Graph theory analysis-based intelligent identification method for stock equity relationship of power grid project suppliers

Abstract

The invention discloses a graph theory analysis-based power grid project supplier equity relationship intelligent identification method. The method comprises the following steps of S1, collecting supplier relationship data; s2, stock equity structure features and transaction features of suppliers in the standardized data set are extracted; s3, constructing a supplier equity relation graph, traversing the supplier equity relation graph by adopting a depth-first search algorithm and a breadth-first search algorithm, and calculating a control path length and an equity control weight; s4, optimizing the stock right control relation identification of the suppliers by using an improved Grapporter network, and predicting the stock right control relation strength between the suppliers; s5, calculating the risk scores of the suppliers, setting a threshold value according to the risk scores, and classifying the risk levels of the suppliers; and S6, an incremental learning method is adopted to optimize the Grapher network. According to the method, graph theory analysis and Grapher network improvement are combined, the supplier stock equity control relation is intelligently identified, the risk is dynamically evaluated, and the method has the advantages of being accurate in identification, high in adaptability and high in real-time performance.

Description

Technical Field

[0001] The present invention relates to the field of power grid project supply chain management and risk identification, and in particular to a method for intelligently identifying equity relationships of power grid project suppliers based on graph theory analysis. Background Art

[0002] In the bidding and supply chain management of power grid engineering projects, supplier qualification review and equity relationship identification are crucial steps in ensuring fair competition and reducing supply chain risks. Traditional supplier equity relationship analysis relies primarily on manual verification or simple database queries. Companies use industrial and commercial information query platforms, annual reports, shareholder structure, and equity change records to preliminarily screen supplier equity relationships. However, this approach presents multiple issues: data fragmentation, difficulty in penetrating identification, high labor costs, and the inability to dynamically update. This makes it difficult to meet the requirements of modern power grid engineering supply chains for supplier equity structure compliance analysis.

[0003] In existing technologies, the identification of supplier equity relationships mainly relies on the comparison of supplier industrial and commercial data and the analysis of transaction relationships. For example, some companies compare the supplier's shareholder list to find out whether there is cross-holding, and analyze the business dealings between suppliers based on supply chain transaction data to determine whether there is equity relationship or potential interest transfer behavior. However, the existing query method can play a certain role in identifying suppliers with a relatively simple equity structure. However, when faced with complex equity penetration relationships, hidden shareholding structures and hierarchical cross-control relationships, the identification ability of traditional methods is very limited, and it is impossible to deeply analyze the ultimate controller behind the supplier and its influence scope. In addition, suppliers often avoid direct equity relationships by establishing multi-level holding companies, cross-holdings, shareholder proxy holding, etc., making it difficult for traditional shareholder information comparison methods to effectively identify hidden related parties, resulting in certain compliance risks in the selection and management of suppliers for power grid engineering projects.

[0004] Another problem is that most existing supplier equity relationship analysis methods are static. Even if shareholder change information is obtained through industrial and commercial databases, it is difficult to dynamically update the supplier control relationship in real time. In actual applications, the supplier's shareholder structure, controlling weight and transaction relationship may change in a relatively short period of time, such as equity transfer, legal person change, business adjustment, etc. The existing static analysis model cannot meet the needs of supervision and risk assessment, and is likely to lead to a lag in the judgment of supplier control relationships. Especially in power grid engineering projects, the supplier's bidding activities are highly complex. If the supplier control relationship data cannot be updated in a timely manner, it may cause suppliers to circumvent bidding rules through hidden control relationships, increasing the risk of supply chain management.

[0005] In addition, the risk assessment methods used in existing technologies are relatively simple, usually relying only on the supplier's industrial and commercial data and financial information, and lacking comprehensive analysis of multi-dimensional data such as the supplier's equity control strength, transaction frequency, and bidding cooperation relationships. This makes the risk assessment results lack accuracy and difficult to comprehensively measure the supplier's equity control risk. In the fields of supply chain finance, bidding management, and supplier compliance review, traditional methods are usually based on simple rule matching or manual experience judgment, and fail to fully utilize artificial intelligence and graph theory analysis methods to deeply model supplier relationships, resulting in low identification accuracy of supplier equity control relationships and inability to effectively warn suppliers of potential compliance risks.

[0006] Based on the above problems, in recent years, some studies have begun to try to introduce graph analysis and artificial intelligence technology to improve the accuracy of identifying supplier equity relationships. For example, some studies construct enterprise association networks based on knowledge graphs and use network analysis methods to identify direct or indirect control relationships between suppliers. However, traditional graph analysis methods usually only focus on a single network structure feature, such as using depth-first search or breadth-first search to traverse the equity relationship network to identify the ultimate controller of the supplier and its equity control path, but ignore dynamic factors such as transaction data and bidding cooperation relationships, making it difficult to comprehensively evaluate the actual control weight of the supplier. In addition, traditional graph theory methods lack consideration of the time dimension and fail to combine historical risk data to dynamically optimize the identification and risk assessment model of supplier equity relationships, resulting in analysis results that are difficult to adapt to changes in supplier equity relationships. Summary of the Invention

[0007] One purpose of the present invention is to propose a method for intelligently identifying equity relationships among power grid project suppliers based on graph theory analysis. The present invention combines graph theory analysis with an improved Graphormer network to construct a supplier equity relationship graph, intelligently identify the control relationships and ultimate controllers among suppliers, and dynamically evaluate supplier risk levels. The method traverses the supplier network through depth-first search and breadth-first search, accurately calculates control paths and control weights, optimizes the accuracy of equity control relationship identification, and adopts an incremental learning strategy to optimize the Graphormer network to ensure identification accuracy after supplier data is updated.

[0008] According to an embodiment of the present invention, a method for intelligently identifying equity relationships of power grid project suppliers based on graph theory analysis includes the following steps:

[0009] S1. Collect supplier relationship data, set the collection frequency, automatically run the collection program, and pre-process the collected data to generate a standardized data set;

[0010] S2. Extract the equity structure characteristics and transaction characteristics of suppliers in the standardized dataset, and calculate the shareholding ratio, equity concentration, transaction amount, transaction frequency, and bidding cooperation relationship. Store the data in the supplier equity relationship database and index it according to the supplier's unique identifier;

[0011] S3. Build a supplier equity relationship graph based on the supplier equity relationship database. Use depth-first search and breadth-first search algorithms to traverse the supplier equity relationship graph, identify the equity control relationships and ultimate controllers between suppliers, and calculate the control path length and equity control weight.

[0012] S4. Optimize the identification of supplier equity control relationships using an improved Graphormer network. Construct an improved Graphormer network. Train the improved Graphormer network based on supplier equity relationship data to learn equity control relationship patterns and predict the strength of equity control relationships between suppliers.

[0013] S5. Calculate the supplier's risk score, set thresholds based on the risk score, and classify the supplier's risk level;

[0014] S6. Use incremental learning methods to optimize the Graphormer network, update the supplier equity relationship graph based on historical risk data, and update the supplier equity relationship database.

[0015] Optionally, the supplier relationship data includes supplier business registration information, shareholder structure, corporate annual reports, equity change records, historical bidding information and supply chain transaction data, and the preprocessing includes format conversion, data deduplication, outlier detection, missing data filling, time synchronization and normalization.

[0016] Optionally, the S2 specifically includes:

[0017] S21. Extract suppliers' business registration information, shareholder structure, annual reports, equity change records, historical bidding information, and supply chain transaction data from standardized data sets, and associate the data based on the supplier's unique identifier to ensure complete matching of all types of data for the same supplier.

[0018] S22. Calculate the supplier's shareholding ratio;

[0019] S23. Calculate the supplier's equity concentration by sorting the shareholders based on their shareholding ratios, extract the shareholding ratios of the top N shareholders, and sum them up;

[0020] S24. Calculate the supplier's transaction amount, extract the supplier's annual transaction records based on the supply chain transaction data, calculate the supplier's annual total transaction amount and annual transaction number, and calculate the supplier's transaction frequency;

[0021] S25. Calculate suppliers’ bidding partnerships and, based on historical bidding data, count suppliers’ joint bidding in different projects.

[0022] S26. Build a supplier equity relationship database, store the calculated shareholding ratio, equity concentration, transaction amount, transaction frequency and bidding cooperation relationship in the database, and index it according to the supplier's unique identifier.

[0023] Optionally, the S3 specifically includes:

[0024] S31. Based on the data in the supplier equity relationship database, construct a supplier equity relationship graph, wherein the supplier equity relationship graph comprises suppliers, shareholders, and ultimate controllers as nodes, equity holding relationships, transaction relationships, and bidding cooperation relationships as edges, and weight values ​​are set for the edges;

[0025] S32. Use a depth-first search algorithm to traverse the supplier equity relationship graph, starting with the supplier as the starting node, and search upward along the shareholder shareholding path until reaching the ultimate controller, and calculate the control path length. The depth-first search algorithm traversal process includes:

[0026] Select the starting supplier node S0, mark it as visited, and push it onto the stack;

[0027] From the current stack top node S t Select an unvisited direct shareholder node S t+1 ;

[0028] If S t+1 If it exists, mark S t+1 If it has been visited, push it onto the stack and continue searching upwards;

[0029] If S t+1 If it does not exist, go back to the previous node S t-1 , find unvisited direct shareholder nodes and repeat the selection and search operations;

[0030] When all shareholder nodes have been visited and backtracking to the starting supplier node S0 with no unvisited nodes, the search ends and the control path length is calculated;

[0031] S33. Calculate the supplier's equity control weight based on the control path length using a depth-first search algorithm;

[0032] S34. Using a breadth-first search algorithm to traverse the supplier equity relationship graph, starting with the ultimate controller as the starting node, searching downward along the shareholder shareholding path to identify the ultimate controller's controlled suppliers. The breadth-first search algorithm traversal process includes:

[0033] Select the starting controller node C0, mark it as visited, and add it to the queue;

[0034] From the current queue head node C t Select all directly controlled supplier nodes C t+1 ;

[0035] C t+1 The unvisited nodes are added to the queue and marked as visited;

[0036] Continue to take the next node from the head of the queue and repeat the selection and addition operations until all supplier nodes have been visited;

[0037] Calculate the scope of control of suppliers, count the number of suppliers directly or indirectly influenced by each controller, and calculate the weight of controlled suppliers;

[0038] S35. Calculate the preliminary equity control relationship strength between suppliers based on the supplier's equity control weight and the transaction amount between suppliers.

[0039] Optionally, the S4 specifically includes:

[0040] S41. Construct an improved Graphormer network and input supplier equity relationship data. The improved Graphormer network includes an input layer, an improved graph structure encoding layer, an adaptive multi-head attention mechanism, a global relationship modeling layer, and a fully connected classification layer. The supplier equity relationship data includes supplier equity structure characteristics, transaction characteristics, and control path data. The supplier equity structure characteristics include shareholding ratio, shareholder level, and equity concentration. The transaction characteristics include transaction amount, transaction frequency, and bidding cooperation relationship. The control path data includes control path length, equity control weight, and ultimate controller identifier.

[0041] S42. Based on the supplier equity relationship data in the improved graph structure encoding layer, the improved Graphormer network is used to identify the supplier equity control relationship, the improved Graphormer network is trained, the equity control relationship model is learned, and the supplier equity relationship feature vector H is set. (l) :

[0042] H (l+1) =σ(D -1 / 2 AD -1 / 2 H (l) W (l) +λdiag(I)H (l) );

[0043] Among them, H (l+1) represents the supplier equity relationship feature vector of the l+1 layer, σ represents the nonlinear activation function, H(l) represents the eigenvector of the supplier equity relationship at level l, D represents the degree matrix of the adjacency matrix, A represents the adjacency matrix of the supplier equity relationship graph, and W (l) represents the training weight matrix of the lth layer, λ represents the adaptive adjustment parameter, I represents the importance vector of the supplier node, and diag represents the diagonalization of the vector;

[0044] S43. In the adaptive multi-head attention mechanism, the attention weights between suppliers are calculated and the feature vectors are weighted updated to optimize the prediction of supplier control relationships:

[0045]

[0046] Among them, Z represents the final multi-head attention output vector, Concat represents the concatenation of the outputs of each attention head, H represents the number of attention heads, Softmax represents normalization, and d k represents the dimension of the key vector, Q h , K h and V h denote the query, key, and value vectors of the supplier feature vector, γ, δ, and η denote weight coefficients, and T i,j 、 and denote the normalized values ​​of the transaction amount, control path length, and equity control weight between supplier i and supplier j, respectively. T denotes the transpose operation of the vector;

[0047] S44. At the global relationship modeling layer, the strength of the equity control relationship between suppliers is calculated based on the multi-head attention output vector:

[0048]

[0049] Among them, P i,j represents the strength of the equity control relationship between supplier i and supplier j, σ represents the nonlinear activation function, and W p and W h Represents the training weight matrix, Z i and Z j Represent the feature vectors of supplier i and supplier j after attention mechanism update, represents the normalized value of the initial equity control relationship strength, represents the normalized value of the bidding cooperation relationship, α and β represent the weight coefficients;

[0050] S45, based on the equity control relationship strength P between suppliers i,j, optimize the prediction of supplier equity control relationships at the fully connected classification layer, set a set of equity control relationship categories, calculate the probability distribution of equity control relationship categories between suppliers, and define a loss function. The loss function uses a weighted cross entropy loss function to assign higher weights to high-risk suppliers:

[0051]

[0052] Where L represents the loss function, O represents the total number of supplier relationship pairs, K represents the total number of equity control relationship categories, ω k represents the weight of category k, y i,j,k Represents the true category label, using one-hot encoding, Represents the predicted category probability, P i,j represents the strength of the equity control relationship between supplier i and supplier j, exp represents the natural exponential function, and W c and b c Represents the training weight matrix and bias of the fully connected layer, and k′ represents the category;

[0053] S46, according to the equity control relationship strength P i,j , calculate the changing trend of equity control weights among suppliers, and adjust the supplier equity relationship database based on time series analysis to optimize the supplier equity control hierarchy structure;

[0054] S47. Update the supplier's equity control hierarchy based on the adjusted equity control weights.

[0055] Optionally, the S6 specifically includes:

[0056] S61. Setting an optimization objective function, training an improved Graphormer network based on historical risk data, wherein the historical risk data includes historical supplier equity structure data, transaction data, and control path data, and constructing a historical risk data matrix;

[0057] S62. Based on the historical risk data matrix, an incremental learning strategy is used to update the parameters of the Graphormer network to adapt to the dynamically changing supplier equity control relationship. The incremental learning objective function is defined as:

[0058]

[0059] in, represents the incremental learning objective function, represents the predicted probability of supplier equity control relationship category, y i,j represents the true label, ω j represents the category weight, W prevRepresents the parameters of the Graphormer network obtained in the previous round of training, W represents the parameters of the Graphormer network, ψ represents the regularization parameter, which is used to control the update amplitude of the parameters, ∥∥ 2 represents the norm operation;

[0060] S63. Based on the Graphormer network optimized by incremental learning, the supplier equity control relationship strength is recalculated, and the risk score is calculated based on the optimized equity control relationship strength, the supplier's risk classification is adjusted, and the supplier equity relationship database is updated.

[0061] The beneficial effects of the present invention are:

[0062] First of all, the present invention can break through the limitations of traditional supplier equity relationship identification methods and realize in-depth analysis of complex supplier control relationships. The existing technology relies on manual query of industrial and commercial information or simple database comparison, which makes it difficult to effectively identify the invisible control relationship between suppliers, especially multi-layer equity nesting, cross-holding and identification of ultimate controllers. The present invention is based on graph theory analysis method to construct a supplier equity relationship graph, with suppliers, shareholders and ultimate controllers as nodes, equity holding relationships, transaction relationships and bidding cooperation relationships as edges, and uses depth-first search and breadth-first search algorithms to traverse the supplier equity network, accurately identify the control path between suppliers, calculate the control path length and control weight, thereby effectively exploring the complex control relationship between suppliers and improving the transparency of supply chain management.

[0063] Secondly, the present invention combines artificial intelligence technology to improve the intelligence level of supplier control relationship identification. By constructing an improved Graphormer network, a graph structure encoding layer, a multi-head attention mechanism and a global relationship modeling layer are introduced on the basis of the traditional graph neural network, making the identification process of supplier control relationships more accurate. The Graphormer network can not only learn the characteristics of the supplier's equity structure, but also comprehensively consider transaction characteristics and control path information to optimize the prediction accuracy of control relationships. Compared with traditional static rule matching methods, this method can automatically mine supplier control patterns and maintain efficient recognition capabilities in complex equity structures and dynamic transaction environments.

[0064] In addition, the present invention adopts a dynamic risk assessment mechanism to make up for the lag of traditional supplier compliance management methods. Most existing supplier risk assessment methods are based on static industrial and commercial data, which makes it difficult to reflect real-time changes in the supplier's equity structure. This method calculates the supplier's risk score, comprehensively considers the supplier's equity concentration, control chain complexity, transaction anomalies and equity change frequency, and sets thresholds based on the risk score to classify the supplier's risk level, providing a more targeted risk assessment solution. It can not only identify high-risk suppliers, but also predict the trend of changes in the supplier's equity relationship, discover possible risk points in advance, and improve the security and compliance of the power grid project supply chain.

[0065] Finally, the present invention also adopts an incremental learning strategy to make the supplier equity relationship analysis more adaptable and scalable. With the continuous updating of supplier data, the supplier equity control relationship may change at any time, while traditional methods usually require the reconstruction of the complete supplier relationship database, which has high computational costs and is difficult to meet the needs of real-time analysis. This method optimizes the Graphormer network through an incremental learning strategy, continuously trains the model using historical risk data, and dynamically adjusts the weight parameters for identifying supplier equity control relationships, thereby ensuring the accuracy and timeliness of the analysis results. Compared with traditional static analysis methods, this method can adapt to the dynamic changes in the equity relationships of power grid project suppliers without affecting the performance of existing models, thereby improving the real-time and accuracy of supplier management. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0067] Figure 1 This is a flow chart of a method for intelligently identifying equity relationships of power grid project suppliers based on graph theory analysis proposed by the present invention. DETAILED DESCRIPTION

[0068] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0069] refer to Figure 1 , an intelligent identification method for equity relations of power grid project suppliers based on graph theory analysis, comprising the following steps:

[0070] S1. Collect supplier relationship data, set the collection frequency, automatically run the collection program, and pre-process the collected data to generate a standardized data set;

[0071] S2. Extract the equity structure characteristics and transaction characteristics of suppliers in the standardized dataset, and calculate the shareholding ratio, equity concentration, transaction amount, transaction frequency, and bidding cooperation relationship. Store the data in the supplier equity relationship database and index it according to the supplier's unique identifier;

[0072] S3. Build a supplier equity relationship graph based on the supplier equity relationship database. Use depth-first search and breadth-first search algorithms to traverse the supplier equity relationship graph, identify the equity control relationships and ultimate controllers between suppliers, and calculate the control path length and equity control weight.

[0073] S4. Optimize the identification of supplier equity control relationships using an improved Graphormer network. Construct an improved Graphormer network. Train the improved Graphormer network based on supplier equity relationship data to learn equity control relationship patterns and predict the strength of equity control relationships between suppliers.

[0074] S5. Calculate the supplier's risk score, set thresholds based on the risk score, and classify the supplier's risk level;

[0075] S6. Use incremental learning methods to optimize the Graphormer network, update the supplier equity relationship graph based on historical risk data, and update the supplier equity relationship database.

[0076] The present invention constructs an intelligent identification method for supplier equity relationships based on graph theory analysis, which can automatically and efficiently identify equity control relationships between suppliers, analyze the ultimate controller, and calculate the control path length and control weight. It traverses the supplier equity relationship graph through depth-first search and breadth-first search algorithms, improving the ability to parse complex equity structures. At the same time, combined with the improved Graphormer network to optimize the prediction of equity control relationships, the identification of supplier control relationships is made more accurate. The incremental learning method is used to continuously optimize the supplier equity relationship database, ensuring the dynamic updating of analysis results, improving the intelligent level of supply chain management, and enhancing supplier compliance management and risk prevention and control capabilities.

[0077] In this embodiment, the supplier relationship data includes supplier business registration information, shareholder structure, annual report, equity change records, historical bidding information and supply chain transaction data. The preprocessing includes format conversion, data deduplication, outlier detection, missing data filling, time synchronization and normalization.

[0078] The present invention optimizes the collection and preprocessing of supplier relationship data, combining industrial and commercial registration information, shareholder structure, corporate annual reports, equity change records, historical bidding information and supply chain transaction data to ensure the comprehensiveness of data sources. It adopts preprocessing steps such as format conversion, data deduplication, outlier detection, missing data filling, time synchronization and normalization to improve data quality, reduce noise and redundant data, and enhance the accuracy of subsequent equity analysis. By constructing a standardized data set, the integrity and consistency of the supplier equity relationship database are improved, laying the foundation for subsequent supplier equity analysis and control relationship identification.

[0079] In this embodiment, S2 specifically includes:

[0080] S21. Extract suppliers' business registration information, shareholder structure, annual reports, equity change records, historical bidding information, and supply chain transaction data from standardized data sets, and associate the data based on the supplier's unique identifier to ensure complete matching of all types of data for the same supplier.

[0081] S22. Calculate the supplier's shareholding ratio;

[0082] S23. Calculate the supplier's equity concentration by sorting the shareholders based on their shareholding ratios, extract the shareholding ratios of the top N shareholders, and sum them up;

[0083] S24. Calculate the supplier's transaction amount, extract the supplier's annual transaction records based on the supply chain transaction data, calculate the supplier's annual total transaction amount and annual transaction number, and calculate the supplier's transaction frequency;

[0084] S25. Calculate suppliers’ bidding partnerships and, based on historical bidding data, count suppliers’ joint bidding in different projects.

[0085] S26. Build a supplier equity relationship database, store the calculated shareholding ratio, equity concentration, transaction amount, transaction frequency and bidding cooperation relationship in the database, and index it according to the supplier's unique identifier.

[0086] The present invention optimizes the extraction of supplier data features. By calculating the shareholding ratio, equity concentration, transaction amount, transaction frequency and bidding cooperation relationship, a supplier equity relationship database is constructed and indexed according to the unique identifier to ensure data traceability and consistency. Through association analysis and feature calculation, the supplier's equity relationship is not limited to direct shareholding, but implicit control relationships can be identified in multi-level shareholding relationships. By statistics on supplier bidding cooperation behavior, the supplier's risk assessment is further optimized, preventing suppliers from making joint bids through affiliated companies, and improving the fairness and transparency of supply chain management.

[0087] In this embodiment, S3 specifically includes:

[0088] S31. Based on the data in the supplier equity relationship database, construct a supplier equity relationship graph, wherein the supplier equity relationship graph comprises suppliers, shareholders, and ultimate controllers as nodes, equity holding relationships, transaction relationships, and bidding cooperation relationships as edges, and weight values ​​are set for the edges;

[0089] S32. Use a depth-first search algorithm to traverse the supplier equity relationship graph, starting with the supplier as the starting node, and search upward along the shareholder shareholding path until reaching the ultimate controller, and calculate the control path length. The depth-first search algorithm traversal process includes:

[0090] Select the starting supplier node S0, mark it as visited, and push it onto the stack;

[0091] From the current stack top node S t Select an unvisited direct shareholder node S t+1 ;

[0092] If S t+1 If it exists, mark S t+1 If it has been visited, push it onto the stack and continue searching upwards;

[0093] If S t+1 If it does not exist, go back to the previous node S t-1 , find unvisited direct shareholder nodes and repeat the selection and search operations;

[0094] When all shareholder nodes have been visited and backtracking to the starting supplier node S0 with no unvisited nodes, the search ends and the control path length is calculated;

[0095] S33. Calculate the supplier's equity control weight based on the control path length using a depth-first search algorithm;

[0096] S34. Using a breadth-first search algorithm to traverse the supplier equity relationship graph, starting with the ultimate controller as the starting node, searching downward along the shareholder shareholding path to identify the ultimate controller's controlled suppliers. The breadth-first search algorithm traversal process includes:

[0097] Select the starting controller node C0, mark it as visited, and add it to the queue;

[0098] From the current queue head node C t Select all directly controlled supplier nodes C t+1 ;

[0099] C t+1 The unvisited nodes are added to the queue and marked as visited;

[0100] Continue to take the next node from the head of the queue and repeat the selection and addition operations until all supplier nodes have been visited;

[0101] Calculate the scope of control of suppliers, count the number of suppliers directly or indirectly influenced by each controller, and calculate the weight of controlled suppliers;

[0102] S35. Calculate the preliminary equity control relationship strength between suppliers based on the supplier's equity control weight and the transaction amount between suppliers.

[0103] The present invention realizes the all-round identification of the control relationship between suppliers by constructing a supplier equity relationship diagram and combining the depth-first search and breadth-first search algorithms. Through the depth-first search, the shareholder shareholding path is traversed upward to calculate the supplier's control path length and equity control weight, accurately identifying the ultimate controller. Through the breadth-first search, the controlled suppliers are identified by traversing downward from the ultimate controller, and the influence of the supplier in the entire equity network is analyzed. Through a comprehensive analysis of the supplier's transaction amount, control weight, etc., the preliminary equity control relationship strength between suppliers is calculated, providing a more accurate basis for risk warning of the supply chain.

[0104] In this embodiment, the S4 specifically includes:

[0105] S41. Construct an improved Graphormer network and input supplier equity relationship data. The improved Graphormer network includes an input layer, an improved graph structure encoding layer, an adaptive multi-head attention mechanism, a global relationship modeling layer, and a fully connected classification layer. The supplier equity relationship data includes supplier equity structure characteristics, transaction characteristics, and control path data. The supplier equity structure characteristics include shareholding ratio, shareholder level, and equity concentration. The transaction characteristics include transaction amount, transaction frequency, and bidding cooperation relationship. The control path data includes control path length, equity control weight, and ultimate controller identifier.

[0106] S42. Based on the supplier equity relationship data in the improved graph structure encoding layer, the improved Graphormer network is used to identify the supplier equity control relationship, the improved Graphormer network is trained, the equity control relationship model is learned, and the supplier equity relationship feature vector H is set. (l) :

[0107] H (l+1) =σ(D -1 / 2 AD -1 / 2 H (l) W (l) +λdiag(I)H (l) );

[0108] Among them, H(l+1) represents the supplier equity relationship feature vector of the l+1 layer, σ represents the nonlinear activation function, H (l) represents the eigenvector of the supplier equity relationship at level l, D represents the degree matrix of the adjacency matrix, A represents the adjacency matrix of the supplier equity relationship graph, and W (l) represents the training weight matrix of the lth layer, λ represents the adaptive adjustment parameter, I represents the importance vector of the supplier node, and diag represents the diagonalization of the vector;

[0109] S43. In the adaptive multi-head attention mechanism, the attention weights between suppliers are calculated and the feature vectors are weighted updated to optimize the prediction of supplier control relationships:

[0110]

[0111] Among them, Z represents the final multi-head attention output vector, Concat represents the concatenation of the outputs of each attention head, H represents the number of attention heads, Softmax represents normalization, and d k represents the dimension of the key vector, Q h , K h and V h denote the query, key, and value vectors of the supplier feature vector, γ, δ, and η denote weight coefficients, and T i,j 、 and denote the normalized values ​​of the transaction amount, control path length, and equity control weight between supplier i and supplier j, respectively. T denotes the transpose operation of the vector;

[0112] S44. At the global relationship modeling layer, the strength of the equity control relationship between suppliers is calculated based on the multi-head attention output vector:

[0113]

[0114] Among them, P i,j represents the strength of the equity control relationship between supplier i and supplier j, σ represents the nonlinear activation function, and W p and W h Represents the training weight matrix, Z i and Z j Represent the feature vectors of supplier i and supplier j after attention mechanism update, represents the normalized value of the initial equity control relationship strength, represents the normalized value of the bidding cooperation relationship, α and β represent the weight coefficients;

[0115] S45, based on the equity control relationship strength P between suppliers i,j, optimize the prediction of supplier equity control relationships at the fully connected classification layer, set a set of equity control relationship categories, calculate the probability distribution of equity control relationship categories between suppliers, and define a loss function. The loss function uses a weighted cross entropy loss function to assign higher weights to high-risk suppliers:

[0116]

[0117] Where L represents the loss function, O represents the total number of supplier relationship pairs, K represents the total number of equity control relationship categories, ω k represents the weight of category k, y i,j,k Represents the true category label, using one-hot encoding, Represents the predicted category probability, P i,j represents the strength of the equity control relationship between supplier i and supplier j, exp represents the natural exponential function, and W c and b c Represents the training weight matrix and bias of the fully connected layer, and k′ represents the category;

[0118] S46, according to the equity control relationship strength P i,j , calculate the changing trend of equity control weights among suppliers, and adjust the supplier equity relationship database based on time series analysis to optimize the supplier equity control hierarchy structure;

[0119] S47. Update the supplier's equity control hierarchy based on the adjusted equity control weights.

[0120] The present invention utilizes an improved Graphormer network to optimize the identification of supplier equity control relationships. On the basis of traditional graph neural networks, it introduces a graph structure encoding layer, an adaptive multi-head attention mechanism, and a global relationship modeling layer, making the identification of supplier control relationships more robust and accurate. Through the Graphormer network, the supplier's equity structure characteristics, transaction characteristics, and control path information can be integrated to improve the accuracy of supplier control relationship prediction. Combined with the multi-head attention mechanism, the feature representation of supplier control relationships is optimized, and the ability to identify implicit control relationships is improved. The probability distribution of equity control relationship categories between suppliers is calculated through the fully connected classification layer, thereby improving the accuracy of supplier risk assessment.

[0121] In this embodiment, S6 specifically includes:

[0122] S61. Setting an optimization objective function, training an improved Graphormer network based on historical risk data, wherein the historical risk data includes historical supplier equity structure data, transaction data, and control path data, and constructing a historical risk data matrix;

[0123] S62. Based on the historical risk data matrix, an incremental learning strategy is used to update the parameters of the Graphormer network to adapt to the dynamically changing supplier equity control relationship. The incremental learning objective function is defined as:

[0124]

[0125] in, represents the incremental learning objective function, represents the predicted probability of supplier equity control relationship category, y i,j represents the true label, ω j represents the category weight, W prev Represents the parameters of the Graphormer network obtained in the previous round of training, W represents the parameters of the Graphormer network, ψ represents the regularization parameter, which is used to control the update amplitude of the parameters, ∥∥ 2 represents the norm operation;

[0126] S63. Based on the Graphormer network optimized by incremental learning, the supplier equity control relationship strength is recalculated, and the risk score is calculated based on the optimized equity control relationship strength, the supplier's risk classification is adjusted, and the supplier equity relationship database is updated.

[0127] This paper uses an incremental learning method to optimize the Graphormer network. Based on historical risk data, it continuously optimizes the supplier equity control relationship identification model, ensuring that the identification of supplier control relationships can adapt to the dynamically changing supply chain environment. By calculating the supplier's risk score and combining equity concentration, control chain complexity, transaction anomalies, and equity change frequency, the supplier's risk level is dynamically assessed. Based on the optimized control relationship strength, the supplier risk classification is adjusted in real time, improving the intelligent level of supplier risk management. By updating the supplier equity relationship database, the timeliness of the data is ensured, enhancing the security and compliance of supply chain management.

[0128] Example 1:

[0129] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the supplier management system of a State Grid Corporation's 2023 power transmission and transformation equipment procurement project. The project covers multiple provinces across the country and involves a total of 432 suppliers, among which there are a large number of cross-holdings, implicit control relationships and potential joint bidding risks. The traditional supplier review method mainly relies on manual comparison of corporate industrial and commercial information, historical bidding records and supply chain transaction data. The review cycle is long, and it is difficult to accurately identify the implicit control relationship between suppliers, resulting in some suppliers repeating bidding through affiliated companies, affecting market fairness.

[0130] In this project, the method of the present invention is integrated into the supplier compliance management system to identify the equity control relationship between suppliers and dynamically evaluate the risk level of suppliers. This method first connects with the National Enterprise Credit Information Publicity System, the stock exchange database, and the third-party enterprise information query platform through a data interface to obtain the industrial and commercial registration information, shareholder structure, annual report, equity change record, historical bidding information and supply chain transaction data of all suppliers. The system sets the collection frequency to once a day, and after collection, the data is formatted, deduplicated, detected for outliers, filled with missing data, time synchronized and normalized to ensure data quality. The preprocessed data is stored in the supplier equity relationship database and indexed according to the supplier's unique identifier to form a standardized data set.

[0131] After the data is collated, the method of the present invention begins to extract and calculate the supplier's equity structure characteristics and transaction characteristics. The system automatically calculates the supplier's shareholding ratio, equity concentration, transaction amount, transaction frequency and bidding cooperation relationship. In the process of constructing the supplier equity relationship diagram, the present invention uses depth-first search and breadth-first search algorithms to identify the supplier's control relationship.

[0132] To further improve the accuracy of control relationship identification, the present invention uses an improved Graphormer network to optimize supplier control relationships. The Graphormer network is trained based on the supplier's equity structure, transaction characteristics, and control paths, automatically learning the equity control model and predicting the strength of control relationships between suppliers.

[0133] During the risk assessment stage, the present invention calculates the supplier's risk score based on the calculated control path and control relationship strength, combined with characteristics such as the degree of transaction abnormality and the frequency of equity changes, and classifies it according to risk thresholds. The risk assessment results are updated in real time to the supplier management system and synchronized to the power grid company's procurement department for decision-making.

[0134] Table 1 Experimental comparison data table

[0135]

[0136]

[0137] From the comparison results of experimental data, it can be seen that the method of the present invention shows significant advantages over traditional methods in supplier equity relationship identification and risk assessment.

[0138] First, in terms of processing time, traditional manual review takes about 240 hours to complete, and the traditional rule-matching-based method takes 72 hours. However, the method of the present invention only takes 2.5 hours to complete the entire supplier equity control relationship identification process, greatly improving the review efficiency.

[0139] In terms of recognition accuracy, the accuracy of manual review is only 83.2%. The traditional rule matching method has a slight improvement of 86.5%, but there are still high rates of false positives and missed positives. The present invention combines graph theory analysis with an improved Graphormer network to achieve an recognition accuracy of 94.8%, which is significantly better than existing methods and ensures accurate identification of supplier equity control relationships.

[0140] In terms of identifying implicitly controlled companies, manual review can only identify 27 suppliers with implicit equity control, while the traditional rule matching method can increase the number to 49. However, it still cannot fully identify complex equity nesting and indirect control relationships. The method of the present invention can identify 74 implicitly related suppliers, indicating that its ability in complex equity penetration analysis far exceeds that of traditional methods, effectively reducing the risk of suppliers using equity structures to evade supervision.

[0141] In terms of supplier review time, manual review usually takes 7 to 10 days, while traditional rule matching methods can shorten it to 5 to 7 days, but there is still the problem of being time-consuming. The method of the present invention can shorten the review time to 1 to 2 days, providing a more efficient and real-time risk control means for supplier management of power grid engineering projects.

[0142] In terms of the amount of supply chain risk prevention, traditional manual review can avoid about 150 million yuan of supply chain procurement risks, and the rule matching method can increase it to 320 million yuan, but there are still large loopholes. The application of the method of the present invention in supply chain management has successfully prevented 630 million yuan of potential procurement risks, providing strong protection for the compliance and security of the power grid project supply chain.

[0143] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for intelligently identifying equity relationships among power grid project suppliers based on graph theory analysis, characterized in that: The steps include: S1. Collect supplier relationship data, set the collection frequency, automatically run the collection program, and pre-process the collected data to generate a standardized data set; S2. Extract the equity structure characteristics and transaction characteristics of suppliers in the standardized dataset, and calculate the shareholding ratio, equity concentration, transaction amount, transaction frequency, and bidding cooperation relationship. Store the data in the supplier equity relationship database and index it according to the supplier's unique identifier; S3. Construct a supplier equity relationship graph based on the supplier equity relationship database. Use depth-first search and breadth-first search algorithms to traverse the supplier equity relationship graph, identify the equity control relationships and ultimate controllers between suppliers, and calculate the control path length and equity control weight. S4. Optimize the identification of supplier equity control relationships using an improved Graphormer network. Construct an improved Graphormer network. Train the improved Graphormer network based on supplier equity relationship data to learn equity control relationship patterns and predict the strength of equity control relationships between suppliers. S5. Calculate the supplier's risk score, set thresholds based on the risk score, and classify the supplier's risk level; S6. Use incremental learning methods to optimize the Graphormer network, update the supplier equity relationship graph based on historical risk data, and update the supplier equity relationship database.

2. The method for intelligently identifying equity relationships of power grid project suppliers based on graph theory analysis according to claim 1 is characterized in that: The supplier relationship data includes supplier business registration information, shareholder structure, corporate annual reports, equity change records, historical bidding information and supply chain transaction data. The preprocessing includes format conversion, data deduplication, outlier detection, missing data filling, time synchronization and normalization.

3. The method for intelligently identifying equity relationships of power grid project suppliers based on graph theory analysis according to claim 1 is characterized in that: The S2 specifically includes: S21. Extract suppliers' business registration information, shareholder structure, annual reports, equity change records, historical bidding information, and supply chain transaction data from standardized data sets, and associate the data based on the supplier's unique identifier to ensure complete matching of all types of data for the same supplier. S22. Calculate the supplier's shareholding ratio; S23. Calculate the supplier's equity concentration by sorting the shareholders based on their shareholding ratios, extract the shareholding ratios of the top N shareholders, and sum them up; S24. Calculate the supplier's transaction amount, extract the supplier's annual transaction records based on the supply chain transaction data, calculate the supplier's annual total transaction amount and annual transaction number, and calculate the supplier's transaction frequency; S25. Calculate suppliers’ bidding partnerships and, based on historical bidding data, count suppliers’ joint bidding in different projects. S26. Build a supplier equity relationship database, store the calculated shareholding ratio, equity concentration, transaction amount, transaction frequency and bidding cooperation relationship in the database, and index it according to the supplier's unique identifier.

4. The method for intelligently identifying equity relationships of power grid project suppliers based on graph theory analysis according to claim 1 is characterized in that: The S3 specifically includes: S31. Based on the data in the supplier equity relationship database, construct a supplier equity relationship graph, wherein the supplier equity relationship graph comprises suppliers, shareholders, and ultimate controllers as nodes, equity holding relationships, transaction relationships, and bidding cooperation relationships as edges, and weight values ​​are set for the edges; S32. Use a depth-first search algorithm to traverse the supplier equity relationship graph, starting with the supplier as the starting node, and search upward along the shareholder shareholding path until reaching the ultimate controller, and calculate the length of the control path. The depth-first search algorithm traversal process includes: Select the starting supplier node S0, mark it as visited, and push it onto the stack; From the current stack top node S t Select an unvisited direct shareholder node S t+1 ; If S t+1 If it exists, mark S t+1 If it has been visited, push it onto the stack and continue searching upwards; If S t+1 If it does not exist, go back to the previous node S t-1 , find unvisited direct shareholder nodes and repeat the selection and search operations; When all shareholder nodes have been visited and backtracking to the starting supplier node S0 with no unvisited nodes, the search ends and the control path length is calculated; S33. Calculate the supplier's equity control weight based on the control path length using a depth-first search algorithm; S34. Using a breadth-first search algorithm to traverse the supplier equity relationship graph, starting with the ultimate controller as the starting node, searching downward along the shareholder shareholding path to identify the ultimate controller's controlled suppliers. The breadth-first search algorithm traversal process includes: Select the starting controller node C0, mark it as visited, and add it to the queue; From the current queue head node C t Select all directly controlled supplier nodes C t+1 ; C t+1 The unvisited nodes are added to the queue and marked as visited; Continue to take the next node from the head of the queue and repeat the selection and addition operations until all supplier nodes have been visited; Calculate the scope of control of suppliers, count the number of suppliers directly or indirectly influenced by each controller, and calculate the weight of controlled suppliers; S35. Calculate the preliminary equity control relationship strength between suppliers based on the supplier's equity control weight and the transaction amount between suppliers.

5. The method for intelligently identifying equity relationships of power grid project suppliers based on graph theory analysis according to claim 1 is characterized in that: The S4 specifically includes: S41. Construct an improved Graphormer network and input supplier equity relationship data. The improved Graphormer network includes an input layer, an improved graph structure encoding layer, an adaptive multi-head attention mechanism, a global relationship modeling layer, and a fully connected classification layer. The supplier equity relationship data includes supplier equity structure characteristics, transaction characteristics, and control path data. The supplier equity structure characteristics include shareholding ratio, shareholder level, and equity concentration. The transaction characteristics include transaction amount, transaction frequency, and bidding cooperation relationship. The control path data includes control path length, equity control weight, and ultimate controller identifier. S42. Based on the supplier equity relationship data in the improved graph structure encoding layer, the improved Graphormer network is used to identify the supplier equity control relationship, the improved Graphormer network is trained, the equity control relationship model is learned, and the supplier equity relationship feature vector H is set. (l) : H (l+1) =σ(D -1 / 2 AD -1 / 2 H (l) W (l) +λdiag(I)H (l) ); Among them, H (l+1) represents the supplier equity relationship feature vector of the l+1 layer, σ represents the nonlinear activation function, H (l) represents the eigenvector of the supplier equity relationship at level l, D represents the degree matrix of the adjacency matrix, A represents the adjacency matrix of the supplier equity relationship graph, and W (l) represents the training weight matrix of the lth layer, λ represents the adaptive adjustment parameter, I represents the importance vector of the supplier node, and diag represents the diagonalization of the vector; S43. In the adaptive multi-head attention mechanism, the attention weights between suppliers are calculated and the feature vectors are weighted updated to optimize the prediction of supplier control relationships: Among them, Z represents the final multi-head attention output vector, Concat represents the concatenation of the outputs of each attention head, H represents the number of attention heads, Softmax represents normalization, and d k represents the dimension of the key vector, Q h , K h and V h denote the query, key, and value vectors of the supplier feature vector, γ, δ, and η denote weight coefficients, and T i,j 、 and denote the normalized values ​​of the transaction amount, control path length, and equity control weight between supplier i and supplier j, respectively. T denotes the transpose operation of the vector; S44. At the global relationship modeling layer, the strength of the equity control relationship between suppliers is calculated based on the multi-head attention output vector: Among them, P i,j represents the strength of the equity control relationship between supplier i and supplier j, σ represents the nonlinear activation function, and W p and W h represents the training weight matrix, Z i and Z j Represent the feature vectors of supplier i and supplier j after attention mechanism update, represents the normalized value of the initial equity control relationship strength, represents the normalized value of the bidding cooperation relationship, α and β represent the weight coefficients; S45, based on the equity control relationship strength P between suppliers i,j , optimize the prediction of supplier equity control relationships at the fully connected classification layer, set a set of equity control relationship categories, calculate the probability distribution of equity control relationship categories between suppliers, and define a loss function. The loss function uses a weighted cross entropy loss function to assign higher weights to high-risk suppliers: Where L represents the loss function, O represents the total number of supplier relationship pairs, K represents the total number of equity control relationship categories, ω k represents the weight of category k, y i,j,k Represents the true category label, using one-hot encoding, Represents the predicted category probability, P i,j represents the strength of the equity control relationship between supplier i and supplier j, exp represents the natural exponential function, and W c and b c Represents the training weight matrix and bias of the fully connected layer, and k′ represents the category; S46, according to the equity control relationship strength P i,j , calculate the changing trend of equity control weights among suppliers, and adjust the supplier equity relationship database based on time series analysis to optimize the supplier equity control hierarchy structure; S47. Update the supplier's equity control hierarchy based on the adjusted equity control weights.

6. The method for intelligently identifying equity relationships of power grid project suppliers based on graph theory analysis according to claim 1 is characterized in that: The S6 specifically includes: S61. Setting an optimization objective function, training an improved Graphormer network based on historical risk data, wherein the historical risk data includes historical supplier equity structure data, transaction data, and control path data, and constructing a historical risk data matrix; S62. Based on the historical risk data matrix, an incremental learning strategy is used to update the parameters of the Graphormer network to adapt to the dynamically changing supplier equity control relationship. The incremental learning objective function is defined as: in, represents the incremental learning objective function, represents the predicted probability of supplier equity control relationship category, y i,j represents the true label, ω j represents the category weight, W prev Represents the parameters of the Graphormer network obtained in the previous round of training, W represents the parameters of the Graphormer network, ψ represents the regularization parameter, which is used to control the update amplitude of the parameters, |||| 2 represents the norm operation; S63. Based on the Graphormer network optimized by incremental learning, the supplier equity control relationship strength is recalculated, and the risk score is calculated based on the optimized equity control relationship strength, the supplier's risk classification is adjusted, and the supplier equity relationship database is updated.

Images