Method and system for constructing power industry dynamic knowledge base based on large language model

Abstract

The application discloses a power industry dynamic knowledge base construction method and system based on a large language model, relates to the field of electric digital data processing, and comprises the following steps: receiving a power field document imported by a user, and converting the power field document into an electronic document; calling a large language model to analyze the electronic document, extracting professional term definitions, chart position information and document logical levels in the electronic document, and generating knowledge units containing multiple contexts; connecting the knowledge units into directed knowledge links, calculating semantic similarity between adjacent knowledge units in the directed knowledge links, segmenting the directed knowledge links, and obtaining multiple knowledge topics; receiving a knowledge retrieval instruction, determining a target knowledge topic corresponding to the knowledge retrieval instruction; extracting a target knowledge unit corresponding to the target knowledge topic, and generating a knowledge retrieval result based on a target context of the target knowledge unit. The application can optimize knowledge base construction and guarantee the integrity of knowledge base data.

Description

Technical Field

[0001] This application relates to the field of electrical digital data processing, and in particular to a method and system for constructing a dynamic knowledge base for the power industry based on a large language model. Background Technology

[0002] With the deepening of information technology construction in the power industry, power companies have accumulated a large amount of diverse documents, including equipment parameters, technical standards, and operation and maintenance manuals. These documents contain important professional knowledge and are of great value in ensuring the safe and stable operation of the power system and improving operation and maintenance efficiency. Establishing a comprehensive power industry knowledge base to achieve efficient management of documents and knowledge reuse has become an important requirement for the industry's development.

[0003] In related technologies, the power industry knowledge base is constructed using an intelligent document processing system. This type of system employs deep learning models for document parsing, capable of recognizing various content formats such as text, tables, and graphics. The system uses a graph database to store the relationships between documents, establishing a multi-dimensional knowledge index. For knowledge organization, the system uses natural language processing technology to perform topic clustering and tag extraction on document content, achieving automatic document classification. During retrieval, the system combines vector retrieval and keyword matching mechanisms, filtering relevant documents through semantic similarity calculation. Knowledge updates are semi-automated; the system automatically extracts key information from new documents, which is then updated to the knowledge base after expert review.

[0004] However, because professional documents in the power industry often contain a large number of mixed diagrams, technical terms, and multi-layered logical relationships, the system is prone to causing context breaks when parsing these documents. For example, when processing cross-page tables or multi-condition operating procedures in equipment manuals, the system can only segment them into independent fragments for storage, making it difficult to restore the complete knowledge context during retrieval. Summary of the Invention

[0005] This application provides a method and system for constructing a dynamic knowledge base for the power industry based on a large language model, which is used to optimize knowledge base construction and ensure the integrity of knowledge base data.

[0006] Firstly, this application provides a method for constructing a dynamic knowledge base for the power industry based on a large language model, applied to a data processing system. The method includes: receiving power industry documents imported by a user and converting them into electronic documents containing unique document identifiers; using a large language model to parse the electronic documents, extracting definitions of technical terms, chart location information, and document logical hierarchy to generate knowledge units containing multiple contexts; connecting the knowledge units into directed knowledge links and calculating the semantic similarity between adjacent knowledge units in the directed knowledge links; segmenting the directed knowledge links based on semantic similarity to obtain multiple knowledge topics; receiving a knowledge retrieval instruction and determining the target knowledge topic corresponding to the instruction; extracting the target knowledge units corresponding to the target knowledge topics and generating knowledge retrieval results based on the target context of the target knowledge units.

[0007] In the above embodiments, the data processing system can parse documents in the power field, extract definitions of professional terms, chart locations, and document logical levels, and generate knowledge units containing complete context. By constructing directed knowledge links and calculating semantic similarity, it can achieve accurate association of knowledge units and topic clustering. During retrieval, the system can generate structured retrieval results based on the context of knowledge units, effectively ensuring the integrity of knowledge base data and retrieval accuracy.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the step of receiving a power sector document imported by a user and converting the power sector document into an electronic document containing a unique document identifier specifically includes: receiving the power sector document imported by the user; extracting the document attribute information of the power sector document to obtain the document type, creation time, and version number; selecting the corresponding layout analysis model based on the document type to identify the layout structure of the power sector document, including text paragraphs, tables, and graphics, to obtain a document structure tree; generating a unique document identifier based on the document structure tree and the version number; and encapsulating the layout structure and the unique document identifier into an electronic document in a preset standard format.

[0009] In the above embodiments, the data processing system performs in-depth layout analysis on the imported documents, extracts document attribute information, and identifies complex layout structures such as text, tables, and graphics; by generating unique identifiers and standard format conversion, it ensures the uniqueness and structured storage of documents in the knowledge base.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, before calling the large language model to parse the electronic document, extract the definitions of professional terms, chart location information and document logical hierarchy in the electronic document, and generate knowledge units containing multiple contexts, the method further includes: constructing a training sample set based on documents in the power field; the training sample set includes definitions of professional terms, equipment operation steps and fault handling procedures; extracting professional terms from the documents in the power field and constructing a power field lexicon; adding the power field lexicon to the pre-trained language model, and using the training sample set to incrementally train the pre-trained language model to obtain the large language model.

[0011] In the above embodiments, the data processing system constructs a professional training sample set based on documents in the power field, including terminology definitions, operation steps, and fault handling procedures; by incrementally training the pre-trained language model, the model is equipped with the ability to understand power professional knowledge, thereby improving the accuracy of document parsing and knowledge extraction.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting the target knowledge unit corresponding to the target knowledge topic and generating knowledge retrieval results based on the target context of the target knowledge unit specifically includes: extracting the target knowledge unit corresponding to the target knowledge topic, obtaining the document location information of the target knowledge unit, and constructing the target context for knowledge retrieval; extracting the graphic content and table content of the target location according to the target context to generate a multimodal knowledge sequence; arranging the multimodal knowledge sequence hierarchically according to the document logical hierarchy of the target knowledge unit to obtain a structured knowledge body; and generating knowledge retrieval results according to the hierarchical relationship of the structured knowledge body.

[0013] In the above embodiments, the data processing system performs multimodal knowledge organization on the retrieval results, extracts graphic and tabular content and generates structured descriptions; it also arranges the knowledge sequence hierarchically based on the document's logical hierarchy to construct a clear knowledge system, thus realizing the structured presentation of the retrieval results.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting graphic content and table content of the target location based on the target context to generate a multimodal knowledge sequence specifically includes: locating and extracting the primitive features of the target graphic and the cell layout features of the target table based on the graphic location information in the target context; identifying the graphic components and connections in the target graphic, and converting the graphic components and connections into a structured graphic description; identifying the header level and data area of ​​the target table, extracting the relationships between cells, and converting them into a structured table description; and combining the structured graphic description and the structured table description with the corresponding text description blocks to generate a multimodal knowledge sequence.

[0015] In the above embodiments, the data processing system accurately identifies the component features and relationships in the charts and graphs, converting complex graphic and tabular information into structured descriptions; through intelligent combination with text descriptions, it generates a complete multimodal knowledge sequence, ensuring the semantic integrity of the graphic and textual information.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after connecting knowledge units into directed knowledge links and calculating the semantic similarity between adjacent knowledge units in the directed knowledge links, the method further includes: calculating the out-degree and in-degree values ​​of each knowledge unit in the directed knowledge links to obtain node weights; constructing a transition probability matrix of the directed knowledge links based on the node weights and calculating the page walk score; identifying key paths in the directed knowledge links based on the node weights and the page walk score; and deleting redundant links with node weights lower than a preset weight threshold based on the key paths and supplementing missing associations to optimize the directed knowledge links.

[0017] In the above embodiments, the data processing system calculates the in-degree and out-degree values ​​and weights of nodes in the knowledge link, constructs a transition probability matrix for importance analysis, and improves the connectivity and retrieval efficiency of the knowledge network by identifying critical paths and optimizing redundant links.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the step of calculating the out-degree and in-degree values ​​of each knowledge unit in the directed knowledge link to obtain the node weight specifically includes: traversing each knowledge unit in the directed knowledge link, counting the number of edges pointing to other knowledge units to obtain the out-degree value; traversing each knowledge unit in the directed knowledge link, counting the number of edges from other knowledge units to obtain the in-degree value; and performing a weighted calculation based on the out-degree and in-degree values ​​to obtain the node weight.

[0019] In the above embodiments, the data processing system calculates node weight indicators by statistically analyzing the connection relationships between knowledge units; and accurately assesses the importance of knowledge units by weighted calculation of out-degree and in-degree values, providing a basis for knowledge network optimization.

[0020] In a second aspect, embodiments of this application provide a data processing system comprising: one or more processors and a memory; the memory is coupled to the one or more processors and is used to store computer program code, the computer program code including computer instructions, wherein the one or more processors invoke the computer instructions to cause the data processing system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a data processing system, cause the data processing system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a data processing system, cause the data processing system to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the data processing system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. By adopting a power document parsing and knowledge link construction method based on a large language model, the system can intelligently identify professional terms, charts, and logical levels in documents and organize them into knowledge units containing complete context. By calculating the semantic similarity between knowledge units to construct directed links, accurate knowledge association is achieved. This effectively solves the problems of fragmented context and inaccurate knowledge association in related technologies, thereby ensuring the integrity of knowledge base data and the semantic coherence of retrieval results.

[0026] 2. By employing a multimodal knowledge extraction and organization method based on document location information, the system can accurately locate and extract graphic and tabular content at the target location, convert it into a structured description, and intelligently combine it with text descriptions. By hierarchically arranging the multimodal knowledge sequence, a clear knowledge system is constructed. This effectively solves the problems of fragmented graphic and textual information and ambiguous knowledge hierarchy in related technologies, thereby realizing the multimodal fusion and structured presentation of retrieval results.

[0027] 3. By adopting a knowledge link optimization method based on node weights and page traversal, the system can accurately assess the importance of knowledge units by calculating the in-degree and transition probability matrices of nodes; by identifying critical paths and optimizing redundant links, the quality of the knowledge network is improved; and the problems of knowledge link redundancy and inaccurate association in related technologies are effectively solved, thereby achieving efficient connectivity and accurate retrieval of the knowledge network. Attached Figure Description

[0028] Figure 1This is a flowchart illustrating a method for constructing a dynamic knowledge base for the power industry based on a large language model, as described in this application.

[0029] Figure 2 This is another flowchart illustrating the method for constructing a dynamic knowledge base for the power industry based on a large language model in this application embodiment;

[0030] Figure 3 This is a schematic diagram of the physical device structure of a data processing system in an embodiment of this application. Detailed Implementation

[0031] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0033] To facilitate understanding, the application scenarios of the embodiments of this application are described below.

[0034] The operations and maintenance department of a power company needs to manage a large number of power equipment documents, including equipment manuals, operating procedures, and troubleshooting manuals. These documents are in various formats and contain a large number of technical terms, diagrams, and operating procedures. When troubleshooting equipment failures, maintenance personnel need to quickly retrieve relevant information, but due to the scattered content, inconsistent formats, and complex relationships between professional knowledge, it is often difficult to find the required information in a timely manner, affecting the efficiency of troubleshooting. This is especially true for new employees, who struggle to quickly grasp the knowledge structure in the face of such a large document library, further hindering their work efficiency.

[0035] In related technologies, keyword matching and simple text classification can be used to store, manage, and retrieve documents in the power industry. However, this method relies primarily on the surface features of the documents and cannot deeply understand their professional content and knowledge relationships. The following section introduces a scenario using a dynamic knowledge base construction method for the power industry based on a large language model, which is part of this related technology.

[0036] Existing document management systems primarily rely on keyword matching for retrieval and simple text classification to organize documents. For example, a substation's document system categorizes equipment manuals by equipment type and document type, allowing retrieval by title and keywords. When maintenance personnel need to query the protection setting scheme for a specific circuit breaker model, the system can only return relevant documents based on keywords. It cannot understand the contextual relationships of technical terms or connect related knowledge scattered across different documents, forcing users to manually browse multiple documents to obtain complete information.

[0037] The method for constructing a dynamic knowledge base for the power industry based on a large language model, as described in this application, achieves deep knowledge association and accurate retrieval through intelligent parsing of document content, construction of directed knowledge links, and calculation of semantic similarity. This not only improves retrieval efficiency but also ensures the integrity and accuracy of knowledge transfer. The following describes a scenario where the method for constructing a dynamic knowledge base for the power industry based on a large language model is used.

[0038] After adopting the proposed solution, the document management system can intelligently parse the professional content of power-related documents. When a new equipment manual is uploaded, the system automatically identifies the document structure, extracts technical terms, diagrams, and operating procedures, and establishes knowledge associations. For example, when processing the setting scheme of a certain type of relay protection device, the system can not only locate the specific parameter setting instructions but also automatically associate related principle explanations, wiring diagrams, and debugging steps, forming a complete knowledge chain. Maintenance personnel can directly obtain structured knowledge content through retrieval, quickly understanding the working principles and operating requirements of the equipment.

[0039] As can be seen, the method for constructing a dynamic knowledge base for the power industry based on a large language model in this application can not only achieve dynamic updates of the knowledge network, but also effectively solve the problems of knowledge fragmentation and inaccurate association, thereby realizing the systematic organization and intelligent service of knowledge in the power field.

[0040] To facilitate understanding, the method provided in this implementation will be described in detail below, using the above scenario as an example. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating a method for constructing a dynamic knowledge base for the power industry based on a large language model, as described in this application.

[0041] S101. Receive power sector documents imported by the user and convert them into electronic documents containing unique document identifiers.

[0042] Among them, power sector documents refer to technical documents related to power systems, including but not limited to equipment manuals, technical specifications, operation manuals, and troubleshooting guidelines; document unique identifiers refer to string codes used to uniquely distinguish different documents, generated through document attribute information and content characteristics; electronic documents are used to represent standardized digital format documents, containing structured text content and layout information.

[0043] The data processing system executes this step when it receives power sector documents imported by a user through the document upload interface. Specifically, the system first reads the binary data stream of the imported document and identifies the document format type; then, it extracts the document's metadata information, including attributes such as creation time, author, and version number; next, it uses the corresponding layout analysis model to process the document content, identifying layout elements such as text paragraphs, tables, and graphics, and constructing a document structure tree; finally, it generates a unique identifier based on the document attributes and structural features, and encapsulates the parsing results into a standard format electronic document.

[0044] In some embodiments, document format conversion and unique identifier generation can be achieved in multiple ways: Optionally, the data processing system can adopt a rule-based approach, first using regular expressions to extract key information such as document titles and chapters, then combining the document creation time and version number to generate a hash value as a unique identifier, and finally converting the content into Markdown format; Optionally, the data processing system can adopt a deep learning approach, using a pre-trained document understanding model to identify document structure and semantic information, generating a unique identifier based on document feature vectors, and outputting a structured representation in JSON format. It is understood that other document parsing and identifier generation methods can also be used, and are not limited here.

[0045] In practical applications, issues may arise due to non-standard document formatting leading to parsing errors. To address this, the data processing system first preprocesses and standardizes the document, including correcting non-standard formatting, standardizing character encoding, and handling special symbols. Simultaneously, the system maintains a format conversion rule library, developing corresponding handling strategies for different types of anomalies. For example, for tables spanning multiple pages, the system analyzes the relationship between table borders and cell correspondences to merge them into a complete data structure; for non-standard chapter hierarchies, the system reconstructs the document's hierarchical structure based on text formatting and indentation.

[0046] S102. Call the large language model to parse the electronic document, extract the definitions of professional terms, the location information of charts and graphs and the logical hierarchy of the document, and generate knowledge units containing multiple contexts.

[0047] Among them, the large language model represents a pre-trained language model trained with knowledge of the power industry, used to understand the content of professional documents; the professional terminology definition refers to the proper nouns in the power industry and their explanations; the chart location information is used to represent the spatial layout and reference relationships of graphics and tables in the document; the document logical hierarchy represents the hierarchical organization structure of the content; and the knowledge unit refers to the smallest knowledge fragment with complete semantics, containing contextual information.

[0048] The data processing system performs this step after completing the electronic document conversion. Specifically, the system first segments the electronic document into sentence-level text fragments while retaining structural information such as paragraphs and chapters; then, it calls a domain-trained large language model to identify technical terms in the text and extract their definitions; next, it locates the chart elements in the document and analyzes their reference relationships with surrounding text; then, it constructs the logical hierarchical relationships of the content based on the document's structure tree; finally, it combines related text fragments, term definitions, and chart references into knowledge units and records their contextual information within the document.

[0049] In some embodiments, intelligent parsing of professional content can be achieved in several ways: Optionally, the data processing system can adopt a rule-based and template-based approach, using a predefined terminology dictionary and text pattern matching rules to identify professional content, determining the boundaries of terminology definitions through syntactic analysis, and constructing a logical hierarchy based on document titles and chapter numbers; Optionally, the data processing system can adopt a deep learning approach, using sequence labeling models to identify professional terms and definitions, modeling the hierarchical structure of documents through graph neural networks, and capturing the relationship between charts and text using attention mechanisms. It is understood that other content understanding and knowledge extraction methods can also be used, and are not limited here.

[0050] In practical applications, ambiguity in technical terms may lead to parsing errors. To address this, the data processing system maintains a domain-specific terminology knowledge base, recording multiple expressions and contextual scenarios for each term. When encountering ambiguous terms, the system analyzes the contextual information of the sentence containing the term, including co-occurring words, syntactic structure, and thematic background, eliminating ambiguity through multi-feature fusion. For example, for the term "voltage," the system determines whether it refers to an electrical parameter or the operating status of equipment based on the context, thereby extracting the appropriate definition.

[0051] S103. Connect knowledge units into directed knowledge links and calculate the semantic similarity between adjacent knowledge units in the directed knowledge links.

[0052] Among them, directed knowledge links represent a network of directional relationships between knowledge units, used to describe the transmission and dependency of knowledge; semantic similarity refers to the degree of closeness between two knowledge units in meaning, calculated through distance measurement in vector space; adjacent knowledge units represent pairs of nodes directly connected in the knowledge link; knowledge association is used to represent the logical relationship between different knowledge units, including sequence, inclusion relationship, reference relationship, etc.

[0053] The data processing system performs this step after extracting the knowledge units. Specifically, the system first analyzes the explicit relationships between knowledge units, including cross-references and terminology definition references in documents; then, based on the logical hierarchy of the documents, it establishes hierarchical relationships between knowledge units; next, it uses a pre-trained language model to convert the knowledge units into semantic vector representations; then, it calculates the cosine similarity between adjacent knowledge unit vectors and sets a similarity threshold to filter valid connections; finally, it constructs a directed knowledge link network based on explicit relationships and semantic similarity.

[0054] In some embodiments, knowledge link construction can be implemented in multiple ways: Optionally, the data processing system can adopt a rule-based approach, using document structure and referencing relationships to establish initial links, calculating the association strength between nodes using traditional text similarity algorithms such as TF-IDF, and optimizing the link structure based on heuristic rules; Optionally, the data processing system can adopt a deep learning approach, using a bidirectional encoder to obtain semantic representations of knowledge units, calculating relevance scores between nodes through an attention mechanism, and combining graph neural networks to learn the neighborhood features of nodes. It is understood that other knowledge association and similarity calculation methods can also be used, and are not limited here.

[0055] In practical applications, inaccurate semantic similarity calculations may lead to incorrect connections. To address this, the data processing system combines multiple similarity metrics, including lexical overlap, topic similarity, and structural similarity. The system first calculates similarity indicators across different dimensions and then obtains a comprehensive score through weighted fusion. Simultaneously, the system considers the domain characteristics of knowledge units, such as the density of technical terms and the type of technical parameters, dynamically adjusting the similarity calculation strategy. For example, for equipment parameter descriptions, the system prioritizes matching numerical features; for operational procedure descriptions, it focuses more on the consistency of the step sequence.

[0056] S104. Segment the directed knowledge link based on semantic similarity to obtain multiple knowledge topics.

[0057] Segmentation refers to dividing closely related groups of knowledge units into independent subgraph structures; knowledge topics represent sets of knowledge units with similar semantic content or common topic features; semantic clustering is used to represent the automatic grouping process of knowledge units based on semantic similarity; topic boundaries refer to the dividing lines between different knowledge topics, which are determined by semantic similarity thresholds.

[0058] The data processing system performs this step after constructing the directed knowledge link. Specifically, the data processing system first constructs a similarity matrix of knowledge units based on the calculated semantic similarity; then, it uses a community detection algorithm to detect tightly connected subgroups in the knowledge link; next, it analyzes the topic characteristics of each subgroup, including the distribution of key terms, core concepts, and technical fields; then, it optimizes topic boundaries and handles transitional knowledge units across topics; finally, it generates descriptive labels for each knowledge topic and establishes relationships between topics.

[0059] In some embodiments, knowledge topic partitioning can be achieved in multiple ways: Optionally, the data processing system can employ a density-based clustering method, using the DBSCAN algorithm to identify dense regions of knowledge units based on semantic similarity, forming knowledge topics through neighborhood expansion, and determining topic boundaries based on connectivity analysis; alternatively, the data processing system can employ a graph-based partitioning method, using a spectral clustering algorithm to convert knowledge links into Laplacian matrices, obtaining the optimal partitioning scheme through eigenvalue decomposition, and optimizing topic partitioning by combining modularity evaluation. It is understood that other topic discovery and boundary optimization methods can also be used, and are not limited here.

[0060] In practical applications, the ambiguity of knowledge topic boundaries may lead to inaccurate classification. To address this, the data processing system employs a multi-level topic segmentation strategy. The system first identifies core topic regions at a high semantic similarity threshold, then gradually lowers the threshold to discover transitional regions. For knowledge units located at topic boundaries, the system analyzes their association strength with multiple topics, allowing them to belong to multiple related topics simultaneously. For example, a knowledge unit describing a device's protection function may be associated with both the topics of device structure and protection principle; the system preserves this multi-topic attribute to reflect the completeness of the knowledge.

[0061] S105. Receive knowledge retrieval instructions and determine the target knowledge topic corresponding to the knowledge retrieval instructions.

[0062] Among them, the knowledge retrieval instruction represents the query request entered by the user, which includes search keywords and query conditions; the target knowledge topic refers to the set of knowledge units that are most semantically relevant to the retrieval instruction; the topic matching degree is used to represent the degree of relevance between the retrieval instruction and the knowledge topic; and the query intent represents the actual needs behind the user's retrieval request, which is identified through semantic understanding.

[0063] The data processing system executes this step upon receiving a user's search request. Specifically, the system first performs semantic understanding of the search command, identifying core concepts and query intent; then, it converts the search command into a vector representation and calculates the semantic similarity with each knowledge topic; next, it ranks the knowledge topics based on topic matching degree and selects the topics with the highest similarity; then, it analyzes the relationships between topics to expand the scope of related topics; finally, it integrates multiple features to determine the final target knowledge topic set.

[0064] In some embodiments, retrieval intent understanding and topic matching can be achieved in multiple ways: Optionally, the data processing system can adopt a rule-based approach, using keyword matching and syntactic analysis to extract query intent, selecting target topics through predefined topic mapping rules, and combining multiple matching conditions based on Boolean logic; alternatively, the data processing system can adopt a deep learning approach, using an intent classification model to understand retrieval needs, calculating the relevance score between the query and the topic through a semantic retrieval model, and combining a multi-head attention mechanism to capture multiple aspects of the query. It is understood that other retrieval understanding and topic matching methods can also be used, and are not limited here.

[0065] In practical applications, ambiguous search commands may lead to inaccurate topic matching. To address this, the data processing system implements an interactive query understanding strategy. When an unclear query intent is detected, the system generates clarifying questions to guide the user to provide more information. Simultaneously, the system maintains the user's search history, utilizing contextual information to aid in intent understanding. For example, when a user searches for "protection device parameter settings," the system will determine the specific device type based on previous search records or prompt the user to select a specific protection type, thereby accurately locating the target knowledge topic.

[0066] S106. Extract the target knowledge units corresponding to the target knowledge topics, and generate knowledge retrieval results based on the target context of the target knowledge units.

[0067] Among them, the target knowledge unit represents a knowledge fragment directly related to the retrieval requirements; the target context refers to the contextual information of the knowledge unit in the original document, including adjacent paragraphs, related charts, etc.; the knowledge retrieval results are used to represent the structured knowledge content returned by the system; and the context association represents the semantic connection relationship between the knowledge unit and its surrounding content.

[0068] The data processing system performs this step after determining the target knowledge topic. Specifically, the system first selects the knowledge units most relevant to the search intent from the target knowledge topic; then, based on the document location information of the knowledge units, it extracts their original context content; next, it analyzes the logical relationships between knowledge units and reconstructs the organizational structure of the knowledge content; then, it integrates text descriptions, chart content, and related information to generate a multimodal knowledge representation; finally, it adjusts the display format and level of detail of the content according to the user's search intent to form the final search results.

[0069] In some embodiments, knowledge retrieval results can be generated in multiple ways: Optionally, the data processing system can adopt a template-based approach, using predefined result organization templates to sort knowledge units according to importance and relevance, assembling contextual information through a rule engine to generate hierarchical retrieval results; Optionally, the data processing system can adopt a generative approach, using sequence-to-sequence models to reorganize and summarize knowledge content, integrating text and image information through a multimodal fusion network, and dynamically generating retrieval responses based on user intent. It is understood that other result generation and display methods can also be used, and are not limited here.

[0070] In practical applications, the problem of overly lengthy contextual information leading to less refined search results may arise. To address this, the data processing system employs an adaptive content filtering strategy. The system first assesses the importance of each piece of contextual content to understanding the target knowledge, identifying key information through saliency analysis. For chart content, the system extracts relevant local areas to avoid including irrelevant content. For example, when searching for parameter settings for a protection function, the system prioritizes directly relevant content such as parameter definitions, value ranges, and typical configurations, while appropriately simplifying other functional characteristics of the device, thus ensuring the accuracy and usability of the search results.

[0071] Furthermore, the knowledge network can be continuously optimized and improved with ongoing system use. For example, after handling a certain type of fault multiple times, the system can supplement knowledge connections based on actual application scenarios, organically organizing information such as fault phenomena, diagnostic steps, and handling methods. When new fault cases appear, the system automatically analyzes the connections with existing knowledge and updates the knowledge network. Simultaneously, the system can adaptively adjust the depth and breadth of knowledge display according to the user's professional level, providing basic guidance for novice maintenance personnel and in-depth principle analysis for professional technicians, thus achieving personalized customization of knowledge services.

[0072] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating the method for constructing a dynamic knowledge base for the power industry based on a large language model in this application.

[0073] S201. Receive power sector documents imported by the user and convert them into electronic documents containing unique document identifiers.

[0074] Referring to step S101, the data processing system will convert power sector documents into electronic documents.

[0075] In some embodiments, the data processing system performs attribute extraction and layout analysis on the imported documents, converting them into standard format electronic documents. Specifically, the data processing system receives power sector documents imported by the user, extracts document attribute information from the power sector documents to obtain document type, creation time, and version number; selects the corresponding layout analysis model based on the document type, identifies the layout structure of the power sector documents including text paragraphs, tables, and graphics, and obtains a document structure tree; generates a unique document identifier based on the document structure tree and version number; and encapsulates the layout structure and unique document identifier into a preset standard format electronic document.

[0076] Among them, power field documents refer to technical documents containing professional knowledge of power systems, including equipment manuals, regulations and specifications; document attribute information refers to the metadata set describing the basic characteristics of the document; layout analysis model is used to represent the layout recognition model for different types of documents; document structure tree represents the hierarchical organization structure of document content; layout structure refers to the spatial arrangement relationship of various content elements in the document; document unique identifier is used to represent the unique code that distinguishes different documents.

[0077] The data processing system performs this step when a user uploads power-related documents through the system interface. Specifically, the system first reads the binary stream of the uploaded document, identifies the file format, and extracts metadata information. Then, it loads the corresponding layout analysis model based on the document type to perform structured parsing of the document. Next, it uses deep learning algorithms to identify text blocks, table areas, and graphic areas in the document, analyzing their positional and hierarchical relationships. The parsing results are then organized into a tree structure, and a unique identifier is generated based on structural features and version information. Finally, the document content is converted into a standard format and packaged into a unified electronic document.

[0078] In some embodiments, document parsing and conversion can be achieved in multiple ways: Optionally, the data processing system can adopt a deep learning-based approach, first using an object detection model to detect different regional structures (headers, footers, body text, images, formulas, etc.) in the document page, and then applying appropriate recognition methods to each region: using OCR technology for text recognition in the body text, and using deep learning models such as UniMERNet for formula recognition in the formula region, and finally fusing all recognition results to generate a document feature vector; Optionally, the data processing system can adopt a layout analysis and rule-based approach, recognizing the document structure through predefined templates, extracting region information based on layout analysis, and generating document identifiers using the SHA-256 algorithm. It is understood that other document processing methods can also be used, and are not limited here.

[0079] In practical applications, complex document formats may lead to layout recognition errors. To address this, the data processing system employs a multi-model collaborative parsing strategy. First, a document type recognizer is built to accurately determine the specific type and format characteristics of the document. Then, the most suitable combination of layout analysis models is selected to perform multi-level parsing of the document. Simultaneously, model performance is continuously optimized through rule validation and human feedback. For example, for equipment manuals containing complex charts, the system uses both table recognition and graph analysis models to ensure accurate parsing of various content types.

[0080] S202. Call the large language model to parse the electronic document, extract the professional terminology definitions, chart location information and document logical hierarchy in the electronic document, and generate knowledge units containing multiple contexts.

[0081] Referring to step S102, the data processing system will generate knowledge units.

[0082] In some embodiments, the data processing system constructs professional domain training data and performs incremental model training. Specifically, the data processing system constructs a training sample set based on documents in the power industry. This training sample set includes definitions of professional terms, equipment operation procedures, and fault handling processes. Professional terms are extracted from the documents in the power industry to construct a power industry lexicon. The power industry lexicon is added to the pre-trained language model, and the pre-trained language model is incrementally trained using the training sample set to obtain a large language model.

[0083] Among them, the training sample set represents the labeled data set used for model training; the professional term definition refers to the professional terms in the power field and their explanations; the power field lexicon is a structured dictionary used to represent power professional terms; the pre-trained language model represents the basic model pre-trained on a general corpus; and incremental training refers to the process of performing domain-adaptive training based on the pre-trained model.

[0084] The data processing system performs this step after completing the collection of power-related documents. Specifically, the system first extracts professional knowledge content from the parsed power-related documents, including concept explanations, operational guidelines, and fault cases; then, it constructs a high-quality training corpus through text analysis and expert annotation; next, it uses a large language model to extract and construct a professional vocabulary from the document content, including terminology definitions, usage scenarios, and related information; then, it integrates the vocabulary into the vocabulary system of the pre-trained model; finally, it uses the constructed training samples to perform targeted training on the model, improving its understanding of the power-related field.

[0085] In some embodiments, model training and optimization can be achieved in multiple ways: Optionally, the data processing system can adopt a knowledge distillation-based approach, first training a large-scale professional domain model, then transferring domain knowledge to a lightweight model, improving model performance through multi-task learning, and periodically updating the model with new corpora; Optionally, the data processing system can adopt a few-shot learning-based approach, constructing domain knowledge prompt templates, achieving rapid domain adaptation through prompt learning methods, and continuously optimizing model performance by combining active learning strategies. It is understood that other model training methods can also be used, and are not limited here.

[0086] In practical applications, uneven training sample quality may lead to unstable model performance. To address this, the data processing system employs a hierarchical quality control data management strategy. First, a sample evaluation system is established to assess sample quality based on dimensions such as accuracy, completeness, and representativeness. Then, different training strategies are applied to samples of different quality levels: high-quality samples are used for core capability training, while other samples are used for knowledge expansion. Simultaneously, adversarial training is used to improve model robustness. For example, when processing equipment operation instructions, the system pays special attention to the accuracy of key steps and safety prompts, ensuring the model can reliably understand and express important information.

[0087] S203. Connect knowledge units into directed knowledge links and calculate the semantic similarity between adjacent knowledge units in the directed knowledge links.

[0088] Referring to step S103, the data processing system will construct a directed knowledge link.

[0089] S204. Calculate the out-degree and in-degree values ​​of each knowledge unit in the directed knowledge link to obtain the node weight.

[0090] Among them, the out-degree value represents the number of connections from this knowledge unit to other knowledge units; the in-degree value refers to the number of connections from other knowledge units to this knowledge unit; the node weight is used to represent the importance of the knowledge unit in the entire knowledge network; and the connection strength represents the tightness of the relationship between knowledge units.

[0091] The data processing system performs this step after constructing the directed knowledge link. Specifically, the data processing system first traverses each knowledge unit in the knowledge link and counts the number of its outgoing and incoming connections; then, it assigns a weight value to each edge based on the semantic similarity of the connections; next, it calculates the initial weight of the node by combining the in-degree value and the edge weight; then, it adjusts the weights considering the hierarchical depth and specialization of the knowledge unit; finally, it normalizes the weights to obtain the final node weight values.

[0092] In some embodiments, node weights can be calculated in several ways: Optionally, the data processing system can use a graph theory-based approach, iteratively calculating node centrality metrics, including proximity centrality and betweenness centrality, and combining the weighted sum of multiple centrality metrics as the node weight; alternatively, the data processing system can use a random walk-based approach, calculating the node's PageRank value as the base weight, and fine-tuning the weights using expert knowledge rules to obtain a comprehensive score reflecting the importance of knowledge. It is understood that other node importance assessment methods can also be used, and are not limited here.

[0093] In practical applications, imbalances in the connections between knowledge units may lead to biased weight calculations. To address this, the data processing system employs a hierarchical weighted calculation strategy. The system first groups knowledge units according to their professional fields and technical levels, calculating a basic weight within each group. Then, considering the importance of different groups, the weights are adjusted across groups. For example, for knowledge units related to equipment protection, the system increases their weight based on their criticality in safe operation; for knowledge units providing supplementary explanations, the system correspondingly decreases their weight, thus ensuring that node weights more accurately reflect the actual importance of the knowledge.

[0094] In some embodiments, the data processing system calculates node importance by analyzing the connection relationships between knowledge units. That is, the data processing system traverses each knowledge unit in the directed knowledge link, counts the number of edges pointing to other knowledge units, and obtains the out-degree value; it traverses each knowledge unit in the directed knowledge link, counts the number of edges from other knowledge units, and obtains the in-degree value; and performs a weighted calculation based on the out-degree value and the in-degree value to obtain the node weight.

[0095] Among them, the out-degree value represents the number of connections from the knowledge unit to other units; the in-degree value refers to the number of connections from other units to the knowledge unit; the number of edges is used to represent the number of direct associations between knowledge units; the weighted calculation represents the weight calculation process that considers the importance of connections; and the node weight is a quantitative indicator of the importance of the knowledge unit in the entire network.

[0096] The data processing system performs this step after constructing the directed knowledge link. Specifically, the system first traverses all nodes in the knowledge network, counts the number of connections sent by each node, and records this as the out-degree value. Then, it traverses the network again, counts the number of connections pointing to each node, and records this as the in-degree value. Next, it analyzes the semantic strength of the connections and assigns weight coefficients to different types of relationships. Then, it combines the in-degree and out-degree values ​​with the connection weights to calculate the initial weight of the node through a weighted summation. Finally, it performs normalization to obtain the final weight value that reflects the importance of the node.

[0097] In some embodiments, node weight calculation can be implemented in multiple ways: Optionally, the data processing system can adopt a network topology-based approach, first calculating the basic centrality indicators of nodes, including degree centrality, proximity centrality, and betweenness centrality, then obtaining a comprehensive score through multi-indicator fusion, and finally adjusting the weight distribution based on expert experience; Optionally, the data processing system can adopt a random walk-based approach, constructing a transition probability matrix, obtaining the steady-state distribution probability of nodes through iterative calculation, and correcting the probability values ​​by combining the semantic importance of connections. It is understood that other weight calculation methods can also be used, and are not limited here.

[0098] In practical applications, uneven distribution of node connections may lead to biases in weight calculation. To address this, the data processing system employs a hierarchical weighted calculation strategy. First, the network is layered according to the type and level of knowledge units, and basic weights are calculated within each layer. Then, considering the differences in importance between layers, hierarchical weight coefficients are set. Finally, the final result is obtained by combining the intra-layer weights and hierarchical weights. For example, for a knowledge network for equipment fault diagnosis, the system divides the network into layers based on fault type, focusing on the weight calculation of nodes related to safety hazards to ensure that key knowledge receives higher importance scores.

[0099] S205. Based on node weights, construct the transition probability matrix of the directed knowledge link and calculate the page walkthrough score.

[0100] Among them, the transition probability matrix represents the numerical representation of the jump probability between knowledge units; the page walk score refers to the node importance index calculated by the random walk algorithm; the jump probability is used to represent the possibility of transitioning from one knowledge unit to another; and the damping factor represents the probability of interrupting the current path at any node and randomly jumping to other nodes.

[0101] The data processing system performs this step after obtaining the node weights. Specifically, the data processing system first constructs an adjacency matrix based on the connection relationships between knowledge units; then, it calculates the transition probabilities between nodes based on node weights and connection strengths; next, it introduces a damping factor to consider the impact of random jumps; then, it solves for the steady-state distribution of the transition probability matrix using a power iteration method; finally, it uses the steady-state distribution value as the page walk score of the knowledge unit to evaluate the importance of the node in the global knowledge network.

[0102] In some embodiments, considering the resource consumption issues of knowledge graph technology in practical applications, the data processing system can choose different implementation schemes based on the actual situation: Optionally, in scenarios with limited computing resources, a simple ranking method based on vector similarity can be used to directly calculate the semantic similarity between knowledge units as the basis for importance scoring; alternatively, when offline processing is allowed and resources are sufficient, a complete knowledge graph scheme can be adopted, obtaining more accurate scoring results through complex graph computation. The system will dynamically evaluate available resources and select appropriate processing strategies to achieve a balance between accuracy and efficiency.

[0103] In practical applications, data processing systems can also employ a tiered processing strategy: for frequently accessed core knowledge subgraphs, knowledge graph calculations can be pre-computed and results cached; for less frequently accessed peripheral knowledge, a lightweight real-time computation solution can be used. Simultaneously, the system continuously monitors resource usage and automatically adjusts processing strategies as needed, such as downgrading to basic similarity calculations when resources are scarce, and resuming the complete graph computation process when resources are sufficient. This ensures system availability while providing high-quality knowledge retrieval services whenever possible.

[0104] S206. Identify the critical path in the directed knowledge link based on node weights and page walk scores.

[0105] Among them, the critical path represents the most important knowledge transfer channel in the knowledge network; the path weight refers to the comprehensive score of the weights of all nodes on the path; the path relevance is used to represent the semantic coherence between adjacent knowledge units on the path; and the key node sequence represents the combination of core knowledge units that constitute the critical path.

[0106] The data processing system performs this step after obtaining the page walkthrough score. Specifically, the system first weights and page walkthrough scores to obtain a comprehensive importance score for each node; then, it sets a node selection threshold based on the comprehensive score to initially determine the set of key nodes; next, it analyzes the connections between key nodes to identify node sequences with high semantic relevance; then, it uses a dynamic programming algorithm to calculate the optimal path combination, considering both path completeness and knowledge coverage; finally, it optimizes the identified key paths to ensure path connectivity and the coherence of knowledge transfer.

[0107] In some embodiments, critical path identification can be achieved in several ways: Optionally, the data processing system can adopt a shortest path-based approach, first assigning weights to each edge based on node importance, then using Dijkstra's algorithm to calculate the optimal path between important nodes, and constructing a knowledge backbone network through path merging; Optionally, the data processing system can adopt a flow maximization-based approach, treating the knowledge network as a flow network, identifying the main channels for knowledge transfer using the maximum flow algorithm, and filtering the final critical path based on node importance. It is understood that other path optimization algorithms can also be used, and are not limited here.

[0108] In practical applications, issues may arise where branches or loops in the critical path lead to confusion in knowledge transfer. To address this, the data processing system employs path simplification and loop handling strategies. The system first detects branch points and loop structures within the path, assessing their impact on knowledge transfer. For necessary branches, the main path with high semantic relevance is retained. For loops, the hierarchical relationships between nodes within the loop are analyzed, breaking the loop while ensuring knowledge integrity. For example, when processing the knowledge path for equipment fault diagnosis, the system organizes the diagnostic process into a linear sequence of steps, maintaining the connection between necessary branch points through cross-referencing.

[0109] S207. Based on the critical path, delete redundant links with node weights lower than a preset weight threshold and supplement missing relationships to optimize the directed knowledge link.

[0110] Among them, redundant links represent redundant connections that do not contribute much to knowledge transfer; preset weight thresholds refer to the weight judgment criteria used to filter effective connections; missing associations represent logical relationships between knowledge units that should exist but have not been identified; and optimized knowledge links are used to represent high-quality knowledge networks that have been simplified and supplemented.

[0111] The data processing system performs this step after identifying the critical path. Specifically, the system first filters connections in the knowledge link based on preset weight thresholds, identifying low-value redundant connections; then it analyzes knowledge transfer patterns on the critical path, summarizing typical association patterns between knowledge units; next, based on these association patterns, it searches for potentially existing but unconnected knowledge unit pairs in the knowledge network; then, it performs semantic similarity verification on these candidate associations; finally, it deletes redundant connections while adding newly verified associations, achieving dynamic optimization of the knowledge link.

[0112] In some embodiments, knowledge link optimization can be achieved in several ways: Optionally, the data processing system can adopt a graph structure-based optimization approach, using community detection algorithms to identify the clustering characteristics of knowledge units, determining redundant links based on the connection density inside and outside clusters, and discovering potential associations through transitive closure analysis; Optionally, the data processing system can adopt a knowledge reasoning-based approach, constructing a semantic relationship model between knowledge units, discovering implicit associations through rule reasoning and pattern matching, and determining supplementary and deleted links based on confidence evaluation. It is understood that other network optimization methods can also be used, and are not limited here.

[0113] In practical applications, improper addition of associations may lead to knowledge contamination. To address this, the data processing system employs a multi-layered association verification mechanism. First, the system constructs association verification rules based on the domain knowledge base, including the hierarchical relationships of technical terms and the logical dependencies of technical concepts. Then, candidate associations are verified against these rules to ensure they conform to the domain knowledge system. Finally, important associations are manually confirmed through expert feedback. For example, when adding associations related to equipment maintenance knowledge, the system rigorously verifies the sequential dependencies between maintenance steps to avoid introducing incorrect operational guidance.

[0114] S208. Based on semantic similarity, the directed knowledge link is segmented to obtain multiple knowledge topics.

[0115] Referring to step S104, the data processing system will determine the knowledge topic.

[0116] S209. Receive knowledge retrieval instructions and determine the target knowledge topic corresponding to the knowledge retrieval instructions.

[0117] Referring to step S105, the data processing system will determine the target knowledge topic based on the retrieval.

[0118] S210. Extract the target knowledge units corresponding to the target knowledge topics, obtain the document location information of the target knowledge units, and construct the target context for knowledge retrieval.

[0119] Among them, the target knowledge unit represents a knowledge fragment that is highly relevant to the retrieval requirements; document location information refers to the precise location of the knowledge unit in the original document, including page number, paragraph number, etc.; target context is used to represent the contextual information of the knowledge unit, including adjacent content and citation relationships; knowledge association strength represents the degree of relevance between the knowledge unit and the retrieval requirements.

[0120] The data processing system performs this step after identifying the target knowledge topic. Specifically, the system first filters the most relevant knowledge units within the target knowledge topic based on the search intent; then, it locates the specific position within the original document based on document location information; next, it extracts the contextual information of the knowledge unit, including preceding and following paragraphs, related diagrams, and citations; then, it analyzes the relevance of the contextual content to the search requirements to determine the contextual scope; finally, it organizes the knowledge unit and its effective context into structured retrieval foundation data.

[0121] In some embodiments, the construction of the target context can be achieved in several ways: Optionally, the data processing system can adopt a syntactic analysis-based approach to identify the linguistic dependencies of knowledge units, determine contextual references through coreference resolution, extract key information based on subject-verb relations, and construct a semantically complete contextual representation; Optionally, the data processing system can adopt a topic coherence-based approach to calculate the semantic coherence between paragraphs, identify the boundaries of topic flow, and divide the contextual scope through hierarchical clustering methods to ensure the coherence of knowledge content. It is understood that other contextual analysis methods can also be used, and are not limited here.

[0122] In practical applications, the problem of increased retrieval noise due to an excessively broad context may arise. To address this, the data processing system employs an adaptive context pruning strategy. First, the system constructs a relevance assessment model based on the search intent, scoring each segment of the context content. Then, it sets dynamic thresholds to retain highly relevant core content while preserving necessary logical transitions and key definitions. For example, when searching for the parameter settings of a protection device, the system prioritizes directly relevant content such as parameter definitions and configuration steps, while appropriately simplifying lengthy descriptions of the device's principles.

[0123] S211. Based on the target context, extract the graphic and tabular content of the target location and generate a multimodal knowledge sequence.

[0124] Among them, graphic content represents visual information such as diagrams and structural diagrams in the document; tabular content refers to structured tables containing data and parameters; multimodal knowledge sequence is used to represent complete knowledge expression that integrates text, graphics, and tables; content relevance represents the semantic correspondence between different modal information.

[0125] The data processing system performs this step after constructing the target context. Specifically, the data processing system first locates the graphics and tables referenced in the target context based on the document location information; then it analyzes the components of the graphics, including primitives, labels, and connections; next, it parses the structural features of the tables, including header hierarchy, data areas, and cell attributes; then it identifies the correspondence between text descriptions and chart content; finally, it organizes the content of each modality according to semantic association order to generate a structured knowledge sequence.

[0126] In some embodiments, multimodal knowledge sequence generation can be achieved in various ways: Optionally, the data processing system can adopt an image processing-based approach, using feature extraction algorithms to identify key components in the image, analyzing the image structure through edge detection and region segmentation, and extracting text information from the image using optical character recognition to establish an image-text correspondence; Optionally, the data processing system can adopt a deep learning-based approach, using a multimodal pre-trained model to understand the semantics of the image, capturing the text-image relationship through a cross-modal attention mechanism, and constructing a coherent knowledge representation based on a sequence generation model. It is understood that other multimodal content understanding methods can also be used, and are not limited here.

[0127] In practical applications, incomplete extraction of information from charts may lead to knowledge gaps. To address this, the data processing system employs a strategy of hierarchical analysis and information complementarity. The system first performs multi-level analysis of the chart, extracting information step-by-step from the overall layout to local details. Then, it supplements information that is difficult to directly identify from the chart using relevant descriptions in the text. Simultaneously, it establishes an association index for chart elements. For example, for complex circuit schematics, the system first identifies the main functional blocks, then analyzes the internal connections, and finally supplements detailed information such as component parameters and operating conditions through text descriptions.

[0128] In some embodiments, the data processing system performs structured processing on the chart content in the document and integrates it with the text content. That is, the data processing system locates and extracts the primitive features of the target graphic and the cell layout features of the target table based on the chart location information in the target context; identifies the graphic components and connections in the target graphic and converts the graphic components and connections into a structured graphic description; identifies the header level and data area of ​​the target table, extracts the relationships between cells, and converts them into a structured table description; and combines the structured graphic description and the structured table description with the corresponding text description blocks to generate a multimodal knowledge sequence.

[0129] Among them, primitive features represent the basic constituent elements and attributes of graphics; cell layout features refer to the spatial distribution and merging relationship of cells in a table; graphic components are used to represent the functional structural units in graphics; structured graphic descriptions represent the standardized textual expression of graphic content; structured table descriptions refer to converting table content into a machine-processable data structure; and multimodal knowledge sequences are used to represent a comprehensive knowledge representation that integrates text, graphics, and tables.

[0130] The data processing system executes this step after acquiring the target context information. Specifically, the data processing system first accurately locates the target chart based on the location information, extracts the basic graphic elements and the grid structure of the table; then it analyzes the functional modules in the chart, identifying the connection methods and signal flow between components; next, it parses the hierarchical structure of the table, determining the correspondence between the table header range and the data area; then it extracts the merging information and data type of the cells, establishing logical relationships between the cells; finally, it integrates the graphic semantics, table data, and related text descriptions into a unified knowledge representation.

[0131] In some embodiments, multimodal content parsing and fusion can be achieved in multiple ways: Optionally, the data processing system can adopt a multimodal large language model-based approach to directly perform semantic understanding and feature extraction on image content, achieving deep fusion of text and image information; Optionally, the data processing system can adopt a deep learning-based approach to establish cross-modal correspondences through an attention mechanism and create structured descriptions based on a semantic generation model; Optionally, the data processing system can adopt a deep learning-based approach to understand chart content using a multimodal pre-trained model, establish cross-modal correspondences through an attention mechanism, and create structured descriptions based on a semantic generation model. It is understood that other multimodal processing methods can also be used, and are not limited here.

[0132] In practical applications, complex chart structures may lead to incomplete information extraction. To address this, the data processing system employs a layered, progressive parsing strategy. First, it performs an overall analysis of the chart, identifying key functional areas and layout features. Then, it delves deeper into the detailed information of each area, including component attributes and data relationships. Simultaneously, it supplements implicit information that is difficult to identify directly with textual descriptions. For example, when parsing an electrical schematic, the system first identifies the main circuit modules, then analyzes the component connections, and finally supplements the information with component parameters and operating conditions through textual descriptions, ensuring the completeness of knowledge extraction.

[0133] S212. Based on the document logical hierarchy of target knowledge units, the multimodal knowledge sequence is arranged hierarchically to obtain a structured knowledge body.

[0134] Among them, document logical hierarchy represents the hierarchical organization structure of knowledge content; structured knowledge body refers to the complete knowledge representation reorganized according to hierarchical relationships; hierarchical arrangement is used to represent the subordinate and parallel relationships of knowledge content; knowledge integrity represents the completeness of each component in the knowledge structure.

[0135] The data processing system performs this step after generating the multimodal knowledge sequence. Specifically, the data processing system first analyzes the hierarchical attributes of the target knowledge units to determine the primary and secondary relationships of the content; then, it reconstructs the hierarchical framework of the knowledge sequence based on the document structure tree; next, it maps the multimodal content to the corresponding levels according to logical relationships; then, it processes cross-level reference relationships to ensure the coherence of knowledge transfer; finally, it optimizes the hierarchical structure to generate a knowledge system that is easy to understand and navigate.

[0136] In some embodiments, structured knowledge bodies can be constructed in multiple ways: Optionally, the data processing system can adopt a tree-based approach, constructing a multi-level directory index, representing knowledge levels through node relationships, and using recursive algorithms to process nested structures to achieve hierarchical organization of knowledge content; alternatively, the data processing system can adopt a semantic network-based approach, establishing a concept hierarchy graph, defining the subordinate relationships between knowledge units through ontology relationships, and constructing a complete knowledge structure in conjunction with inference rules. It is understood that other knowledge organization methods can also be used, and are not limited here.

[0137] In practical applications, problems may arise from unreasonable knowledge hierarchy divisions leading to structural chaos. To address this, the data processing system employs a hierarchy optimization and structural standardization strategy. First, the system designs standardized hierarchy templates based on the domain knowledge system, including fixed levels such as basic concepts, principle explanations, and operational procedures. Then, it categorizes and maps knowledge content to ensure appropriate granularity at each level. Simultaneously, it handles hierarchy adjustments in special cases. For example, when organizing equipment maintenance knowledge, the system arranges maintenance requirements, operational procedures, and precautions according to a standard hierarchy, while elevating the hierarchy of important safety warnings for prominent display.

[0138] S213. Generate knowledge retrieval results based on the hierarchical relationship of the structured knowledge body.

[0139] Among them, knowledge retrieval results represent the final knowledge content returned by the system in response to user query requirements; hierarchical relationship refers to the organizational structure and logical connection of each part of the knowledge system; result presentation format is used to represent the display and interaction methods of the retrieved content; knowledge adaptability represents the degree of matching between the retrieval results and user needs.

[0140] The data processing system performs this step after constructing the structured knowledge body. Specifically, the data processing system first determines the presentation hierarchy and level of detail of the results based on the search intent; then it extracts content from the corresponding levels of the structured knowledge body; next, it adjusts the layout of the multimodal content to ensure the rationality of the visual presentation; then, it adds interactive navigation elements to support users in exploring knowledge; and finally, it personalizes and optimizes the search results based on user characteristics and application scenarios.

[0141] In some embodiments, search results can be generated in multiple ways: Optionally, the data processing system can adopt a template-based rendering approach, designing result templates for different types of queries, filling structured knowledge into corresponding positions, achieving standardized content display through style customization, and adding related recommendations and quick navigation functions; Optionally, the data processing system can adopt a dynamic generation approach, analyzing user demand characteristics in real time, adaptively adjusting the organization of content, generating summary descriptions through deep learning models, and enhancing expressiveness through visualization technology. It is understood that other result generation methods can also be used, and are not limited here.

[0142] In practical applications, users may encounter situations where search results are too technical and difficult to understand. To address this, the data processing system employs an intelligent explanation and hierarchical display strategy. The system first identifies technical terms and complex concepts in the results and automatically generates easy-to-understand explanations. Then, it organizes the content according to different levels, such as basic knowledge, specialized knowledge, and in-depth knowledge. Simultaneously, it provides knowledge path navigation to help users understand the relevant content step-by-step. For example, when a user queries the working principle of a protection device, the system will first display basic functional descriptions and then gradually delve into specific technical parameters and implementation mechanisms, ensuring that users with different levels of expertise can obtain appropriate knowledge support.

[0143] In this embodiment, innovative technologies such as document parsing based on a large language model, knowledge link construction, and multimodal content fusion are employed to achieve intelligent processing and knowledge reorganization of documents in the power sector. These core functions include terminology recognition, chart content parsing, and knowledge unit association. Through dynamically optimized knowledge links and adaptive retrieval services, the system effectively solves problems such as knowledge fragmentation, low relevance, and inaccurate retrieval found in traditional document management systems. This results in the systematic organization, accurate retrieval, and intelligent service of knowledge in the power sector, significantly improving knowledge management efficiency and service quality.

[0144] The data processing system in the embodiments of this invention is described below from a hardware processing perspective. Please refer to [link / reference needed]. Figure 3 This is a schematic diagram of the physical device structure of a data processing system in an embodiment of this application.

[0145] It should be noted that,Figure 3 The structure of the data processing system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0146] like Figure 3 As shown, the data processing system includes a CPU 301, which can perform various appropriate actions and processes according to a program stored in ROM 302 or a program loaded from storage section 308 into RAM 303, such as executing the methods described in the above embodiments. RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via bus 304. I / O interface 305 is also connected to bus 304.

[0147] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including hard disks, etc.; and communication section 309 including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0148] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by CPU 301, it performs the various functions defined in the present invention.

[0149] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0150] Specifically, the data processing system in this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the method for constructing a dynamic knowledge base for the power industry based on a large language model provided in the above embodiment.

[0151] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the data processing system described in the above embodiments; or it may exist independently and not assembled into the data processing system. The storage medium carries one or more computer programs, which, when executed by a processor of the data processing system, cause the data processing system to implement the method for constructing a dynamic knowledge base for the power industry based on a large language model provided in the above embodiments.

[0152] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0153] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

Claims

1. A method for constructing a dynamic knowledge base for the power industry based on a large language model, characterized in that, Applied to a data processing system, the method includes: Receive power sector documents imported by users and convert the power sector documents into electronic documents containing unique document identifiers; The electronic document is parsed by calling a large language model, extracting the definitions of professional terms, chart location information and document logical hierarchy in the electronic document, and generating knowledge units containing multiple contexts; The knowledge units are connected to form a directed knowledge link, and the semantic similarity between adjacent knowledge units in the directed knowledge link is calculated; The directed knowledge link is segmented based on the semantic similarity to obtain multiple knowledge topics; Receive a knowledge retrieval instruction and determine the target knowledge topic corresponding to the knowledge retrieval instruction; Extracting target knowledge units corresponding to the target knowledge topic, and generating knowledge retrieval results based on the target context of the target knowledge units, specifically includes: extracting target knowledge units corresponding to the target knowledge topic, obtaining document location information of the target knowledge units, and constructing a target context for knowledge retrieval; extracting graphic and table content of the target location based on the target context to generate a multimodal knowledge sequence; arranging the multimodal knowledge sequence hierarchically based on the document logical hierarchy of the target knowledge units to obtain a structured knowledge body; and generating knowledge retrieval results based on the hierarchical relationship of the structured knowledge body.

2. The method according to claim 1, characterized in that, The step of receiving power-related documents imported by the user and converting the power-related documents into electronic documents containing unique document identifiers specifically includes: Receive power sector documents imported by users, extract document attribute information from the power sector documents, and obtain document type, creation time and version number; Based on the document type, select the corresponding layout analysis model, identify the layout structure of the power field documents including text paragraphs, tables, and graphics, and obtain the document structure tree; Generate a unique document identifier based on the document structure tree and the version number; The layout structure and the unique document identifier are encapsulated into an electronic document in a preset standard format.

3. The method according to claim 1, characterized in that, Before the step of calling a large language model to parse the electronic document, extracting the definitions of technical terms, chart location information, and document logical hierarchy from the electronic document, and generating knowledge units containing multiple contexts, the method further includes: A training sample set is constructed based on the aforementioned documents in the power industry; the training sample set includes definitions of professional terms, equipment operation procedures, and fault handling processes. Extract the technical terms from the power industry documents and construct a power industry thesaurus. The power domain vocabulary is added to the pre-trained language model, and the pre-trained language model is incrementally trained using the training sample set to obtain the large language model.

4. The method according to claim 1, characterized in that, The step of extracting the graphical and tabular content of the target location based on the target context and generating a multimodal knowledge sequence specifically includes: Based on the chart location information in the target context, locate and extract the primitive features of the target graphic and the cell layout features of the target table; Identify the graphic components and connections in the target graphic, and convert the graphic components and connections into a structured graphic description; Identify the header level and data area of ​​the target table, extract the relationships between cells, and convert them into a structured table description; The structured graphical description and the structured tabular description are combined with the corresponding text description blocks to generate a multimodal knowledge sequence.

5. The method according to claim 1, characterized in that, After the steps of connecting the knowledge units into directed knowledge links and calculating the semantic similarity between adjacent knowledge units in the directed knowledge links, the method further includes: Calculate the out-degree and in-degree values ​​of each knowledge unit in the directed knowledge link to obtain the node weight; Based on the node weights, the transition probability matrix of the directed knowledge link is constructed, and the page walkthrough score is calculated. Based on the node weights and page traversal scores, identify the key paths in the directed knowledge link; Based on the critical path, redundant links with node weights lower than a preset weight threshold are deleted, and missing associations are added to optimize the directed knowledge link.

6. The method according to claim 5, characterized in that, The step of calculating the out-degree and in-degree values ​​of each knowledge unit in the directed knowledge link to obtain the node weight specifically includes: Traverse each knowledge unit in the directed knowledge link, count the number of edges pointing to other knowledge units, and obtain the out-degree value; Traverse each knowledge unit in the directed knowledge link, count the number of edges from other knowledge units, and obtain the in-degree value; The node weight is obtained by weighting the out-degree value and the in-degree value.

7. A data processing system, characterized in that, The data processing system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the data processing system to perform the method as described in any one of claims 1-6.

8. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the data processing system, the data processing system performs the method as described in any one of claims 1-6.

9. A computer program product, characterized in that, When the computer program product is run on the data processing system, the data processing system performs the method as described in any one of claims 1-6.

Images