A hierarchical entropy-oriented substation engineering feasibility study report evidence graph method

Abstract

The application discloses a hierarchical entropy-oriented substation engineering feasibility study report evidence graph method. In view of the problems that the structural relationship of specification clauses, historical reports, table annexes and calculation data is easy to be lost in intelligent writing of the substation engineering feasibility study report, evidence retrieval noise is large, and a large model lacks traceable constraints, the application proposes a hierarchical entropy-oriented evidence graph construction framework. Specifically, the application first analyzes project parameters, site data, load data, equipment lists, specification files and historical reports to construct an evidence graph containing chapters, indexes, clauses, tables, formulas and source coordinates; then calculates hierarchical structure entropy based on node correlation degree, chapter ownership and source credibility to generate chapter evidence subgraphs; and finally binds the subgraphs to a report template to provide structured prior for chapter generation and verification. The application improves evidence organization integrity, retrieval accuracy and traceability, and effectively reduces large model illusion.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and intelligent processing technology of power engineering documents, specifically to a hierarchical entropy-oriented evidence mapping method for substation engineering feasibility study reports. Background Technology

[0002] With the increasing demands for refined management of power grid construction projects, the preparation of feasibility study reports for substation projects requires a comprehensive approach, incorporating project parameters, site conditions, load forecasts, equipment lists, technical specifications, historical reports, tables and attachments, and calculation bases. Traditional manual preparation methods typically rely on engineers reviewing documents item by item and manually organizing the supporting evidence, resulting in low data analysis efficiency, weak cross-document evidence connections, and difficulties in tracing the basis of each chapter. Especially in substation engineering scenarios, there are strong structural relationships between specification clauses, indicator data, table units, and calculation formulas. Using only ordinary text retrieval or template filling methods can easily disrupt the hierarchical relationships, table row and column relationships, and formula variable relationships in the original data, leading to missing evidence, mismatched bases, and inconsistent data citations when generating chapters.

[0003] In recent years, large language models have demonstrated strong text generation capabilities in assisting the writing of feasibility study reports. However, their generation process typically relies on contextual fragment input and lacks explicit modeling of the structural relationships of multi-source evidence. Existing knowledge-based or retrieval-enhanced generation methods often focus on semantic similarity matching, making it difficult to accurately retain chapter affiliations, indicator references, normative constraints, source coordinates, and computational dependencies in substation engineering data. Furthermore, retrieval results often contain evidence nodes with low relevance to the current chapter, resulting in a lack of stable, structured prior constraints in the model-generated content. These issues make it difficult for existing technologies to meet the requirements of evidence completeness, traceability, and accuracy of chapter argumentation in substation engineering feasibility study reports. Summary of the Invention

[0004] This invention provides a hierarchical entropy-oriented evidence graph method for substation engineering feasibility study reports. This method analyzes project parameters, site information, load data, equipment lists, specifications, historical reports, table attachments, and calculation formulas to extract text paragraphs, table cells, indicator values, formula variables, and source locations. It constructs an evidence graph containing chapter nodes, indicator nodes, specification clause nodes, table nodes, formula nodes, project parameter nodes, and source nodes. Edge connections are established based on semantic relationships, chapter affiliation relationships, indicator reference relationships, table row and column relationships, formula variable relationships, and source tracing relationships. Hierarchical structure entropy is calculated by combining node degree, connection strength, chapter affiliation, source credibility, and evidence relevance. Based on the constraint of minimizing hierarchical structure entropy, the graph hierarchy is divided and evidence clusters are merged to generate chapter evidence subgraphs corresponding to the report chapters. These chapter evidence subgraphs are bound to the feasibility study report chapter templates to form a structured evidence index, providing traceable evidence support for chapter generation, basis verification, and manual review. This improves the completeness of evidence organization, retrieval accuracy, and reliability of large model generation in substation engineering feasibility study reports.

[0005] The present invention provides a hierarchical entropy-guided evidence mapping method for substation engineering feasibility study reports, comprising the following steps: S1: Analyze multi-source data from the feasibility study report of the substation project, including project parameters, site information, load data, equipment list, standard documents, historical reports, table attachments and calculation formulas, and extract text paragraphs, chapter titles, standard clauses, table rows and columns, index values, calculation formulas, unit information and source location to form unified basic evidence data; S2: Standardize the basic evidence data and generate chapter nodes, indicator nodes, normative clause nodes, table nodes, formula nodes, project parameter nodes, and source nodes according to the evidence type. Configure node type, chapter, data unit, time version, source file, page number, and table coordinate attributes for each node. S3: Based on the semantic relationships, chapter affiliation relationships, indicator reference relationships, table row and column relationships, formula variable relationships, and source tracing relationships between the nodes, establish edge connections, construct the evidence map of the substation engineering feasibility study report, and record the supporting relationships between chapters and evidence, indicators and data, clauses and conclusions, formulas and variables, tables and cells, and evidence and sources; S4: Calculate the hierarchical structure entropy based on the connection strength, node degree, chapter affiliation, source credibility and evidence relevance of the nodes in the evidence graph to obtain the degree of organization and information uncertainty of the evidence nodes under the current graph division, and identify the node set with loose evidence distribution, weak chapter association and high retrieval noise. S5: With the constraint of minimizing the hierarchical structure entropy, the evidence graph is hierarchically divided and iteratively merged. Nodes related to the same chapter, the same argumentation indicator, the same normative basis, and the same data source are aggregated into chapter evidence subgraphs, while retaining the original source information and cross-node relationships. S6: Bind the chapter evidence sub-graphs to the feasibility study report chapter templates to form a structured evidence index for each chapter, and output the corresponding evidence nodes, citation sources, indicator data, normative clauses and calculation basis.

[0006] According to a specific implementation of an embodiment of the present invention, the specific steps of S1 are as follows: S11. Analyze the project parameters, site information, load data, equipment list, specifications, historical reports, table attachments and calculation formulas in the feasibility study report of the substation project to obtain multimodal evidence units from the first document; From the The data was broken down into text segments, table cells, calculation formulas, indicator values, and source locations to obtain a data analysis set:

[0007] in, Indicates the first Analysis results of the data. Represents a collection of text fragments. Represents a set of table cells. Represents a set of computational expressions. Represents a set of indicator values. This represents a set of source locations, which includes the filename, page number, paragraph number, cell coordinates, and time version. S12, Extract the first from the data parsing set. From each content block, we obtain the evidence unit:

[0008] in, Indicates the first The first document One unit of evidence, Indicate the content of the evidence, Indicate the type of evidence. Represents the metadata of the indicator. Indicating the source location, the aforementioned evidence units collectively constitute the basic evidence data; According to a specific implementation of an embodiment of the present invention, the specific steps of S2 are as follows: S21. Extract the node type, chapter, data unit, time version, source file, page number, and table coordinate attributes from the basic evidence data to obtain the attribute group of the $i$-th evidence node:

[0009] in, Indicates the first One piece of evidence, Indicates the node type. Indicates the chapter to which it belongs. Indicates the unit of data. Indicates the time version. Indicates the source document, Indicates page number and table coordinate attributes; S22. Perform discrete attribute embedding and numerical attribute normalization mapping based on the attribute group to obtain the standardized feature vector of the evidence node:

[0010] in, Indicates the first The standardized feature vector of each evidence node, This represents a discrete attribute embedding function. This represents the normalization mapping function for numerical attributes. , , , , and These are mapping matrices representing node type, chapter, data unit, time version, source file, and coordinate attributes, respectively. S23. Based on the distances between the source attribute vector and the ideal source vector and the negative source vector, the source credibility is obtained:

[0011] in, Indicates the first The credibility of the source of each piece of evidence. This represents the distance from the source attribute vector to the ideal source vector. This represents the distance from the source attribute vector to the negative source vector. The source credibility is determined by the authority of the data, the time version, the relevance of the chapter, and the completeness of the data. According to a specific implementation of an embodiment of the present invention, the specific steps of S3 are as follows: S31. Based on the standardized node set, edge set, node type mapping, relation type mapping, and source attribute mapping, the evidence graph is obtained:

[0012] in, Representing the evidence spectrum, Represents the set of evidence nodes. Denotes the set of edges. Represents node type mapping, Represents relation type mapping, Indicates source attribute mapping; S32. Calculate the node association strength based on node feature cosine similarity, chapter affiliation consistency, source relevance, table row and column binding strength, and formula variable dependency strength:

[0013] in, Indicates the first The evidence node and the first The strength of the association between each piece of evidence node Represents the cosine similarity of node features. This indicates consistency in chapter affiliation. Indicates the degree of relevance to the source. Indicates the row and column binding strength of the table. Indicates the strength of the dependency between the variables in the formula. to Indicates the corresponding weight; S33. Based on the node association strength and edge connection threshold, obtain the graph edge set:

[0014] in, Represents a node With nodes Relationship types This represents the edge connection threshold; the set of edges records the supporting relationships between chapters, indicators, clauses, tables, formulas, and sources. According to a specific implementation of an embodiment of the present invention, the specific steps of S4 are as follows: S41, according to the... Calculate the first-order structural entropy based on the node degree of each evidence node and the graph volume of the evidence graph:

[0015] in, Indicates the first The degree of each evidence node. The volume of the evidence map represents the graph volume. This represents the node degree normalization coefficient. This represents the first-order structural entropy of the evidence map; S42. Based on the volume of the evidence cluster in the current hierarchical division, calculate the node normalization coefficient within the evidence cluster:

[0016] in, This represents the evidence cluster in the current hierarchical division. Indicates a cluster of evidence Cluster volume, Represents a node In the cluster of evidence The normalization coefficient within; S43. Calculate the hierarchical structure entropy based on the internal entropy of the evidence cluster and the inter-cluster cut edge information:

[0017] in, Indicates the current level division. Indicates a cluster of evidence The number of cut edges between the node and the external node. This represents the entropy of the hierarchical structure; the lower the entropy of the hierarchical structure, the higher the degree of aggregation of evidence nodes in terms of chapter, indicator, and source dimensions. According to a specific implementation of an embodiment of the present invention, the specific steps of S5 are as follows: S51. Based on the hierarchical structure entropy and the partitioning penalty term, the optimal hierarchical partitioning is obtained:

[0018] in, This indicates the optimized hierarchical division. Represents the canonical balance coefficient. This indicates the division of penalty items; S52. Based on the intra-cluster association loss and inter-cluster edge cuts, the partitioning penalty term is obtained:

[0019] in, Indicates the strength of association between nodes in the same cluster. Indicates a cluster of evidence The number of external cut edges; this penalty term constrains weakly associated nodes from entering the same evidence cluster and suppresses excessive cross-cluster edges; S53. Based on the difference in hierarchical entropy before and after the merger, obtain the decrease in structural entropy between any two evidence clusters:

[0020] in, Indicates a cluster of evidence With cluster of evidence The decrease in structural entropy after the merge is performed. This represents the new division after merging two evidence clusters; when Higher than the preset threshold At that time, a merge operation is performed, and the original source information and cross-node relationships are preserved during the iteration process; According to a specific implementation of an embodiment of the present invention, the specific steps of S6 are as follows: S61. Based on the chapter evidence subgraph and the chapter to which the evidence node belongs, obtain the chapter binding marker:

[0021] in, Indicates the first Chapter and the The binding markers between evidence nodes Indicates the first Evidence sub-graphs corresponding to each chapter Indicates the first The chapter to which each piece of evidence belongs; S62. Calculate the argumentation requirement matching score based on the chapter argumentation requirement vector, the standardized feature vector of evidence nodes, source credibility, source traceability score, and chapter relevance:

[0022] in, Indicates the first The first chapter The evidence matching score for each argument requirement Indicates the first The first chapter A vector of argumentation requirements, Indicates the credibility of the source. Indicates a score that allows for traceability of the source. Indicates the relevance of chapters. to Indicates the matching weight; S63. Calculate the chapter evidence index quality score based on the argumentation requirement matching score, source credibility, source traceability score, and chapter relevance:

[0023] in, Indicates the first The evidence index quality score for each chapter to Indicates the weight of quality evaluation. This indicates that the average value of all argument requirements within the chapter is taken. This indicates that the average value is taken for the bound evidence nodes; when Reaching the threshold When, output the chapter evidence index; when Below If necessary, return to S41 to S43 to recalculate the hierarchy.

[0024] The beneficial effects of this invention are as follows: First, this invention parses project parameters, site information, load data, equipment lists, specifications, historical reports, table attachments, and calculation formulas into evidence units with source locations, and further forms evidence nodes. This allows for the unified organization of indicators, clauses, tables, and calculation bases scattered across different documents during the feasibility study report writing process. This approach helps reduce omissions during data extraction, manual comparison, and source tracing. Secondly, this invention constructs an evidence graph through node association strength and introduces hierarchical structure entropy to constrain the graph division, making it easier to aggregate evidence nodes related to the same chapter, the same argumentation indicator, the same normative basis, and the same data source into chapter evidence subgraphs. Compared to methods that simply rely on keyword retrieval or vector similarity retrieval, this invention can, to a certain extent, reduce the probability of irrelevant evidence entering chapter materials and improve the tightness of chapter evidence organization; Third, this invention binds the chapter evidence sub-graphs to the feasibility study report chapter templates, calculates the argumentation requirement matching score and the evidence index quality score, and returns to the hierarchical structure entropy calculation stage for re-division when the quality score is insufficient. Thus, the system can form a structured evidence index before generating or verifying chapter content, ensuring a clear source chain for the evidence nodes, indicator data, normative clauses, and calculation basis referenced in the chapters, thereby improving traceability and review efficiency during the report writing process. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method; Figure 2 This is the architecture diagram of this method. Detailed Implementation

[0026] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to examples and accompanying drawings.

[0027] As attached Figure 1 and attached Figure 2 As shown, a hierarchical entropy-oriented evidence mapping method for substation engineering feasibility study reports includes the following steps: Step 1: Receive the project parameters, site information, load data, equipment list, specifications, historical reports, tables and attachments and calculation formulas required for the preparation of the substation project feasibility study report. Establish a data sequence according to the data source, document version and creation time, and assign a data number to each document. For the The data was processed through text segmentation, table cell recognition, calculation expression recognition, indicator value extraction, and source location positioning to obtain a data parsing set:

[0028] in, Indicates the first The analysis results of the data; Represents a collection of text fragments; Represents a set of table cells; Represents a set of computational expressions; Represents a set of indicator values; This represents a set of source locations; the source locations include filename, page number, paragraph number, cell coordinates, and time version. Extract the first from the data parsing set Each content block combines the evidence content, evidence type, indicator metadata, and source location to obtain an evidence unit:

[0029] in, Indicates the first The first document One unit of evidence; Indicate the content of the evidence; Indicate the type of evidence; This represents the metadata of the indicator; Indicates the source location; the aforementioned evidence units collectively constitute the basic evidence data; Step 2: Read the basic evidence data, generate chapter nodes, indicator nodes, standard clause nodes, table nodes, formula nodes, project parameter nodes, and source nodes according to evidence type, and extract the node type, chapter, data unit, time version, source file, page number, and table coordinate attributes from each node to obtain the [previous node]. The attribute group of each evidence node:

[0030] in, Indicates the first One evidence node; Indicates the node type; Indicates the chapter to which it belongs; Indicates the unit of data; Indicates the time version; Indicates the source document; Indicates page number and table coordinate attributes; Based on the attribute group, discrete attribute embedding and numerical attribute normalization mapping are performed to map evidence nodes from different sources to a unified feature space, resulting in standardized feature vectors for the evidence nodes:

[0031] in, Indicates the first The standardized feature vector of each evidence node; Represents a discrete attribute embedding function; This represents a function for normalizing numerical attributes. , , , , and These are mapping matrices representing node type, chapter, data unit, time version, source file, and coordinate attributes, respectively. Calculate the source credibility of each evidence node based on the distance between the source attribute vector and the ideal source vector and the negative source vector:

[0032] in, Indicates the first The credibility of the source of each piece of evidence; This represents the distance from the source attribute vector to the ideal source vector; This represents the distance from the source attribute vector to the negative source vector; source credibility is determined by the authority of the data, the time version, the relevance of the chapter, and the completeness of the data. Step 3: Establish edge connections based on the semantic relationships, chapter affiliation relationships, indicator reference relationships, table row and column relationships, formula variable relationships, and source tracing relationships between evidence nodes. Based on the standardized node set, edge set, node type mapping, relationship type mapping, and source attribute mapping, obtain the evidence graph:

[0033] in, Representing the evidence spectrum; Represents the set of evidence nodes; Represents the set of edges; Indicates node type mapping; Represents relation type mapping; Indicates source attribute mapping; For any two evidence nodes, the node association strength is calculated by combining node feature cosine similarity, chapter affiliation consistency, source relevance, table row and column binding strength, and formula variable dependency strength:

[0034] in, Indicates the first The evidence node and the first The strength of the association between each piece of evidence node; Represents the cosine similarity of node features; This indicates consistency in chapter affiliation; Indicates the degree of relevance to the source; Indicates the row and column binding strength of the table; Indicates the strength of dependency between formula variables; to Indicates the corresponding weight; The node association strength is compared with the edge connection threshold to filter node pairs that meet the connection conditions, and these pairs are added to the graph edge set.

[0035] in, Represents a node With nodes Relationship types; This represents the edge connection threshold; the edge set records the supporting relationships between chapters, metrics, clauses, tables, formulas, and sources. Step 4: Read the connection relationships of each evidence node in the evidence graph, and calculate the first-order structural entropy by combining the node degree and the graph volume of the evidence graph:

[0036] in, Indicates the first The degree of each evidence node; The volume of the evidence map; This represents the node degree normalization coefficient; This represents the first-order structural entropy of the evidence map; Based on the volume of the evidence cluster in the current hierarchical division, calculate the node normalization coefficient within the evidence cluster:

[0037] in, This represents the evidence cluster in the current hierarchical division; Indicates a cluster of evidence Cluster volume; Represents a node In the cluster of evidence The normalization coefficient within; By combining the entropy within the evidence cluster and the inter-cluster cut edge information, calculate the hierarchical structure entropy under the current level partition:

[0038] in, Indicates the current level division; Indicates a cluster of evidence The number of cut edges between the node and external nodes; This represents the entropy of the hierarchical structure; as the entropy of the hierarchical structure decreases, the degree of aggregation of evidence nodes increases in the dimensions of chapter, indicator, and source. Step 5: Combine the hierarchical structure entropy and the partitioning penalty term to obtain the optimization objective for the evidence graph hierarchical partitioning, and calculate the optimized hierarchical partitioning:

[0039] Where, represents the optimized hierarchical division; Indicates the canonical balance coefficient; This indicates the division of penalty items; Consistent with the aforementioned meaning; The penalty term is calculated based on intra-cluster association loss and inter-cluster edge cutting to constrain weakly associated nodes from entering the same evidence cluster and reduce interference caused by an excessive number of cross-cluster edges.

[0040] in, Indicates the strength of association between nodes in the same cluster; Indicates a cluster of evidence The number of external cut edges; and Consistent with the aforementioned meaning; During the iterative merging process, any two candidate evidence clusters are selected, and the difference in hierarchical structure entropy before and after merging is calculated to obtain the decrease in structural entropy:

[0041] in, Indicates a cluster of evidence With cluster of evidence The decrease in structural entropy after the merge; This represents the new division after merging the two evidence clusters; Indicates a preset threshold; when Higher than At that time, a merge operation is performed, and the original source information and cross-node relationships are preserved during the iteration process; Step 6: Obtain the chapter evidence subgraph based on the optimized hierarchical division, bind the chapter evidence subgraph to the feasibility study report chapter template, and obtain the chapter binding marker based on the chapter evidence subgraph and the chapter to which the evidence node belongs:

[0042] in, Indicates the first Chapter and the The binding markers between evidence nodes; Indicates the first Evidence sub-graphs corresponding to each chapter; Indicates the first The chapter to which each piece of evidence belongs; Indicates the first One evidence node; Based on the chapter argumentation demand vector, the standardized feature vector of evidence nodes, source credibility, source traceability score, and chapter relevance, calculate the argumentation demand matching score:

[0043] in, Indicates the first The first chapter Evidence matching score for each argument requirement; Indicates the first The first chapter A vector of argumentation requirements; Indicates the credibility of the source; Indicates a score indicating traceability of origin; Indicates the relevance of chapters; to Indicates the matching weight; Calculate the chapter evidence index quality score based on the argumentation need matching score, source credibility, source traceability score, and chapter relevance:

[0044] in, Indicates the first The evidence index quality score for each chapter; to Indicates the weight of quality evaluation; This indicates that the average value of all argument requirements within the chapter is taken. This indicates that the average value is taken for the bound evidence nodes; when Reaching the threshold When, output the chapter evidence index; when Below If necessary, return to step 4 to recalculate the hierarchy.

[0045] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A hierarchical entropy-guided evidence mapping method for substation engineering feasibility study reports, characterized in that... Includes the following steps: S1: Analyze multi-source data from the feasibility study report of the substation project, including project parameters, site information, load data, equipment list, standard documents, historical reports, table attachments and calculation formulas, and extract text paragraphs, chapter titles, standard clauses, table rows and columns, index values, calculation formulas, unit information and source location to form unified basic evidence data; S2: Standardize the basic evidence data and generate chapter nodes, indicator nodes, normative clause nodes, table nodes, formula nodes, project parameter nodes, and source nodes according to the evidence type. Configure node type, chapter, data unit, time version, source file, page number, and table coordinate attributes for each node. S3: Based on the semantic relationships, chapter affiliation relationships, indicator reference relationships, table row and column relationships, formula variable relationships, and source tracing relationships between the nodes, establish edge connections, construct the evidence map of the substation engineering feasibility study report, and record the supporting relationships between chapters and evidence, indicators and data, clauses and conclusions, formulas and variables, tables and cells, and evidence and sources; S4: Calculate the hierarchical structure entropy based on the connection strength, node degree, chapter affiliation, source credibility and evidence relevance of the nodes in the evidence graph to obtain the degree of organization and information uncertainty of the evidence nodes under the current graph division, and identify the node set with loose evidence distribution, weak chapter association and high retrieval noise. S5: With the constraint of minimizing the hierarchical structure entropy, the evidence graph is hierarchically divided and iteratively merged. Nodes related to the same chapter, the same argumentation indicator, the same normative basis, and the same data source are aggregated into chapter evidence subgraphs, while retaining the original source information and cross-node relationships. S6: Bind the chapter evidence sub-graphs to the feasibility study report chapter templates to form a structured evidence index for each chapter, and output the corresponding evidence nodes, citation sources, indicator data, normative clauses and calculation basis.

2. The hierarchical entropy-guided evidence mapping method for substation engineering feasibility study reports according to claim 1, characterized in that... The specific method for step S1 is as follows: S11. Analyze the project parameters, site information, load data, equipment list, specifications, historical reports, tables and attachments, and calculation formulas from the feasibility study report of the substation project. The data yielded a multimodal evidence unit; From the The data was broken down into text segments, table cells, calculation formulas, indicator values, and source locations to obtain a data analysis set: in, Indicates the first Analysis results of the data. Represents a collection of text fragments. Represents a set of table cells. Represents a set of computational expressions. Represents a set of indicator values. This represents a set of source locations, which includes the filename, page number, paragraph number, cell coordinates, and time version. S12, Extract the first from the data parsing set. From each content block, we obtain the evidence unit: in, Indicates the first The first document One unit of evidence, Indicate the content of the evidence, Indicate the type of evidence. Represents the metadata of the indicator. The source location is indicated, and the evidence units together constitute the basic evidence data.

3. The hierarchical entropy-guided evidence mapping method for substation engineering feasibility study reports according to claim 1, characterized in that... The specific method for step S2 is as follows: S21. Extract the node type, chapter, data unit, time version, source file, page number, and table coordinate attributes from the basic evidence data to obtain the... The attribute group of each evidence node: in, Indicates the first One piece of evidence, Indicates the node type. Indicates the chapter to which it belongs. Indicates the unit of data. Indicates the time version. Indicates the source document, Indicates page number and table coordinate attributes; S22. Perform discrete attribute embedding and numerical attribute normalization mapping based on the attribute group to obtain the standardized feature vector of the evidence node: in, Indicates the first The standardized feature vector of each evidence node, This represents a discrete attribute embedding function. This represents the normalization mapping function for numerical attributes. , , , , and These are mapping matrices representing node type, chapter, data unit, time version, source file, and coordinate attributes, respectively. S23. Based on the distances between the source attribute vector and the ideal source vector and the negative source vector, the source credibility is obtained: in, Indicates the first The credibility of the source of each piece of evidence. This represents the distance from the source attribute vector to the ideal source vector. This represents the distance from the source attribute vector to the negative source vector. The source credibility is determined by the authority of the data, the time version, the relevance of the chapter, and the completeness of the data.

4. The hierarchical entropy-guided evidence mapping method for substation engineering feasibility study reports according to claim 1, characterized in that... The specific method in step S3 is as follows: S31. Based on the standardized node set, edge set, node type mapping, relation type mapping, and source attribute mapping, the evidence graph is obtained: in, Representing the evidence spectrum, Represents the set of evidence nodes. Denotes the set of edges. Represents node type mapping, Represents relation type mapping, Indicates source attribute mapping; S32. Calculate the node association strength based on node feature cosine similarity, chapter affiliation consistency, source relevance, table row and column binding strength, and formula variable dependency strength: in, Indicates the first The evidence node and the first The strength of the association between each piece of evidence node Represents the cosine similarity of node features. This indicates consistency in chapter affiliation. Indicates the degree of relevance to the source. Indicates the row and column binding strength of the table. Indicates the strength of the dependency between the variables in the formula. to Indicates the corresponding weight; S33. Based on the node association strength and edge connection threshold, obtain the graph edge set: in, Represents a node With nodes Relationship types This represents the edge connection threshold; the set of edges records the support relationships between chapters, metrics, clauses, tables, formulas, and sources.

5. The hierarchical entropy-guided evidence mapping method for substation engineering feasibility study reports according to claim 1, characterized in that... The specific method in step S4 is as follows: S41, according to the... Calculate the first-order structural entropy based on the node degree of each evidence node and the graph volume of the evidence graph: in, Indicates the first The degree of each evidence node. The volume of the evidence map represents the graph volume. This represents the node degree normalization coefficient. This represents the first-order structural entropy of the evidence map; S42. Based on the volume of the evidence cluster in the current hierarchical division, calculate the node normalization coefficient within the evidence cluster: in, This represents the evidence cluster in the current hierarchical division. Indicates a cluster of evidence Cluster volume, Represents a node In the cluster of evidence The normalization coefficient within; S43. Calculate the hierarchical structure entropy based on the internal entropy of the evidence cluster and the inter-cluster cut edge information: in, Indicates the current level division. Indicates a cluster of evidence The number of cut edges between the node and the external node. This represents the entropy of the hierarchical structure; the lower the entropy of the hierarchical structure, the higher the degree of aggregation of evidence nodes in terms of chapter, indicator, and source dimensions.

6. The hierarchical entropy-guided evidence mapping method for substation engineering feasibility study reports according to claim 1, characterized in that... The specific method in step S5 is as follows: S51. Based on the hierarchical structure entropy and the partitioning penalty term, the optimal hierarchical partitioning is obtained: in, This indicates the optimized hierarchical division. Represents the canonical balance coefficient. This indicates the division of penalty items; S52. Based on the intra-cluster association loss and inter-cluster edge cuts, the partitioning penalty term is obtained: in, Indicates the strength of association between nodes in the same cluster. Indicates a cluster of evidence The number of external cut edges; this penalty term constrains weakly associated nodes from entering the same evidence cluster and suppresses excessive cross-cluster edges; S53. Based on the difference in hierarchical entropy before and after the merger, obtain the decrease in structural entropy between any two evidence clusters: in, Indicates a cluster of evidence With cluster of evidence The decrease in structural entropy after the merge is performed. This represents the new division after merging two evidence clusters; when Higher than the preset threshold At that time, a merge operation is performed, and the original source information and cross-node relationships are preserved during the iteration process.

7. The hierarchical entropy-guided evidence mapping method for substation engineering feasibility study reports according to claim 1, characterized in that... The specific method in step S6 is as follows: S61. Based on the chapter evidence subgraph and the chapter to which the evidence node belongs, obtain the chapter binding marker: in, Indicates the first Chapter and the The binding markers between evidence nodes Indicates the first Evidence sub-graphs corresponding to each chapter Indicates the first The chapter to which each piece of evidence belongs; S62. Calculate the argumentation requirement matching score based on the chapter argumentation requirement vector, the standardized feature vector of evidence nodes, source credibility, source traceability score, and chapter relevance: in, Indicates the first The first chapter The evidence matching score for each argument requirement Indicates the first The first chapter A vector of argumentation requirements, Indicates the credibility of the source. Indicates a score that allows for traceability of the source. Indicates the relevance of chapters. to Indicates the matching weight; S63. Calculate the chapter evidence index quality score based on the argumentation requirement matching score, source credibility, source traceability score, and chapter relevance: in, Indicates the first The evidence index quality score for each chapter to Indicates the weight of quality evaluation. This indicates that the average value of all argument requirements within the chapter is taken. This indicates that the average value is taken for the bound evidence nodes; when Reaching the threshold When, output the chapter evidence index; when Below If the time is not met, return to S4 to recalculate the hierarchy.

Images