamically generate multi-layer coupling networks and establish the coupling strength between nodes based on the comprehensive characteristics and dependencies of nodes; The breakpoint risk identification and dynamic evolution prediction module is used to quantify risks, screen candidate nodes, and predict multi-step risk evolution by utilizing network structure and node characteristics, and to generate multi-level early warning levels. The adaptive governance strategy generation and closed-loop optimization module is used to intelligently generate multi-objective governance strategies based on early warning information, and dynamically adjust the strategy parameters through a closed-loop optimization algorithm to perform adaptive iteration of the governance strategy.

Images