A Comprehensive Assessment and Early Warning Method and System for Supply Chain Security Risks
By combining graph convolutional neural networks and entropy weighting, the subjective and unsystematic nature of supply chain security risk assessment in existing technologies is addressed, enabling comprehensive and accurate risk assessment and early warning of the supply chain, and ensuring the reliability and timeliness of the assessment results.
Patent Information
- Application Number
- CN202311480582.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-08
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2043-11-08
AI Technical Summary
Existing supply chain security risk assessment methods rely on qualitative analysis and expert judgment, lacking scientific quantitative assessment. This results in highly subjective assessment results, making it difficult to conduct reliable comparisons and quantitative analysis. They also ignore the internal connections between different links and participants in the supply chain, failing to capture and respond to dynamically changing risks in a timely manner, and lacking a comprehensive and systematic assessment of the entire supply chain.
By combining graph convolutional neural networks (GCN) and entropy weighting, a supply chain network is constructed to identify key enterprises, establish a three-level indicator system, and calculate the weight coefficients of each level of indicators, thereby enabling security assessment and early warning for each link in the supply chain.
It provides a more accurate and comprehensive supply chain security risk assessment and early warning capability, enabling early detection and response to potential risks and ensuring the stable operation of the supply chain.
Smart Images

Figure CN117495094B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of comprehensive safety risk assessment technology, and in particular to a method and system for comprehensive assessment and early warning of supply chain safety risks. Background Technology
[0002] Given the increasingly prominent global industrial and supply chain security risks, an accurate and efficient method is needed to assess and provide early warnings of these risks. Existing industrial chain assessment methods generally fall into two categories: The first is qualitative assessment. This method focuses on subjectively judging and describing the risks of each link and participant in the industrial chain, typically using expert opinions, experience, and professional knowledge. Information can be obtained through questionnaires, expert interviews, etc., followed by comprehensive analysis and evaluation. Research literature on this type of method is relatively abundant. For example, Chinese Patent No. CN114580970A, published on June 3, 2022, entitled "A Defense Industrial Chain Assessment Method Based on Risk Index Assessment," starts with an overall framework, assesses the overall situation of the company, decomposes the risk assessment indicators of the defense industrial chain item by item, and combines expert scoring to give a reasonable risk assessment value. Chinese Patent No. CN114493078A, Publication Date: May 13, 2022, Patent Title: A Risk Early Warning Method, System, Equipment, and Medium for the Energy and Power Industry Chain. This method, based on historical data of the energy and power industry chain, constructs a risk assessment index system, determines weights and comprehensive evaluation scores using analytic hierarchy process (AHP) and expert evaluation, and compares the comprehensive evaluation score with a risk threshold to determine the risk early warning level of the energy and power industry chain. However, due to reliance on expert opinions, experience, and professional knowledge, this type of method is easily influenced by the subjective biases and personal experiences of the assessors, leading to uncertainty and subjectivity in the assessment results. Simultaneously, qualitative assessment methods struggle to provide precise quantitative results, limited to subjective judgments and descriptions, lacking objective indicators and data support. The second type is quantitative assessment methods: These methods assess industry chain risks based on data and indicators, using quantitative analysis to identify and measure potential risks. Commonly used methods include risk indicator models, statistical analysis, and mathematical modeling. Quantitative assessment methods can provide more objective and quantifiable risk assessment results; however, there is still limited literature on this type of method. For example, Chinese patent number CN111898916A, publication date: November 6, 2020, patent title: A risk monitoring system and monitoring method for the coal industry chain. This method assesses the risk of a single node in the industry chain by collecting, associating and merging operational data, and then obtains the risk of the entire industry chain based on the risk assessment results of the single node.Chinese Patent No. CN115860474A, Publication Date: March 28, 2023, Patent Title: A Supply Chain Security Early Warning Method and System Based on Ensemble Learning. This method iteratively analyzes the supply chain and service chain within the supply chain to generate supply feature vectors and service feature vectors. These are then concatenated with the basic feature vectors of midstream enterprises to obtain supply aggregation indicators and service aggregation indicators for midstream enterprises. After aggregation at the same level, supply chain feature vectors and service chain feature vectors are obtained. Finally, a machine learning model is trained using ensemble learning to obtain an evaluation model. The evaluation model then predicts the input feature vectors to obtain the supply chain risk level security early warning prediction result. However, quantitative assessment methods require reliable data support. If the data quality is low or the necessary data is lacking, the assessment results may be inaccurate or unreliable. Furthermore, selecting an appropriate model for analysis and calculation is crucial, and choosing a suitable model can be challenging, as different models may have different impacts on the results.
[0003] In summary, existing supply chain security risk assessment methods primarily rely on qualitative analysis and expert judgment, lacking scientific quantitative assessment methods. This leads to highly subjective assessment results, making reliable comparisons and quantitative analysis difficult. Furthermore, when assessing supply chain security risks, the internal connections between different links and participants within the supply chain are often overlooked. However, the transmission and chain reactions of risks within the supply chain are often caused by internal connections; ignoring this may result in missed identification and misjudgment of potential risks. Existing methods focus more on the current state of the supply chain, conducting static assessments. However, supply chain security risk is a dynamic process, constantly influenced by changes in the external environment and internal factors. Static assessments cannot promptly capture and respond to dynamic changes in risks, limiting the effectiveness of risk management. Simultaneously, existing methods often focus only on a specific link or type of risk within the supply chain, lacking a comprehensive and systematic assessment of the entire supply chain, which may lead to misjudgments and omissions of the overall risk situation. To address these issues, this invention proposes a comprehensive supply chain security risk assessment and early warning method and system based on graph convolutional neural networks and entropy weighting, aiming to solve the problems of existing methods and improve the accuracy and practicality of supply chain security risk assessment. Summary of the Invention
[0004] The purpose of this invention is to provide a comprehensive assessment and early warning method and system for supply chain security risks in order to solve the above-mentioned problems. This invention can comprehensively assess supply chain security risks and identify potential risks, providing effective risk management and control basis for each link and participant.
[0005] The present invention achieves the above objectives through the following technical solutions:
[0006] A comprehensive assessment and early warning method for supply chain security risks includes the following steps:
[0007] Step 1: Construct the upstream, midstream, and downstream network structure of the industrial chain based on the data of the industrial chain to be evaluated, and determine the main links of the industrial chain;
[0008] Step 2: Use the Graphical Graph Neural Network (GCN) to learn node attributes and network relationship structure features to screen key enterprises in the main links of the industrial chain;
[0009] Step 3: Establish a three-level indicator system for supply chain security evaluation and assess the third-level indicators of key enterprises. Calculate the scores of the third-level indicators for the main links according to the ranking weight of key enterprises.
[0010] Step 4: Use the entropy weight method to calculate the weight coefficients of each level of the indicator system and combine them with the scores of the third-level indicators of the main links at the industry level to calculate the safety scores of the main links according to the weight coefficients of each level of indicators.
[0011] Step 5: Calculate the weight coefficients of each major link in the industrial chain and calculate the overall security score of the industrial chain based on the weight coefficients of the major links in the industrial chain.
[0012] A further proposed solution is that, in step 1, the data of the industrial chain to be evaluated includes the basic attribute characteristics of enterprises and the characteristics of the industrial chain network structure; the upstream and downstream network structure of the industrial chain is that each node in the industrial chain network represents an enterprise or a link, while the edges represent the relationship between enterprises or links; the industrial chain network is constructed using graph theory and network analysis methods. By constructing the industrial chain network, the enterprises and links in the industrial chain are displayed in the form of a graph. Determining the main links of the industrial chain refers to further analyzing the industrial chain network based on the constructed industrial chain network to determine the main links of the industrial chain.
[0013] A further solution is that, in step 2, the screening of key enterprises includes data processing, feature extraction, key node identification model, network key node prediction, and screening of key enterprises.
[0014] Data processing involves removing a very small number of isolated nodes from the network data, ensuring that all input graphs are connected.
[0015] Feature extraction involves constructing a subgraph network by extracting a two-degree subgraph for each node in the network. A two-degree subgraph network for each node is constructed by selecting neighboring nodes no more than three degrees away from the target node. Features of the subgraph network are extracted, and a feature vector consisting of five features is constructed for each node. The subgraph network features and the five feature vectors for each node are used as the original input to the model. The five features are degree centrality, betweenness centrality, k-shell decomposition, compact centrality, and node embedding. Degree centrality represents the degree k of node i. i The formula for the number of neighboring nodes directly connected to this node is: Betweenness centrality represents the load during information propagation in a network; calculating the shortest path between any two nodes in the network, the more times a node is included in the shortest path, the greater its load and the more important it is; the betweenness of node v is defined as...
[0016]
[0017] Among them, g st This represents the number of shortest paths from point s to point t. The shortest path from point s to point t passes through point i; the k-shell decomposition method indicates that the K-shell value of a node is a score of the node's position in the network; the formula for calculating compact centrality is...
[0018]
[0019] Where, d ij It is the distance between node i and node j; node embedding is compiled using node2vec to preserve the network's structural information, mapping nodes to the embedding space, performing one-hot encoding, and then multiplying by the embedding matrix to obtain the embedding vector of each node; the five features comprehensively consider the network's local, global, and positional attributes.
[0020] The key node identification model consists of a four-layer network architecture: an input layer, a GCN layer, a fully connected layer, and an output layer. The input layer comprises node feature vectors consisting of subgraph network features and five feature vectors. The GCN layer is defined as follows:
[0021] H i+1 =δ(AH i W i +b i )
[0022] Where A is the symmetric normalized Laplacian operator for a two-degree subgraph network; H i W represents the node of the i-th GCN layer; i and b i δ represents the learnable weights and error parameters, respectively; δ is a non-linear activation function, which is input into the key node identification model of the industrial chain network based on the feature vector extracted from each node for training; the fully connected layer plays a classification role and includes a three-layer architecture. The first fully connected layer fits the dataset based on the dropout technique with a degenerate learning rate; the second and third fully connected layers use the exponential linear unit (ELU) activation function to shorten training time and improve accuracy; the output layer is the keyness score output value of each node in the network.
[0023] A further approach is that, in step 2, the prediction of key network nodes involves inputting network feature data, calculating the minimum loss using the regression loss function MSE, and finally obtaining the predicted keyness score for each node in the network.
[0024] The selection of key enterprises is based on the upstream, midstream and downstream network structure of the industrial chain and the main links of the industrial chain. The top N node enterprises with the highest predicted key scores in the main links are selected as key representative enterprises in each main link, and the ranking weight of key representative enterprises is determined.
[0025] A further proposed solution is that, in step 3, the three-tiered indicator system for industrial chain security evaluation consists of three primary indicators: industrial chain competitiveness, industrial chain control, and industrial chain development environment. Industrial chain competitiveness includes technological innovation capability, profitability, and market competitiveness, and at the micro level, it requires enhancing the technological innovation capability and technological innovation output capability of core enterprises in key links of the industrial chain. Industrial chain control includes the external dependence or foreign capital control rate of core technologies, industrial capital, and markets. The industrial chain development environment focuses on examining the talent guarantee conditions of the entire industry under evaluation. Calculating the tiered indicator scores of major links based on the ranking weight of key enterprises refers to calculating the tiered indicator scores of each major link of the industrial chain by weighted summing the ranking weights of each enterprise in the major links and the scores of the tiered indicators.
[0026] A further proposed solution involves step 4, where the entropy weight method calculates the weight coefficients of the indicator system by standardizing the third-level indicator data, transforming it into dimensionless relative indicators to eliminate dimensional differences between different indicators; constructing an n×m decision matrix based on the standardized data, where n represents the number of indicators and m represents the number of samples; and calculating the entropy value for each column of data for each indicator, where the entropy value reflects the information content and degree of difference of the indicator; the formula for calculating the entropy value is: Where p represents the data proportion, k = 1 / ln(n), (0 ≤ e j <1); Calculate the weight coefficient of each indicator based on the entropy value; the formula for calculating the indicator weight coefficient is: w j = (1-E j ) / ∑(1-E j ), where E j The entropy value of the indicator is given. According to the hierarchical structure of the indicator system, the weight coefficients of the sub-indicators are multiplied level by level according to the hierarchical structure to obtain the weight coefficients of the indicator system at each level. The safety score of the main links of the industrial chain is calculated layer by layer by weighted summation based on the weight coefficients of the indicator system at each level and the scores of the three-level indicators of the main links of the industrial chain to obtain the safety score of the main links of the industrial chain.
[0027] A further proposed solution is that, in step 5, the weight coefficients of each major link in the industrial chain are calculated by multiplying the ranking weights of enterprises in each major link according to the hierarchical structure of the industrial chain to obtain the weight coefficients of the major links in the industrial chain; the weight coefficients are normalized so that their sum equals 1 to ensure that the relative relationship of the weights remains unchanged; the overall security score of the industrial chain is calculated by combining the weight coefficients of the major links in the industrial chain with the security scores of each major link in the industrial chain, and then calculating the final security score of the industrial chain by weighted summation layer by layer.
[0028] In another aspect, the present invention provides a comprehensive assessment and early warning system for supply chain security risks, comprising:
[0029] Data Acquisition and Preprocessing Module: The data acquisition and preprocessing module is used to collect network node attribute data, network relationship structure data, and indicator data of the indicator system related to the industrial chain to be evaluated, and to perform data preprocessing operations on the relevant data;
[0030] Node Keyness Score Prediction Module: The node keyness score prediction module is used to implement the GCN-based key node identification model of the industrial chain network. It extracts node features and network structure features from the industrial chain network node data and network relationship structure data, inputs them into the identification model, and outputs the keyness score prediction value of each node in the industrial chain network.
[0031] Key Enterprise Identification Module: Based on the upstream, midstream and downstream network structure of the industrial chain and the main links of the industrial chain, the key enterprise identification module extracts the top N node enterprises with the highest predicted key scores in the main links as key representative enterprises in each main link;
[0032] Key Enterprise Ranking Weight Calculation Module: The key enterprise ranking weight calculation module calculates the ranking weight of key enterprises in each major link of the industrial chain based on the predicted criticality scores of key enterprises and normalizes them according to the maximum-minimum standardization method.
[0033] The module for calculating the weight coefficient of key links: Based on the hierarchical structure of the industrial chain, the module for calculating the weight coefficient of key links calculates the weight of each key link by multiplying the weight of each key link by the weight of each key link according to the hierarchical structure and obtains the weight coefficient of key links in the industrial chain.
[0034] Risk assessment indicator weight coefficient calculation module: The risk assessment indicator weight coefficient calculation module is used to calculate the weight coefficient of each level of the indicator system by multiplying the weight coefficients of the sub-indicators according to the hierarchical structure of the indicator system based on the weight coefficient calculation formula of the entropy weight method.
[0035] The module for calculating the safety score of the three-level indicators for key links: This module calculates the safety score of the three-level indicators for each key link in the industrial chain by combining the ranking weight of each enterprise in the key links with the scores of the three-level indicators in a weighted summation manner.
[0036] The safety score calculation module for key links: The safety score calculation module for key links calculates the safety score of key links in the industrial chain by combining the weight coefficients of the indicator system at each level with the scores of the three-level indicators of each key link in the industrial chain in a weighted summation manner.
[0037] The supply chain security assessment and early warning module calculates the final supply chain security score by combining the weight coefficients of the main links in the supply chain with the security scores of each main link in the supply chain through a weighted summation method.
[0038] A further embodiment is that the network node attribute data includes enterprise name, industry, size, number of patents, market share, company size, registered capital, and financing amount; the network relationship structure data includes supply relationship, investment relationship, cooperation relationship, and competition relationship.
[0039] The beneficial effects of this invention are as follows:
[0040] This invention discloses a comprehensive assessment and early warning method and system for supply chain security risks. By employing a comprehensive assessment method, the system completes the assessment and early warning tasks for supply chain security risks. This method combines the advantages of qualitative and quantitative assessment methods. By comprehensively considering multiple factors and indicators, it conducts a comprehensive assessment of supply chain risks. The comprehensive assessment method typically includes the comprehensive calculation of multiple assessment dimensions, multiple indicators, and multiple weights to obtain a comprehensive risk assessment result. Compared with individual qualitative or quantitative methods, the comprehensive assessment method can more comprehensively consider all aspects of the supply chain and provide a more comprehensive and complete assessment result.
[0041] This invention first introduces Graph Convolutional Neural Networks (GCNs) to identify key companies in each link of the industrial chain. GCNs can accurately capture the complex relationships between nodes in the industrial chain by learning and transmitting the features of the graph structure, thereby identifying key companies. Second, it establishes an evaluation system for industrial chain security risk assessment indicators and assesses the risks of key enterprises in each link of the industrial chain. Furthermore, it employs the entropy weight method to assign weights to indicators at each level of the indicator system, calculating the security score of the main links in the industrial chain. The entropy weight method objectively measures the importance and contribution of indicators, thus rationally allocating the weights of each indicator. The application of the entropy weight method ensures more accurate and reliable assessment results. Finally, based on the weight coefficients of the main links in the industrial chain and the security scores of each main link, a weighted summation method is used to calculate and ultimately complete the assessment of the risks of each link and the overall risk of the industrial chain. This provides a more accurate and comprehensive industrial chain security risk assessment and early warning capability, helping to identify and respond to potential risks early and ensure the stable operation of the industrial chain. Attached Figure Description
[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0043] Figure 1 This is a flowchart of the present invention.
[0044] Figure 2 This is a system structure diagram of the present invention. Detailed Implementation
[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0046] In any embodiment, such as Figure 1 As shown, the present invention provides a comprehensive assessment and early warning method for supply chain security risks, comprising the following steps:
[0047] Step S1: Construct the upstream, midstream, and downstream network structure of the industrial chain based on the data of the industrial chain to be evaluated and determine the main links of the industrial chain. The data of the industrial chain to be evaluated includes basic enterprise attributes (such as enterprise name, industry, scale, number of patents, market share, company size, registered capital, financing amount, etc.) and industrial chain network structure relationship characteristics such as supply relationships, investment relationships, cooperation relationships, and competition relationships. The upstream, midstream, and downstream network structure of the industrial chain is such that each node represents an enterprise or a link, and edges represent the relationships between enterprises or links. Graph theory and network analysis methods can be used to construct the industrial chain network. By constructing the industrial chain network, the enterprises and links in the industrial chain can be displayed in the form of a graph, which helps to better understand the composition and relationships of the entire industrial chain. Determining the main links of the industrial chain refers to further analyzing the constructed industrial chain network to determine the main links of the industrial chain. Only the main links are identified because, considering the numerous risk points in the enterprise's industrial chain and supply chain, it is difficult to exhaustively cover them all. Therefore, we can only focus on the core links and key risks from the perspective of industrial chain security research.
[0048] Step S2: Use the Graphical Graph Neural Network (GCN) to learn node attributes and network relationship structure features to screen key enterprises in the main links of the industrial chain; the key enterprise screening steps include data processing, feature extraction, key node identification model generation, network key node prediction, and key enterprise screening.
[0049] In particular, during data processing, some isolated nodes in the network data are removed, so that the input graphs used in this method are all connected graphs.
[0050] Specifically, feature extraction involves extracting a two-degree subgraph of each node in the network to construct a subgraph network. Based on the three-degree influence principle, this invention selects neighboring nodes no more than three degrees away from the target node to construct the two-degree subgraph network for that node. Further, features of the subgraph network are extracted, and a feature vector consisting of five features is constructed for each node. The subgraph network features and the five feature vectors of each node are used as the original input to the model. The five features refer to degree centrality, betweenness centrality, k-shell decomposition, compact centrality, and node embedding. Degree centrality represents the degree k of node i. i The formula for the number of neighboring nodes directly connected to this node is: Betweenness centrality represents the load during information propagation in a network; calculating the shortest path between any two nodes in the network, the more times a node is included in the shortest path, the greater its load and the more important it is; the betweenness of node v is defined as...
[0051]
[0052] Among them, gst This represents the number of shortest paths from point s to point t. The K-shell value represents the number of points i that pass through in the shortest path from point s to point t; the K-shell decomposition method indicates that the K-shell value of a node is a score of the node's position in the network; the closer a node is to the interior of the network, the greater its influence; conversely, if a node is located at the edge of the network, its criticality is less; the implementation method is to remove the nodes at the outermost edge of the network layer by layer, leaving the nodes in the core of the network as critical nodes; the formula for calculating compact centrality is...
[0053]
[0054] Where, d ij is the distance between node i and node j. Node compact centrality describes how quickly a node can reach other nodes, reflecting the node's density within the network. The higher the compact centrality, the closer the node is to other nodes, and thus, the more central it is in the network. Node embedding is performed using node2vec to preserve the network's structural information. After mapping nodes to the embedding space, one-hot encoding is performed, and then multiplied by the embedding matrix to obtain the embedding vector for each node. These five features comprehensively consider the network's local, global, and positional attributes.
[0055] Specifically, a key node identification model for the industrial chain network is constructed. The identification model consists of a four-layer network architecture: an input layer, a GCN layer, a fully connected layer, and an output layer. The input layer comprises node feature vectors consisting of subgraph network features and five feature vectors. The GCN layer is defined as follows:
[0056] H i+1 =δ(AH i W i +b i )
[0057] Where A is the symmetric normalized Laplacian operator for a two-degree subgraph network; H i W represents the node of the i-th GCN layer; i and b iδ represents the learnable weights and error parameters, respectively; δ is a non-linear activation function, which can be ReLU (Rectified Linear Unit), Leaky ReLU, ELU (Exponential Linear Unit), Sigmoid, Tanh (Hyperbolic Tangent), etc. This invention preferentially chooses the Exponential Linear Unit (ELU) activation function; the feature vectors extracted from each node are input into the key node identification model of the industrial chain network for training. It should be noted that before inputting the features into the model for training, the extracted features need to be normalized using the min-max normalization method, mapping each feature value to the range [0,1]; the fully connected layer mainly plays a classification role, including a three-layer architecture. The first fully connected layer fits the dataset based on the dropout technique with a degenerate learning rate, which avoids overfitting; the second and third fully connected layers use the Exponential Linear Unit (ELU) activation function to shorten training time and improve accuracy; the output layer obtains the keyness score output value of each node in the network.
[0058] Specifically, the prediction of key nodes in the network involves inputting network feature data, calculating the minimum loss using the regression loss function MSE, and finally obtaining the predicted keyness score for each node in the network.
[0059] Specifically, the selection of key enterprises is based on the upstream, midstream, and downstream network structure of the industrial chain and the main links of the industrial chain. The top N node enterprises in the main links with the highest predicted key scores are extracted as key representative enterprises of each main link, and the ranking weight of key representative enterprises is also determined. It should be noted that the value of N depends on the congestion of the industrial chain and the completeness of the data. This paper prefers N to be 5, but other suitable values are also possible. The ranking weight is between [0, 1] and can be calculated by normalizing the predicted key scores of key enterprises in each main link of the industrial chain using the min-max normalization method.
[0060] Step 3: Establish a three-tiered indicator system for supply chain security evaluation and assess the third-tier indicators of key enterprises. Calculate the scores of the third-tier indicators for major links based on the ranking weight of key enterprises. The three-tiered indicator system for supply chain security evaluation consists of three first-tier indicators: supply chain competitiveness, supply chain control, and supply chain development environment. Specifically, supply chain competitiveness, from the perspective of core capabilities, mainly includes technological innovation capability, profitability, and market competitiveness. At the micro level, it is necessary to enhance the technological innovation capability and technological innovation output capability of core enterprises in key links of the supply chain, optimize and improve the profitability of core enterprises, and improve the competitiveness of core enterprises' main products or services in domestic and international markets. Specifically, supply chain control, from the perspective of major risks... This mainly includes core technologies, industrial capital, and the degree of foreign dependence or foreign control in the market; in particular, the industrial chain development environment mainly includes talent support conditions, external natural and political environment, and domestic and foreign policy environment; the impact of the industrial chain development environment on the security of enterprise supply chains is more due to external uncontrollable factors. Among them, talent support conditions are not limited to a single core enterprise, but focus on examining the talent support conditions of the entire industry under evaluation. Therefore, the score of this indicator is calculated based on industry data; calculating the tertiary indicator score of the main links according to the ranking weight of key enterprises refers to calculating the tertiary indicator score of each main link of the industrial chain by weighted summation based on the ranking weight of each enterprise in the main links and the tertiary indicator score.
[0061] Step 4: Calculate the weight coefficients of each level of the indicator system using the entropy weight method, and combine this with the scores of the third-level indicators of the main links at the industry level to calculate the safety score of the main links according to the weight coefficients of each level of indicators. The steps for calculating the weight coefficients of the indicator system using the entropy weight method include standardizing the third-level indicator data, converting it into dimensionless relative indicators to eliminate dimensional differences between different indicators; constructing an n×m decision matrix based on the standardized data, where n represents the number of indicators and m represents the number of samples; calculating the entropy value for each column of data for each indicator, the entropy value reflecting the information content and degree of difference of the indicator; the formula for calculating the entropy value is: Where p represents the data proportion, k = 1 / ln(n), (0 ≤ e j <1); Calculate the weight coefficient of each indicator based on the entropy value; the formula for calculating the indicator weight coefficient is: w j = (1-E j ) / ∑(1-E j ), where E jThe entropy value of the indicator is used. Based on the hierarchical structure of the indicator system, the weight coefficients of the sub-indicators are multiplied level by level according to the hierarchical structure to obtain the weight coefficients of each level of the indicator system. The safety score of the main links in the industrial chain is calculated layer by layer by weighted summation based on the weight coefficients of each level of the indicator system and the scores of the three-level indicators of each main link in the industrial chain. Here's an example to illustrate how the entropy weight method calculates the indicator weight coefficients. Assume there are three sub-indicators: technological innovation capability, profitability, and market competitiveness. First, the collected data is standardized, transforming it into dimensionless relative indicators between 0 and 1. Then, based on the standardized data, a decision matrix is constructed. Assuming there are 3 samples (enterprises) and 3 sub-indicators, the decision matrix is as follows:
[0062] Technological innovation capability Profitability Market competitiveness Company 1 0.8 0.7 0.6 Company 2 0.6 0.9 0.5 Company 3 0.7 0.8 0.9
[0063] Next, the percentage p of each column of data is calculated by dividing each value by the sum of the data in the column to obtain the percentage value;
[0064] Technological innovation capability Profitability Market competitiveness Company 1 0.8 / 2.1 0.7 / 2.4 0.6 / 2.0 Company 2 0.6 / 2.1 0.9 / 2.4 0.5 / 2.0 Company 3 0.7 / 2.1 0.8 / 2.4 0.9 / 2.0
[0065] Furthermore, based on the entropy calculation formula, the entropy value of each column of data for each sub-index is calculated;
[0066] Technological innovation capability Profitability Market competitiveness Company 1 0.380 0.389 0.315 Company 2 0.442 0.358 0.500 Company 3 0.393 0.372 0.286
[0067] Finally, the weight coefficient of each sub-indicator is further calculated according to the weight coefficient formula;
[0068] Technological innovation capability Profitability Market competitiveness Weighting coefficient 0.294 0.303 0.403
[0069] Step S5: Calculate the weight coefficients of each major link in the industrial chain and calculate the overall security score of the industrial chain based on the weight coefficients of the major links. The weight coefficients of each major link are calculated by multiplying the ranking weights of enterprises in each major link according to the hierarchical structure of the industrial chain. Specifically, the weight coefficients are normalized so that their sum equals 1 to ensure that the relative weights remain unchanged. The overall security score of the industrial chain is calculated by combining the weight coefficients of the major links with the security scores of each major link in the industrial chain, using a weighted summation method to calculate the final security score of the industrial chain.
[0070] In one specific embodiment, such as Figure 2 As shown, the present invention provides a comprehensive assessment and early warning system for supply chain security risks, comprising:
[0071] Data Acquisition and Preprocessing Module 510: This module is responsible for collecting network node attribute data related to the industry chain to be evaluated, such as company name, industry, scale, number of patents, market share, company size, registered capital, financing amount, etc.; network relationship structure data, such as supply relationship, investment relationship, cooperation relationship and competition relationship, etc.; and various indicator data of the indicator system (various sub-data under the three primary indicators of industry chain competitiveness, industry chain control and industry chain development environment), and performing data preprocessing operations on the relevant data.
[0072] Node Keyness Score Prediction Module 520: The node keyness score prediction module is responsible for implementing the GCN-based key node identification model of the industrial chain network. It extracts node features and network structure features from the industrial chain network node data and network relationship structure data, inputs them into the identification model, and outputs the keyness score prediction value of each node in the industrial chain network.
[0073] Key Enterprise Identification Module 530: The Key Enterprise Identification Module is responsible for extracting the top N node enterprises with the highest predicted key scores in each major link based on the upstream, midstream and downstream network structure of the industrial chain and the main links of the industrial chain.
[0074] Key Enterprise Ranking Weight Calculation Module 540: The Key Enterprise Ranking Weight Calculation Module is responsible for calculating the ranking weight of key enterprises in each major link of the industrial chain based on the predicted key scores of key enterprises and normalizing them according to the maximum-minimum standardization method.
[0075] Module 550 for calculating the weight coefficient of major links: This module is responsible for multiplying the ranking weights of enterprises in each major link according to the hierarchical structure of the industrial chain to obtain the weight coefficient of the major links in the industrial chain.
[0076] Risk assessment indicator weight coefficient calculation module 560: The risk assessment indicator weight coefficient calculation module is responsible for calculating the weight coefficient of each level of the indicator system by multiplying the weight coefficients of the sub-indicators according to the hierarchical structure of the indicator system based on the weight coefficient calculation formula of the entropy weight method.
[0077] The main link three-level indicator safety score calculation module 570 is responsible for calculating the three-level indicator safety score of each main link in the industrial chain by combining the ranking weight of each enterprise in the main link with the score of the three-level indicator in a weighted summation manner.
[0078] The main link safety score calculation module 580 is responsible for calculating the safety score of the main link of the industrial chain by combining the weight coefficients of the indicator system at each level with the scores of the three-level indicators of each main link of the industrial chain in a weighted summation manner.
[0079] The supply chain security assessment and early warning module 590 is responsible for calculating the final supply chain security score by combining the weight coefficients of the main links of the supply chain with the security scores of each main link in the supply chain through a weighted summation method.
[0080] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims. It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately. Furthermore, various different embodiments of the present invention can also be arbitrarily combined, as long as they do not violate the spirit of the present invention, they should also be considered as the content disclosed in the present invention.
Claims
1. A comprehensive assessment and early warning method for supply chain security risks, characterized in that, Includes the following steps: Step 1: Construct the upstream, midstream, and downstream network structure of the industrial chain based on the data of the industrial chain to be evaluated, and determine the main links of the industrial chain; Step 2: Use Graph Convolutional Neural Network (GCN) to learn node attributes and network relationship structure features to screen key enterprises in the main links of the industrial chain; Step 3: Establish a three-level indicator system for supply chain security evaluation and assess the third-level indicators of key enterprises. Calculate the scores of the third-level indicators for the main links according to the ranking weight of key enterprises. Step 4: Use the entropy weight method to calculate the weight coefficients of each level of the indicator system and combine them with the scores of the third-level indicators of the main links at the industry level to calculate the safety score of the main links according to the weight coefficients of each level of indicators. Step 5: Calculate the weight coefficients of each major link in the industrial chain and calculate the overall security score of the industrial chain based on the weight coefficients of the major links in the industrial chain. In step 1, the data of the industrial chain to be evaluated includes the basic attribute characteristics of enterprises and the characteristics of the industrial chain network structure. The upstream, midstream and downstream network structure of the industrial chain is that each node in the industrial chain network represents an enterprise or a link, and the edge represents the relationship between enterprises or links. The industrial chain network is constructed using graph theory and network analysis methods. By constructing the industrial chain network, the enterprises and links in the industrial chain are displayed in the form of a graph. Determining the main links of the industrial chain refers to further analyzing the industrial chain network based on the constructed industrial chain network to determine the main links of the industrial chain. In step 2, the screening of key enterprises includes data processing, feature extraction, key node identification model, network key node prediction, and screening of key enterprises. Data processing involves removing a very small number of isolated nodes from the network data, ensuring that all input graphs are connected. Feature extraction involves constructing a subgraph network by extracting a two-degree subgraph for each node in the network. A two-degree subgraph network for each node is constructed by selecting neighboring nodes no more than three degrees away from the target node. Features of the subgraph network are extracted, and a feature vector consisting of five features is constructed for each node. The subgraph network features and the five feature vectors for each node are used as the original input to the model. The five features are degree centrality, betweenness centrality, k-shell decomposition, compact centrality, and node embedding. Degree centrality represents the degree of node i. The formula for the number of neighboring nodes directly connected to this node is: Betweenness centrality represents the load during information propagation in a network; it calculates the shortest path between any two nodes in the network. The more times a node is included in the shortest path, the greater its load and the more important it is. The betweenness of a number is defined as: ; in, This represents the number of shortest paths from point s to point t. The shortest path from point s to point t passes through point i; the k-shell decomposition method indicates that the K-shell value of a node is a score of the node's position in the network; the formula for calculating compact centrality is: ; in, It is the distance between node i and node j; node embedding is compiled using node2vec to preserve the network's structural information, mapping nodes to the embedding space, performing one-hot encoding, and then multiplying by the embedding matrix to obtain the embedding vector of each node; the five features comprehensively consider the network's local, global, and positional attributes. The key node identification model consists of a four-layer network architecture: an input layer, a GCN layer, a fully connected layer, and an output layer. The input layer comprises node feature vectors consisting of subgraph network features and five feature vectors. The GCN layer is defined as follows: ; Where A is the symmetric normalized Laplace operator for a two-degree subgraph network; This represents the node of the i-th GCN layer; and These represent the learnable weights and the error parameters, respectively. It is a non-linear activation function that is used to train the key node identification model of the industrial chain network based on the feature vector extracted from each node. The fully connected layer plays a classification role and includes a three-layer architecture. The first fully connected layer fits the dataset based on the dropout technique with a degenerate learning rate. The second and third fully connected layers use the exponential linear unit (ELU) activation function to shorten the training time and improve the accuracy. The output layer is the keyness score output value of each node in the network.
2. The method for comprehensive assessment and early warning of supply chain security risks as described in claim 1, characterized in that, In step 2, the prediction of key nodes in the network involves inputting network feature data, calculating the minimum loss using the regression loss function MSE, and finally obtaining the predicted key score value for each node in the network. The selection of key enterprises is based on the upstream, midstream and downstream network structure of the industrial chain and the main links of the industrial chain. The top N node enterprises with the highest predicted key scores in the main links are selected as key representative enterprises in each main link, and the ranking weight of key representative enterprises is determined.
3. The comprehensive assessment and early warning method for supply chain security risks as described in claim 1, characterized in that, In step 3, the three-level indicator system for industrial chain security evaluation consists of three primary indicators: industrial chain competitiveness, industrial chain control, and industrial chain development environment. Industrial chain competitiveness includes technological innovation capability, profitability, and market competitiveness. At the micro level, it is necessary to enhance the technological innovation capability and technological innovation output capability of core enterprises in key links of the industrial chain. Industrial chain control includes the external dependence or foreign capital control rate of core technologies, industrial capital, and markets. The industrial chain development environment focuses on examining the talent guarantee conditions of the entire industry under evaluation. Calculating the scores of the three-level indicators of the main links according to the ranking weight of key enterprises refers to calculating the scores of the three-level indicators of each main link of the industrial chain by weighted summation based on the ranking weight of each enterprise in the main links and the scores of the three-level indicators.
4. The method for comprehensive assessment and early warning of supply chain security risks as described in claim 1, characterized in that, In step 4, the entropy weight method for calculating the weight coefficients of the indicator system includes standardizing the third-level indicator data, transforming it into dimensionless relative indicators to eliminate dimensional differences between different indicators; constructing an n×m decision matrix based on the standardized data, where n represents the number of indicators and m represents the number of samples; and calculating the entropy value for each column of data for each indicator. The entropy value reflects the information content and degree of difference of the indicator. The formula for calculating the entropy value is: Where p represents the percentage of data, ; Calculate the weight coefficient for each indicator based on the entropy value; the formula for calculating the indicator weight coefficient is: ,in The entropy value of the indicator is given. According to the hierarchical structure of the indicator system, the weight coefficients of the sub-indicators are multiplied level by level according to the hierarchical structure to obtain the weight coefficients of the indicator system at each level. The safety score of the main links of the industrial chain is calculated layer by layer by weighted summation based on the weight coefficients of the indicator system at each level and the scores of the three-level indicators of the main links of the industrial chain to obtain the safety score of the main links of the industrial chain.
5. The method for comprehensive assessment and early warning of supply chain security risks as described in claim 1, characterized in that, In step 5, the weight coefficients of each major link in the industrial chain are calculated by multiplying the ranking weights of enterprises in each major link according to the hierarchical structure of the industrial chain to obtain the weight coefficients of the major links in the industrial chain. The weight coefficients are normalized so that their sum equals 1 to ensure that the relative relationship of the weights remains unchanged. The total security score of the entire industrial chain is calculated by combining the weight coefficients of the major links in the industrial chain with the security scores of each major link in the industrial chain and calculating the final security score of the industrial chain layer by layer in a weighted summation manner.
6. A comprehensive industrial chain security risk assessment and early warning system for implementing the comprehensive industrial chain security risk assessment and early warning method as described in claim 1, characterized in that, include: Data Acquisition and Preprocessing Module: The data acquisition and preprocessing module is used to collect network node attribute data, network relationship structure data, and indicator data of the indicator system related to the industrial chain to be evaluated, and to perform data preprocessing operations on the relevant data; Node Keyness Score Prediction Module: The node keyness score prediction module is used to implement the GCN-based key node identification model of the industrial chain network. It extracts node features and network structure features from the industrial chain network node data and network relationship structure data, inputs them into the identification model, and outputs the keyness score prediction value of each node in the industrial chain network. Key Enterprise Identification Module: Based on the upstream, midstream and downstream network structure of the industrial chain and the main links of the industrial chain, the key enterprise identification module extracts the top N node enterprises with the highest predicted key scores in the main links as key representative enterprises in each main link; Key Enterprise Ranking Weight Calculation Module: The key enterprise ranking weight calculation module calculates the ranking weight of key enterprises in each major link of the industrial chain based on the predicted criticality scores of key enterprises and normalizes them according to the maximum-minimum standardization method. The module for calculating the weight coefficient of key links: Based on the hierarchical structure of the industrial chain, the module for calculating the weight coefficient of key links calculates the weight of each key link by multiplying the weight of each key link by the weight of each key link according to the hierarchical structure and obtains the weight coefficient of key links in the industrial chain. Risk assessment indicator weight coefficient calculation module: The risk assessment indicator weight coefficient calculation module is used to calculate the weight coefficient of each level of the indicator system by multiplying the weight coefficients of the sub-indicators according to the hierarchical structure of the indicator system based on the weight coefficient calculation formula of the entropy weight method. The module for calculating the safety score of the three-level indicators for key links: This module calculates the safety score of the three-level indicators for each key link in the industrial chain by combining the ranking weight of each enterprise in the key links with the scores of the three-level indicators in a weighted summation manner. The safety score calculation module for key links: The safety score calculation module for key links calculates the safety score of key links in the industrial chain by combining the weight coefficients of the indicator system at each level with the scores of the three-level indicators of each key link in the industrial chain in a weighted summation manner. The supply chain security assessment and early warning module calculates the final supply chain security score by combining the weight coefficients of the main links in the supply chain with the security scores of each main link in the supply chain through a weighted summation method.
7. The supply chain security risk comprehensive assessment and early warning system as described in claim 6, characterized in that, The network node attribute data includes enterprise name, industry, size, number of patents, market share, company size, registered capital, and financing amount; the network relationship structure data includes supply relationship, investment relationship, cooperation relationship, and competition relationship.
Citation Information
Patent Citations
Coal industry chain risk monitoring system and monitoring method thereof
CN111898916A
Risk early warning method, system and equipment for energy power industry chain and medium
CN114493078A
National defense industry chain assessment method based on risk index assessment
CN114580970A
Risk identification and evaluation method based on complex product supplier supply chain
CN116070906A
