GraphRAG-based power generation professional knowledge multi-source fusion and global semantic understanding method and system

Abstract

The invention discloses a GraphRAG-based power generation professional knowledge multi-source fusion and global semantic understanding method and system, and belongs to the technical field of artificial intelligence, and the method comprises the steps: carrying out the automatic analysis of a plurality of data sources in the power generation industry; based on the analyzed multiple data sources of the power generation industry, establishing a knowledge graph suitable for the power generation industry; a GraphRAG technology is adopted, the knowledge graph and a large language model are combined, and a GraphRAG global semantic enhancement model is constructed; and based on a GraphRAG global semantic enhancement model, carrying out cross-document reasoning type question and answer. According to the method, the problem that global semantic understanding cannot be realized by traditional search and common RAG is solved.

Description

A GraphRAG-based method and system for multi-source fusion and global semantic understanding of power generation expertise Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to a method and system for multi-source fusion and global semantic understanding of power generation expertise based on GraphRAG. Background Technology

[0002] The power generation industry is characterized by diverse knowledge types, complex professional data, and intricate relationships between regulations and clauses. This includes operating procedures, operating standards, maintenance reports, accident analyses, drawings, design specifications, operating ledgers, parameter records, historical fault cases, inspection records, and expert conclusions. Traditional knowledge retrieval methods primarily rely on keyword matching, failing to understand deep semantics and cross-document relationships. While ordinary vector retrieval / RAG technology supports semantic search, it typically depends only on local information from a single document and cannot understand the professional relationships between multiple documents.

[0003] Therefore, the industry urgently needs a technology that can achieve global knowledge understanding across documents, devices, and disciplines in order to quickly obtain more complete and accurate professional answers in complex power systems.

[0004] GraphRAG, as a novel knowledge enhancement method, can enhance the global understanding of models through graph structures. However, it currently lacks a dedicated implementation path for the power generation industry, as well as a structured processing and reasoning mechanism adapted to the actual document characteristics of the industry. Therefore, there is still room for innovation in its application in the power generation field. Summary of the Invention

[0005] This invention aims to provide a GraphRAG-based method and system for multi-source knowledge fusion and global semantic understanding applicable to the power generation industry, achieving the following objectives: constructing a cross-document and cross-system knowledge graph covering the entire institute; realizing semantic fusion of multiple document types such as procedures, drawings, and reports; supporting global reasoning and cross-document semantic understanding of professional knowledge; providing high-precision professional knowledge question answering and business assistance capabilities; and solving the problem that traditional search and ordinary RAG cannot achieve "global semantic understanding".

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a GraphRAG-based method for multi-source fusion and global semantic understanding of power generation expertise, comprising: automatically parsing multiple data sources in the power generation industry; establishing a knowledge graph suitable for the power generation industry based on the parsed multiple data sources in the power generation industry; using GraphRAG technology to combine the knowledge graph with a large-scale language model to construct a GraphRAG global semantic enhancement model; and performing cross-document reasoning question answering based on the GraphRAG global semantic enhancement model.

[0007] A further improvement of this invention lies in the automatic parsing of various data sources in the power generation industry, including: operation logs, maintenance reports, historical drawings, operation procedures, parameter ledgers, and accident analysis documents in PDF, Word, PPT, and scanned formats; and the automatic parsing of these data sources using NLP and OCR technologies, including: paragraph parsing and heading level recognition, semantic unit segmentation, table extraction and parameter standardization, cross-document reference relationship recognition, technical terminology recognition, and synonym normalization.

[0008] A further improvement of this invention lies in establishing a knowledge graph suitable for the power generation industry based on multiple data sources from the parsed power generation industry data. This graph includes: node types: equipment, fault phenomena, fault causes, operating procedures, parameter indicators, procedure clauses, case reports, components and models, where equipment includes boilers, steam turbines, electrical equipment, and thermal control devices; automatic graph construction is achieved through entity extraction, relation extraction, and professional rule matching; during the knowledge graph construction process, entity extraction, relation extraction, and professional rule matching steps are executed sequentially: Entity extraction: Based on the named entity recognition model and industry lexicon, candidate entities are extracted from the structured parsing results; and according to predefined node type rules, the candidate entities are instantiated into knowledge graph nodes of the corresponding type; Relation extraction: Based on syntactic analysis, Semantic role labeling and cross-document reference relationships automatically identify semantic relationships between entities, including: "Equipment – ​​Occurrence – Fault Phenomenon"; "Fault Phenomenon – Corresponding – Fault Cause"; "Fault Cause – Association – Operation Steps"; "Procedure Clause – Provision – Operation Steps"; "Case Report – Involving – Equipment / Fault"; Professional rule matching: Based on the power generation industry expert rule base, the extracted results are verified and supplemented, including: if a fault cause does not match a specific equipment type, the confidence level is reduced or the fault is removed; if the procedure clause clearly stipulates a causal relationship, a corresponding relationship edge is forcibly established; if the same equipment number appears in multiple documents, it is automatically merged into the same equipment node; ultimately, the automatic construction of knowledge graph nodes and relationships from multiple data sources in the parsed power generation industry is achieved.

[0009] A further improvement of this invention is that it employs GraphRAG technology to combine knowledge graphs with large-scale language models to construct a GraphRAG global semantic enhancement model. This model includes: performing node-level and subgraph-level retrievals in the knowledge graph based on user questions and obtaining a global knowledge domain across documents; or aggregating and deduplicating the content of multiple nodes and documents to automatically generate a "cross-document semantic summary"; and injecting subgraph summaries, hierarchical relationships, and causal chain structures into the context of the LLM to give the model a "global knowledge background".

[0010] A further improvement of this invention is that the construction method of the GraphRAG global semantic enhancement model includes the following steps: (1) Problem-driven graph-aware retrieval When a problem is received from a user, the problem is first semantically parsed to identify the equipment type, system scope, fault phenomenon or parameter index involved in the problem, and the semantic information is mapped to one or more graph query starting nodes; (2) Node-level and subgraph-level joint retrieval Centered on the query starting node, the knowledge graph is traversed according to the preset relationship type to obtain several nodes and their relationships related to the problem semantics, and a problem-related subgraph is constructed; the subgraph includes at least: multiple types of nodes, namely equipment, fault, procedure and case; hierarchical relationship, causal relationship and dependency relationship between nodes; (3) Subgraph semantic aggregation and redundancy removal The node content in the problem-related subgraph is semantically aggregated and the following processing is performed: nodes from different documents but with the same or similar semantics are merged; redundant descriptions of the same fault or parameter are removed; weights are calculated based on the centrality or confidence of the node in the graph; and cross-document semantic summaries are generated based on the aggregation results; (4) Graph semantics to LLM The injection of context will organize cross-document semantic summary, relationship structure information between nodes and key original text references into structured prompt content according to the preset context template, including: key entity list; causal relationship chain; hierarchical structure description; corresponding original text evidence fragments; the structured prompts are injected into the large language model as context input, so that the language model has a global knowledge background when generating answers; (5) Language model generation and verification The large language model generates answer results based on the injected graph semantic context, performs rule verification and reference consistency check on the generated results, and ensures that the answer content is consistent with the knowledge graph.

[0011] A further improvement of this invention is that, based on the GraphRAG global semantic enhancement model, cross-document reasoning question answering is performed, including: identifying the device, system, and parameters to which the user's question belongs; locating relevant subgraphs in the knowledge graph; automatically expanding reasoning along relational paths; combining RAG to retrieve the original text content; using LLM to fuse the results to generate the final answer; and realizing the integration of steps across multiple procedures or the understanding of related questions across systems, as well as the difference analysis between different clauses.

[0012] A further improvement of this invention is that it also includes: a dynamic updating and continuous learning mechanism for professional knowledge, including supporting online updates of knowledge; automatically generating new knowledge nodes based on new cases; and automatically supplementing weak knowledge areas based on user questions and answers.

[0013] The GraphRAG-based multi-source fusion and global semantic understanding system for power generation expertise includes: a data parsing unit for automatically parsing various data sources in the power generation industry; a knowledge graph building unit for building a knowledge graph suitable for the power generation industry based on the parsed data sources; a model building unit that uses GraphRAG technology to combine the knowledge graph with a large-scale language model to build a GraphRAG global semantic enhancement model; and a reasoning question answering unit that performs cross-document reasoning question answering based on the GraphRAG global semantic enhancement model.

[0014] A further improvement of this invention lies in the automatic parsing of multiple data sources in the power generation industry within the data parsing unit. These data sources include: operation logs, maintenance reports, historical drawing database content, operation procedures, parameter ledgers, and accident analysis documents in PDF, Word, PPT, and scanned formats. NLP and OCR technologies are used to automatically parse these multiple data sources, including: paragraph parsing and heading level recognition, semantic unit segmentation, table extraction and parameter standardization, cross-document reference relationship recognition, and professional terminology recognition and synonym normalization.

[0015] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the GraphRAG-based multi-source fusion and global semantic understanding method for power generation expertise.

[0016] Compared with existing technologies, the present invention has at least the following beneficial technical effects: The multi-source knowledge fusion and global semantic understanding method and system based on GraphRAG provided by the present invention can achieve cross-document, cross-system, and global semantic understanding, and can automatically associate multiple procedures, drawings, and case content to achieve "industry expert-style" reasoning. The present invention can provide highly accurate professional question-and-answer services, which are more accurate, interpretable, and comprehensively referenced than ordinary RAG question-and-answer.

[0017] Furthermore, this invention has strong knowledge integration capabilities and can process multiple formats of data, including text, drawings, case studies, and data.

[0018] Furthermore, this invention has strong industry adaptability and is optimized for equipment, technical terminology, and causal relationships between systems in the power generation industry. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 is a flowchart of the multi-source fusion and global semantic understanding method for power generation expertise based on GraphRAG according to the present invention; Figure 2 is a structural block diagram of the multi-source fusion and global semantic understanding system for power generation expertise based on GraphRAG according to the present invention. Detailed Implementation

[0021] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0022] In the description of this invention, it should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0026] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] As shown in Figure 1, the GraphRAG-based multi-source fusion and global semantic understanding method for power generation expertise provided by this invention includes: automatically parsing multiple data sources in the power generation industry; establishing a knowledge graph suitable for the power generation industry based on the parsed multiple data sources; using GraphRAG technology to combine the knowledge graph with a large-scale language model to construct a GraphRAG global semantic enhancement model; and performing cross-document reasoning question answering based on the GraphRAG global semantic enhancement model.

[0028] In this embodiment, multiple data sources from the power generation industry are automatically parsed, including: operation logs, maintenance reports, historical drawings, operation procedures, parameter ledgers, and accident analysis documents in PDF, Word, PPT, and scanned formats; NLP and OCR technologies are used to automatically parse these multiple data sources from the power generation industry, including: paragraph parsing and heading level recognition, semantic unit segmentation, table extraction and parameter standardization, cross-document reference relationship recognition, technical terminology recognition, and synonym normalization.

[0029] In this embodiment, a knowledge graph suitable for the power generation industry is established based on various data sources from the parsed power generation industry, including: node types: equipment, fault phenomena, fault causes, operation steps, parameter indicators, procedure clauses, case reports, components and models, where equipment includes boilers, steam turbines, electrical equipment and thermal control devices; the graph is automatically constructed through entity extraction, relation extraction and professional rule matching; during the knowledge graph construction process, the following steps are executed sequentially: Entity extraction: Based on the named entity recognition model and industry thesaurus, candidate entities are extracted from the structured parsing results; and according to predefined node type rules, the candidate entities are instantiated into knowledge graph nodes of the corresponding type; Relation extraction: Based on syntactic analysis and semantic angle... Color-coded annotations and cross-document references automatically identify semantic relationships between entities, including: "Equipment – ​​Occurrence – Fault Phenomenon"; "Fault Phenomenon – Corresponding – Fault Cause"; "Fault Cause – Association – Operation Steps"; "Procedure Clauses – Provisions – Operation Steps"; "Case Reports – Involving – Equipment / Fault"; Professional rule matching: Based on the power generation industry expert rule base, the extracted results are verified and supplemented, including: if a fault cause does not match a specific equipment type, the confidence level is reduced or the fault is removed; if the procedure clauses clearly stipulate causal relationships, corresponding relationship edges are forcibly established; if the same equipment number appears in multiple documents, it is automatically merged into the same equipment node; ultimately, the automatic construction of knowledge graph nodes and relationships from parsed power generation industry data sources is achieved.

[0030] In this embodiment, GraphRAG technology is used to combine knowledge graphs with large-scale language models to construct a GraphRAG global semantic enhancement model, including: performing node-level and subgraph-level retrieval in the knowledge graph based on user questions and obtaining a global knowledge domain across documents; or aggregating and deduplicating the content of multiple nodes and multiple documents to automatically generate a "cross-document semantic summary"; and injecting subgraph summaries, hierarchical relationships, and causal chain structures into the context of the LLM to give the model a "global knowledge background".

[0031] In this embodiment, the construction method of the GraphRAG global semantic enhancement model includes the following steps: (1) Question-driven graph-aware retrieval When a question is received from the user, the question is first semantically parsed to identify the equipment type, system scope, fault phenomenon or parameter index involved in the question, and the semantic information is mapped to one or more graph query starting nodes; (2) Node-level and subgraph-level joint retrieval Centered on the query starting node, the knowledge graph is traversed according to the preset relationship type to obtain several nodes and their relationships related to the question semantics, and a question-related subgraph is constructed; the subgraph includes at least: multiple types of nodes, namely equipment, fault, procedure and case; hierarchical relationship, causal relationship and dependency relationship between nodes; (3) Subgraph semantic aggregation and redundancy removal The node content in the question-related subgraph is semantically aggregated and the following processing is performed: nodes from different documents but with the same or similar semantics are merged; the repeated descriptions of the same fault or parameter are deredundant; the weight is calculated based on the centrality or confidence of the node in the graph; and a cross-document semantic summary is generated based on the aggregation result; (4) Graph semantics to LLM The injection of context will organize cross-document semantic summary, relationship structure information between nodes and key original text references into structured prompt content according to the preset context template, including: key entity list; causal relationship chain; hierarchical structure description; corresponding original text evidence fragments; the structured prompts are injected into the large language model as context input, so that the language model has a global knowledge background when generating answers; (5) Language model generation and verification The large language model generates answer results based on the injected graph semantic context, performs rule verification and reference consistency check on the generated results, and ensures that the answer content is consistent with the knowledge graph.

[0032] In this embodiment, cross-document reasoning question answering is performed based on the GraphRAG global semantic enhancement model, including: identifying the device, system, and parameters to which the user's question belongs; locating relevant subgraphs in the knowledge graph; automatically expanding reasoning along relational paths; combining the original text content retrieved by RAG; using LLM to fuse the results to generate the final answer; and realizing the integration of steps across multiple procedures or the understanding of related questions across systems, as well as the difference analysis between different clauses.

[0033] In this embodiment, it also includes: a dynamic updating and continuous learning mechanism for professional knowledge, including supporting online updates of knowledge; automatically generating new knowledge nodes based on new cases; and automatically supplementing weak knowledge areas based on user questions and answers.

[0034] Example 2 The method for multi-source fusion and global semantic understanding of power generation professional knowledge based on GraphRAG provided by the present invention includes the following steps: (1) Automatic structuring of multi-source documents Automatically parse multiple data sources in the power generation industry, including: operation logs, maintenance reports, drawings (automatic OCR + block structuring) in PDF, Word, PPT, scanned documents and other formats, operation procedures, parameter ledgers, accident analysis documents.

[0035] The following functions are achieved using technologies such as NLP and OCR: paragraph parsing and heading level recognition, semantic unit segmentation (knowledge blocks), table extraction and parameter standardization, cross-document reference relationship recognition (referenced chapters, device numbers, etc.), and professional terminology recognition and synonym normalization.

[0036] (2) Knowledge graph construction based on professional rules: A knowledge graph suitable for the power generation industry is established, including: Node type: equipment (boiler, steam turbine, electrical equipment, thermal control device, etc.), fault phenomenon, fault cause, operation steps, parameter index, regulations and clauses, case report, components and models. The graph is automatically constructed through entity extraction, relationship extraction and professional rule matching.

[0037] (3) Construction of GraphRAG Global Semantic Enhancement Model: This invention adopts GraphRAG technology to combine knowledge graph with large-scale language model to achieve global understanding: The model performs node-level and subgraph-level retrieval in the knowledge graph based on user questions and obtains global knowledge domain across documents; it can also aggregate and remove redundancy from the content of multiple nodes and multiple documents, and automatically generate "cross-document semantic summary"; the subgraph summary, hierarchical relationship, and causal chain structure are injected into the context of LLM to give the model a "global knowledge background". (4) Cross-document reasoning question answering system: When answering professional questions, the system performs the following steps: identify the device, system, and parameters to which the user question belongs; locate the relevant subgraph in the knowledge graph; automatically expand reasoning along the relational path (causal chain / dependency chain), combine the original text content retrieved by RAG, and use LLM to fuse the results to generate the final answer. It realizes the integration of steps across multiple procedures or the understanding of related questions across systems, and the analysis of differences between different clauses.

[0038] (5) Dynamic updating and continuous learning mechanism of professional knowledge This invention supports online updating of knowledge: such as automatic re-parsing after document update; automatic formation of new knowledge nodes based on new cases; and automatic supplementation of weak knowledge areas based on user questions and answers.

[0039] Example 3: Data Collection and Preprocessing Structured Analysis: Clarify and organize the scope of knowledge resources, including the collection of unstructured documents and multimodal field image data, such as power industry expert knowledge, fault cases, standards and specifications, and operating condition records.

[0040] After identifying the data source, the data file is cleaned, converted to a unified format, and subjected to noise reduction and missing value filling.

[0041] The document is then parsed, including tasks such as heading level recognition, paragraph segmentation, table parsing, professional OCR, and cross-document reference recognition (e.g., automatically linking to the corresponding content in Appendix 1 in the procedure). The final structured knowledge units and relational knowledge graph are constructed, including the following nodes and relationships: Node examples (excerpt): Equipment nodes: Turbine body → Rotor → Fourth bearing; Speed ​​control system, lubrication system; Fault phenomenon nodes: Increased vibration; Local temperature rise; Fault cause nodes: Bearing wear; Rotor imbalance; Unstable oil film; Loose foundation; Parameter nodes: Vibration value, mm / s; Oil temperature, ℃; Document nodes: Operating procedure chapters; Maintenance record paragraphs; Case node: Historical accident "20210327 Unit 1 rotor imbalance of a power plant"; The system automatically extracts the following relationships: "Equipment – ​​Parameter": Fourth bearing → Related → Vibration value; "Phenomenon – Cause": Increased vibration → Possible cause → Rotor imbalance; "Cause – Case": Rotor imbalance → Typical case → 2021 accident report; "Procedure clause – Step": Procedure 3.2.1 → Recommended inspection → Bearing oil temperature inspection; "Drawing – Equipment": Drawing number T-14 → Corresponding equipment → The fourth bearing; finally, a sub-diagram centered on "steam turbine vibration" is obtained.

[0042] Taking "Analysis of Causes of Excessive Vibration in Steam Turbines" as an example, GraphRAG global semantic enhancement and retrieval: A user asks: "What are the possible causes of increased vibration in a steam turbine?" The system identifies keywords, and the large model categorizes the user's question as: Equipment = Steam Turbine; Phenomenon = Abnormal Vibration; Domain = Steam Turbine Specialty; Requirement: Cause + Troubleshooting Process.

[0043] Automatic graph retrieval of associated nodes: bearing wear; oil temperature rise; rotor imbalance; protection logic trigger records; fault chains of historical cases.

[0044] GraphRAG automatically aggregates relevant document content from each node, such as: turbine operation procedures; maintenance weekly reports; historical accident reports; vibration monitoring parameter standards.

[0045] LLM automatically generates causal analysis reports based on the graph relationship chain, including: possible causes sorted by probability; why these causes are relevant; references to relevant procedures and clauses; comparison of historical cases; recommended inspection steps; enabling users to obtain answers with truly "expert-level understanding".

[0046] Example 4, as shown in Figure 2, describes the GraphRAG-based multi-source fusion and global semantic understanding system for power generation expertise provided by this invention. It includes: a data parsing unit for automatically parsing various data sources in the power generation industry; a knowledge graph building unit for establishing a knowledge graph suitable for the power generation industry based on the parsed data sources; a model building unit that uses GraphRAG technology to combine the knowledge graph with a large-scale language model to construct a GraphRAG global semantic enhancement model; and a reasoning-based question answering unit for performing cross-document reasoning-based question answering based on the GraphRAG global semantic enhancement model.

[0047] In the data parsing unit of this embodiment, multiple data sources in the power generation industry are automatically parsed, including: operation logs, maintenance reports, historical drawings, parameter ledgers, and accident analysis documents in PDF, Word, PPT, and scanned formats; NLP and OCR technologies are used to automatically parse these multiple data sources in the power generation industry, including: paragraph parsing and heading level recognition, semantic unit segmentation, table extraction and parameter standardization, cross-document reference relationship recognition, technical terminology recognition, and synonym normalization.

[0048] Example 5: The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the GraphRAG-based multi-source fusion and global semantic understanding method for power generation expertise.

[0049] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0050] This application is described with reference to flowchart illustrations and / or block diagrams of methods, systems, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a system for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0051] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0052] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0053] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0054] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for multi-source fusion and global semantic understanding of power generation expertise based on GraphRAG, characterized in that, include: The system automatically parses multiple data sources in the power generation industry; based on the parsed data sources, it establishes a knowledge graph suitable for the power generation industry; using GraphRAG technology, it combines the knowledge graph with a large-scale language model to construct a GraphRAG global semantic enhancement model; and based on the GraphRAG global semantic enhancement model, it performs cross-document reasoning question answering.

2. The method for multi-source fusion and global semantic understanding of power generation expertise based on GraphRAG according to claim 1, characterized in that, Automatic parsing of various data sources in the power generation industry, including: operation logs, maintenance reports, historical drawings, operation procedures, parameter ledgers, and accident analysis documents in PDF, Word, PPT, and scanned formats; automatic parsing using NLP and OCR technologies, including: paragraph parsing and heading level recognition, semantic unit segmentation, table extraction and parameter standardization, cross-document reference relationship recognition, technical terminology recognition, and synonym normalization.

3. The method for multi-source fusion and global semantic understanding of power generation expertise based on GraphRAG according to claim 2, characterized in that, Based on various data sources from the parsed power generation industry, a knowledge graph suitable for the power generation industry is established, including: node types: equipment, fault phenomena, fault causes, operating procedures, parameter indicators, procedure clauses, case reports, components and models, where equipment includes boilers, steam turbines, electrical equipment and thermal control devices; automatic graph construction is achieved through entity extraction, relation extraction and professional rule matching; in the knowledge graph construction process, the following steps are executed sequentially: Entity extraction: Based on the named entity recognition model and industry lexicon, candidate entities are extracted from the structured parsing results; and according to predefined node type rules, the candidate entities are instantiated into knowledge graph nodes of the corresponding type; Relation extraction: Based on syntactic analysis and semantic role labeling... It also includes cross-document reference relationships, automatically identifying semantic relationships between entities, including: "equipment – ​​occurrence – fault phenomenon"; "fault phenomenon – corresponding – fault cause"; "fault cause – associated – operation steps"; "procedure clause – stipulate – operation steps"; "case report – involved – equipment / fault"; professional rule matching: based on the power generation industry expert rule base, the extraction results are verified and supplemented, including: if a fault cause does not match a specific equipment type, the confidence level is reduced or the fault is removed; if the procedure clause clearly stipulates a causal relationship, a corresponding relationship edge is forcibly established; if the same equipment number appears in multiple documents, it is automatically merged into the same equipment node; ultimately, it realizes the automatic construction of knowledge graph nodes and relationships from multiple data sources in the parsed power generation industry.

4. The method for multi-source fusion and global semantic understanding of power generation expertise based on GraphRAG according to claim 3, characterized in that, By employing GraphRAG technology, a GraphRAG global semantic enhancement model is constructed by combining knowledge graphs with large-scale language models. This model includes: performing node-level and subgraph-level retrievals in the knowledge graph based on user questions to obtain a global knowledge domain across documents; or aggregating and deduplicating the content of multiple nodes and documents to automatically generate a "cross-document semantic summary"; and injecting subgraph summaries, hierarchical relationships, and causal chain structures into the context of the LLM to give the model a "global knowledge background".

5. The method for multi-source fusion and global semantic understanding of power generation expertise based on GraphRAG according to claim 4, characterized in that, The construction method of the GraphRAG global semantic enhancement model includes the following steps: (1) Problem-driven graph-aware retrieval When a problem is received from the user, the problem is first semantically parsed to identify the equipment type, system scope, fault phenomenon or parameter index involved in the problem, and the semantic information is mapped to one or more graph query starting nodes; (2) Joint retrieval at the node level and subgraph level Centered on the query starting node, the knowledge graph is traversed according to the preset relationship type to obtain several nodes and their relationships related to the problem semantics, and a problem-related subgraph is constructed; the subgraph includes at least: multiple types of nodes, namely equipment, fault, procedure and case; hierarchical relationship, causal relationship and dependency relationship between nodes; (3) Semantic aggregation and redundancy removal of subgraphs Semantic aggregation of the node content in the problem-related subgraphs is performed, and the following processing is performed: nodes from different documents but with the same or similar semantics are merged; redundancy removal is performed on repeated descriptions of the same fault or parameter; weights are calculated based on the centrality or confidence of the nodes in the graph; and cross-document semantic summaries are generated based on the aggregation results; (4) Graph semantics to LLM The injection of context will organize cross-document semantic summary, relationship structure information between nodes and key original text references into structured prompt content according to the preset context template, including: key entity list; causal relationship chain; hierarchical structure description; corresponding original text evidence fragments; the structured prompts are injected into the large language model as context input, so that the language model has a global knowledge background when generating answers; (5) Language model generation and verification The large language model generates answer results based on the injected graph semantic context, performs rule verification and reference consistency check on the generated results, and ensures that the answer content is consistent with the knowledge graph.

6. The method for multi-source fusion and global semantic understanding of power generation expertise based on GraphRAG according to claim 5, characterized in that, Based on the GraphRAG global semantic enhancement model, cross-document reasoning question answering is performed, including: identifying the device, system, and parameters to which the user's question belongs; locating relevant subgraphs in the knowledge graph; automatically expanding reasoning along relational paths; combining RAG to retrieve the original text content; using LLM to fuse the results to generate the final answer; and realizing the integration of steps across multiple procedures or the understanding of related questions across systems, as well as the difference analysis between different clauses.

7. The method for multi-source fusion and global semantic understanding of power generation expertise based on GraphRAG according to claim 1, characterized in that, Also includes: The mechanism for dynamic updating and continuous learning of professional knowledge includes supporting online updates of knowledge; automatically generating new knowledge nodes based on new cases; and automatically supplementing weak knowledge areas based on user questions and answers.

8. A GraphRAG-based system for multi-source fusion and global semantic understanding of power generation expertise, characterized in that: include: The data parsing unit is used to automatically parse various data sources in the power generation industry; the knowledge graph building unit is used to build a knowledge graph suitable for the power generation industry based on the parsed data sources. The model building unit uses GraphRAG technology to combine knowledge graphs with large-scale language models to build a GraphRAG global semantic enhancement model; the reasoning question answering unit performs cross-document reasoning question answering based on the GraphRAG global semantic enhancement model.

9. The GraphRAG-based multi-source fusion and global semantic understanding system for power generation expertise according to claim 8, characterized in that, The data parsing unit automatically parses various data sources in the power generation industry, including: operation logs, maintenance reports, historical drawings, operation procedures, parameter ledgers, and accident analysis documents in PDF, Word, PPT, and scanned formats; it employs NLP and OCR technologies to automatically parse these data sources, including: paragraph parsing and heading level recognition, semantic unit segmentation, table extraction and parameter standardization, cross-document reference relationship recognition, and professional terminology recognition and synonym normalization.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the GraphRAG-based multi-source fusion and global semantic understanding method for power generation expertise as described in any one of claims 1-7.

Images