Technical supervision system of power equipment and data processing method

Abstract

The application relates to the technical field of power equipment detection and management, and particularly provides a power equipment technical supervision system and a data processing method, which comprises the following steps: a user terminal is used for receiving a user-inputted supervision question expression; a cloud server is in communication connection with the user terminal; the cloud server comprises a large language model, a query module and a storage module; the query module queries matched relevant clauses in the storage module according to the supervision question expression, and rearranges the relevant clauses based on the subject description of the standard file to which the relevant clauses in the preliminary search result belong; and the large language model rearranges the relevant clauses according to the relevance of the subject description and the supervision question expression. The technical supervision mode of the power equipment in the related art needs manual query of relevant clauses, which is not only low in efficiency, but also requires a high understanding degree of various standards for the supervision construction personnel, thereby leading to the problem of high supervision difficulty.

Description

Technical Field

[0001] This invention relates to the field of power equipment testing and management technology, and in particular to a technical supervision system and data processing method for power equipment. Background Technology

[0002] In power technology supervision scenarios, supervisors need to match relevant standard clauses with on-site issues. This process is of great significance in promoting the standardized operation of the power system and ensuring the safe and stable operation of equipment. Compared with information retrieval in other fields, clause matching in power technology supervision is more professional and complex in terms of handling technical terminology, equipment relationships, and standard document structures, requiring precise matching of on-site issues with rigorous standard clauses.

[0003] In related technologies, existing clause matching methods often rely on human experience or traditional keyword-based retrieval methods, neglecting the support of professional knowledge in the power field. While some studies attempting to use pre-trained language models have improved processing efficiency to some extent, the lack of accurate understanding of domain terminology and equipment relationships, as well as the inability to incorporate related clauses or equipment background information, results in a lack of contextual logical support.

[0004] The current technical supervision methods for power equipment require manual review of relevant clauses, which is not only inefficient but also requires supervisors to have a high level of understanding of various standards, making supervision difficult. No effective solution has yet been proposed. Summary of the Invention

[0005] The present invention provides a technical supervision system and data processing method for power equipment, which at least solves the problem that the technical supervision of power equipment in the related art requires manual querying of relevant clauses, which is not only inefficient, but also requires the supervising construction personnel to have a high degree of understanding of various standards, resulting in great difficulty in supervision.

[0006] According to one aspect of the present invention, a technical supervision system for power equipment is provided, comprising: a user terminal for receiving a supervision question statement input by a user; a cloud server communicatively connected to the user terminal, the cloud server including a large language model, a query module, and a storage module, the storage module for storing standard documents, the standard documents including relevant clauses; the query module being connected to the storage module and the large language model, for querying matching relevant clauses in the storage module according to the supervision question statement to obtain preliminary search results, and using the large language model to rearrange the relevant clauses of the preliminary search results based on the topic description of the standard document to which the relevant clauses in the preliminary search results belong; the large language model being used to rearrange the relevant clauses according to the relevance between the topic description and the supervision question statement.

[0007] As an optional solution, the cloud server further includes: a rewriting module; the rewriting module includes a first call port connected to the large language model, the rewriting module includes functions for inputting the supervision problem statement into the large language model, receiving semantic expansion of the supervision problem statement based on the large language model to obtain pseudo-document content that is more in line with professional semantics, and concatenating the supervision problem statement and the pseudo-document content to generate semantically enhanced text.

[0008] As an optional solution, the cloud server further includes: a knowledge graph creation module, used to extract entities from the standard file based on examples of various entities using a large language model to obtain an entity set; using entities in the entity set as seeds, using automated collection tools to collect data from professional information websites to expand the entity set; and storing the entities and entity relationships in the entity set in a structured manner in a pre-built knowledge graph to obtain an enhanced knowledge graph.

[0009] As an optional solution, the storage module further includes: a cleaning submodule, used to extract text and clean data from the acquired standard file with a preset format, removing irrelevant format characters and retaining the text content; a regular expression matching submodule 122, connected to the cleaning submodule, used to identify the clause number for each line of text by regular expression pattern matching, obtaining structured line data with number; a document tree construction submodule, connected to the regular expression matching submodule 122, used to construct a multi-branch document tree based on the structured line data, using a stack structure according to the organizational hierarchy of the standard file, wherein the multi-branch document tree includes multiple nodes, each node storing at least the clause number, clause content, and node hierarchy path; and a persistent storage submodule, connected to the document tree construction submodule, used to persistently store the multi-branch document tree, wherein the multi-branch document tree is used for retrieval and backtracking.

[0010] As an optional solution, the cloud server further includes a backtracking and reordering module, connected to the query module, used to reorder the relevant terms in the preliminary search results based on the topic description and the supervision question statement using a large language model, to obtain the final search terms results.

[0011] According to another aspect of the present invention, a data processing method for a technical supervision system for power equipment is also provided, comprising: receiving a supervision question statement sent by a user terminal; retrieving corresponding preliminary search results from a pre-stored standard document library based on the semantically enhanced text of the supervision question statement using a query algorithm, wherein the preliminary search results include a preset number of relevant clauses selected from a relevance sequence, the relevance sequence including multiple relevant clauses sorted by relevance degree; determining the topic description of each relevant clause in its respective standard document based on the preliminary search results; reordering the relevant clauses in the preliminary search results using a large language model based on the topic description and the supervision question statement to obtain a final search clause result; and generating relevant clauses matching the supervision question statement using a large language model based on the search clause result and the supervision question statement.

[0012] As an optional approach, before retrieving the corresponding preliminary search results from a pre-stored standard document library using a query algorithm based on the semantically enhanced text of the supervision problem statement, the method further includes: generating a first prompt word based on the supervision problem statement, the format requirements of the standard document, and the professional field it belongs to, and inputting the first prompt word into the large language model; semantically expanding the supervision problem statement using the large language model based on a preset enhanced knowledge graph to obtain pseudo-document content that better fits the professional semantics; and concatenating the supervision problem statement and the pseudo-document content to generate semantically enhanced text.

[0013] As an optional approach, a large language model is used to semantically expand the supervised question statement based on a pre-defined augmented knowledge graph to obtain pseudo-document content that better fits the professional semantics. This includes: extracting entities from the supervised question statement using the large language model based on examples of various entities to obtain the statement entities; finding adjacent entities adjacent to the statement entities based on the augmented knowledge graph; and semantically expanding the pseudo-document content based on the adjacent entities using the large language model.

[0014] As an optional approach, before semantically expanding the supervised problem statement using a large language model based on a pre-built enhanced knowledge graph to obtain pseudo-document content that better fits professional semantics, the method further includes: acquiring a standard file in plain text format; extracting entities from the standard file using a large language model based on examples of various entities to obtain an entity set; using entities in the entity set as seeds, collecting data from professional information websites using automated collection tools to expand the entity set; and storing the entities and entity relationships in the entity set in a structured manner within a pre-constructed knowledge graph to obtain the enhanced knowledge graph.

[0015] As an optional approach, before retrieving preliminary search results from a pre-stored standard file library based on the semantically enhanced text of the supervision problem statement using a query algorithm, the method further includes: obtaining standard files in a preset format; performing text extraction and data cleaning on the standard files, removing irrelevant format characters and retaining the text content; identifying the clause number for each line of text using regular expression pattern matching to obtain structured line data with numbers; constructing a multi-branch document tree based on the structured line data using a stack structure according to the organizational hierarchy of the standard files, wherein the multi-branch document tree includes multiple nodes, each node storing at least the clause number, clause content, and node hierarchy path; and persistently storing the multi-branch document tree, wherein the multi-branch document tree is used for retrieval and backtracking.

[0016] As an optional approach, based on the preliminary search results, the topic description of each relevant clause in the corresponding standard file is determined, including: obtaining the corresponding chapter number in the corresponding standard file based on the preliminary search results; obtaining the node number of the target level by combining the corresponding chapter number with the corresponding multi-branch document tree; traversing all parent nodes above the node number of the target level in the multi-branch document tree of the corresponding standard file; and concatenating the semantic content corresponding to the parent nodes in hierarchical order to construct the topic description of the node number in the standard file.

[0017] As an optional approach, based on the search terms results and the supervision question statement, a large language model generates relevant terms matching the supervision question statement, including: using an enhanced knowledge graph as a supplementary knowledge base; generating a second prompt word based on the supplementary knowledge base, the search terms results, and the supervision question statement; inputting the second prompt word into the large language model, and outputting the corresponding relevant terms, the question answer, and supplementary information of the relevant terms.

[0018] According to another aspect of the present invention, an electronic device is also provided, comprising: a processor, and a memory storing a program, the program including instructions that, when executed by the processor, cause the processor to perform the method described in any of the preceding claims.

[0019] The technical supervision system for power equipment provided in this invention employs a user terminal and a remote server. The cloud server includes a large language model, a query module, and a storage module. The storage module stores standard documents containing relevant clauses. After receiving the user's input of a supervision question, the user terminal communicates with the cloud server. The query module searches the storage module for matching clauses based on the supervision question to obtain preliminary search results. It then rearranges the relevant clauses based on the topic description of the standard documents to which they belong in the preliminary search results. The large language model further rearranges the relevant clauses based on the relevance between the topic description and the supervision question. This solves the problem in related technologies where technical supervision of power equipment requires manual querying of relevant clauses, which is inefficient and requires supervisors to have a high level of understanding of various standards, leading to significant supervision difficulties. The system achieves the technical effect of eliminating the need for manual clause querying, improving efficiency, reducing the requirements for personnel's understanding of standards, and decreasing the difficulty of supervision. Attached Figure Description

[0020] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other embodiments based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the technical supervision system structure according to an embodiment of the present invention.

[0022] Figure 2 This is a flowchart of the data processing method of the technical supervision system according to an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of the structure of the electronic device created by this invention. Detailed Implementation

[0024] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0025] In power technology supervision scenarios, supervisors need to match relevant standard clauses with on-site issues. This process is of great significance in promoting the standardized operation of the power system and ensuring the safe and stable operation of equipment. Compared with information retrieval in other fields, clause matching in power technology supervision is more professional and complex in terms of handling technical terminology, equipment relationships, and standard document structures, requiring precise matching of on-site issues with rigorous standard clauses.

[0026] Existing clause matching methods often rely on human experience or traditional keyword-based retrieval methods, neglecting the support of professional knowledge in the power industry. While some studies attempting to use pre-trained language models have improved processing efficiency to some extent, the lack of accurate understanding of domain terminology and equipment relationships, as well as the inability to incorporate related clauses or equipment background information, results in a lack of contextual logical support.

[0027] To improve efficiency, reduce the requirements for personnel's understanding of standards, and decrease the difficulty of supervision, embodiments of this invention provide a technical supervision system for power equipment, such as... Figure 1 As shown, the system includes: a user terminal 101, which is used to receive a supervised problem statement input by a user.

[0028] The cloud server 102 is communicatively connected to the user terminal 101. The cloud server 102 includes a large language model 111, a query module 112, and a storage module 113. The storage module 113 is used to store standard files, which include relevant clauses.

[0029] The query module 112 is connected to the storage module 113 and the large language model 111. It is used to query the relevant clauses that match in the storage module 113 according to the supervision question statement, obtain the preliminary search results, and use the large language model 111 to rearrange the relevant clauses in the preliminary search results based on the topic description of the standard document to which the relevant clauses in the preliminary search results belong.

[0030] The large language model 111 is used to rearrange relevant clauses based on the relevance of topic descriptions and supervision question statements.

[0031] Power maintenance personnel need to regularly inspect and test in-operation equipment, and must quickly look up inspection items, testing cycles, and qualification standards to avoid omissions or improper operation. Before equipment is put into operation and after major overhauls, professional testing is required. Testing personnel must be familiar with testing methods, instrument requirements, and result judgment rules to ensure the testing process is compliant and the results are valid. When equipment malfunctions, maintenance and repair personnel must quickly look up fault diagnosis criteria, handling procedures, and rectification standards to prevent the fault from escalating. Before new equipment is put into operation, acceptance personnel must verify equipment parameters, installation quality, and testing items against standards to ensure they meet commissioning conditions. Power companies need to provide technical supervision training to maintenance personnel to ensure they understand standard requirements; simultaneously, compliance auditors must verify whether supervision work complies with standards to mitigate safety risks. These scenarios all require accurate acquisition of standard data and efficient understanding of professional semantics. The power equipment technical supervision system created in this invention can be effectively applied to these scenarios.

[0032] User terminal 101 serves as the human-computer interaction entry point, responsible for receiving the user's input of the supervision question and transmitting the question to cloud server 102. Cloud server 102, as the core processing unit, contains three key sub-modules.

[0033] The aforementioned user terminal can be a smartphone, which can realize input and communication functions through a developed APP (Application), or it can be a specially developed private communication device to ensure data security, etc.

[0034] Storage module 113 is used to store standard documents related to the technical supervision of power equipment, such as industry standards, national standards, and enterprise specifications. These documents contain specific relevant clauses, such as testing methods, limit requirements, and operating procedures, and serve as the system's basic knowledge base.

[0035] The query module 112 connects the storage module 113 and the large language model 111, undertaking the functions of preliminary retrieval and initial reordering. First, based on the user's supervision question, the storage module 113 retrieves potentially relevant clauses through keyword matching, semantic screening, and other methods to form preliminary search results. Then, based on the subject descriptions of the standard documents to which each clause belongs in the preliminary results, such as equipment categories: "transformer", "circuit breaker", "surge arrester", etc.; power system components: such as "busbar", "cable", "grounding device", etc.; work types: such as "maintenance", "test", "acceptance", etc.; safety measures: such as "work order", "risk assessment", "isolation measures", etc.; and accident categories: such as "short circuit fault", "overvoltage", "misoperation", etc., the preliminary results are reordered for the first time based on the relevance priority of the subject descriptions and the supervision question. For example, clauses related to transformers are prioritized in the search results related to transformer issues.

[0036] The large language model 111 performs a secondary reordering function. Based on the deep semantic relevance between the topic description and the user's supervised question expression, rather than simple keyword matching, it further optimizes the order of the clauses after the first reordering. For example, if the user's question is "Insulation resistance testing requirements before transformer commissioning," the large language model 111 will determine that the topic "transformer insulation testing" is more relevant to the question than the topic "transformer routine maintenance," thus placing the clauses corresponding to the former at the top. After obtaining the final ranking result through optimization, the final result is returned to the user terminal 101 for the user's reference.

[0037] A large language model is a deep learning-based artificial intelligence model that learns from massive amounts of text data to master the grammar, semantics, and knowledge of human language. It can generate coherent text, answer questions, and translate. The model contains billions or even trillions of parameters, forming the foundation of its powerful language capabilities. It can understand the logical relationships in long texts, generate context-appropriate subsequent content based on preceding information, and perform various tasks such as text generation, summarization, translation, coding, and logical reasoning without requiring separate training for a single task. Large language models can be developed through artificial intelligence or be specially developed private AI models to ensure data security.

[0038] User oversight issues may include: issues related to routine operation oversight, such as "How often should the insulation resistance of a 110kV oil-immersed transformer be tested? Are there any corresponding standard clauses?"; issues related to testing / inspection oversight, such as "What are the environmental temperature and humidity requirements for dielectric loss testing before transformer commissioning? Are there any relevant provisions in DL / T596?"; issues related to fault diagnosis and handling, such as "The cable insulation resistance test value is 50% lower than the last time, what could be the problem? Are there any standard clauses stating whether shutdown is necessary in this situation?"; and issues related to operation and acceptance oversight, such as "What should the insulation oil breakdown voltage value of the box-type transformer in a photovoltaic power station meet before commissioning to be considered compliant?".

[0039] The technical supervision system and data processing method for power equipment provided by the embodiments of this invention avoids the problems of missed detections and false detections in traditional keyword retrieval through hierarchical processing of preliminary retrieval and two rearrangements. The first rearrangement based on topic description narrows the scope, while the second rearrangement based on semantic analysis of the large language model 111 further focuses on core needs, prioritizing the presentation of the most relevant clauses. Technical supervision of power equipment involves a large number of professional standards, with numerous clauses and rigorous expressions. User questions may be expressed in diverse ways, and the semantic understanding capability of the large language model 111 can adapt to such differences in expression, improving matching efficiency across expression scenarios. The large language model 111 can continuously optimize the relevance judgment logic between topic descriptions and question expressions through training. As data accumulates, the ranking results will better match the needs of actual supervision scenarios. Thus, it achieves the technical effects of eliminating the need for manual clause queries, improving efficiency, reducing the requirements for personnel's understanding of standards, and reducing the difficulty of supervision.

[0040] As an optional solution, the cloud server 102 also includes: a rewriting module 114; the rewriting module 114 includes a first call port, which is connected to the large language model 111. The rewriting module 114 includes functions for inputting the supervision problem statement into the large language model 111, receiving semantic expansion of the supervision problem statement based on the large language model 111 to obtain pseudo-document content that is more in line with professional semantics, and concatenating the supervision problem statement and the pseudo-document content to generate semantically enhanced text.

[0041] The rewriting module 114 addresses the potential issues of colloquialisms and ambiguities in user question expressions by introducing the semantic expansion capabilities of the large language model 111, thereby making subsequent searches more accurate.

[0042] The first call port serves as the connection interface with the large language model 111, responsible for passing the user-input supervision problem statement to the large language model 111 and receiving the return result.

[0043] Based on its expertise in power equipment technical supervision, the Big Language Model 111 interprets and expands the original question. For example, if a user inputs a colloquial question, "When should the insulation of a transformer be tested?", the Big Language Model 111 will expand it into "pseudo-document content" that includes professional terminology and relevant scenarios, such as "the timing requirements for insulation resistance testing of transformers before commissioning, after major overhauls, and after dampness, including the testing cycle regulations in preventive testing."

[0044] The rewriting module 114 combines the original supervision problem statement with the expanded pseudo-document content to form an enhanced text that retains the user's original intent while also containing professional semantics.

[0045] The system workflow incorporates a rewriting module 114 between user input and initial retrieval. This addresses the issue of users using non-technical terminology to ask questions, leading to discrepancies in expression. Semantic enhancement transforms these questions into more professionally accurate language, closer to standard document expressions, thus reducing retrieval omissions. Pseudo-document content includes potential needs related to the question, enabling the initial retrieval to cover a more comprehensive range of relevant clauses.

[0046] The introduction of the rewriting module 114 provided in this embodiment allows users to ask questions without needing precise knowledge of technical terms. The system supplements professional information through semantic expansion, making it more suitable for the actual usage scenarios of front-line maintenance personnel. The original question may contain incomplete information, such as only asking how to do it without mentioning any equipment. The pseudo-document content can supplement reasonable assumptions based on the context, reducing missed detections due to missing information. The semantically enhanced text provides richer semantic basis for the initial retrieval of the query module 112 and the secondary ranking of the large language model 111, making the subsequent ranking more in line with the user's actual needs.

[0047] As an optional solution, the cloud server 102 also includes: a knowledge graph creation module 115, which is used to extract entities from standard files based on examples of various entities using a large language model 111 to obtain an entity set; using entities in the entity set as seeds, data is collected from professional information websites through automated collection tools to expand the entity set; and the entities and entity relationships in the entity set are stored in a structured manner in a pre-built knowledge graph to obtain an enhanced knowledge graph.

[0048] The knowledge graph creation module 115, based on the large language model 111, combines entity examples from the power equipment field, such as predefined equipment entity examples: transformers, GIS (Gas Insulated Switchgear) equipment, cables, etc.; parameter entity examples: insulation resistance, dielectric loss value, temperature rise limit; and operation entity examples: dielectric loss testing, airtightness testing, etc., to automatically identify and extract entities from the standard documents in the storage module 113, forming an entity set. The entity set contains all core professional concepts from the standard documents, such as equipment name, testing items, technical parameters, standard number, etc.

[0049] Using entities in the entity set as seeds, relevant data is collected from professional information websites in the power industry (such as the National Energy Administration's standard platform, power equipment testing database, and industry journal database) through automated data collection tools, such as targeted web crawlers and API (Application Programming Interface) interfaces. This further enriches the types and number of entities, forming a more comprehensive expanded entity set that covers more granular knowledge such as equipment sub-types, parameter value ranges, and operational details.

[0050] The expanded entity set and the relationships between entities, such as inclusion relationships ("Transformers include iron cores and windings"), requirement relationships ("GIS equipment requires an annual air leakage rate of ≤0.5%), and reference relationships ("DL / T596-2021 references the testing methods in GB50150-2016"), are stored in a pre-built knowledge graph framework to form an enhanced knowledge graph.

[0051] Structured storage is a method of organizing and storing data according to predefined formats or rules. Its purpose is to give data a clear logical structure and explicit relationships, thereby facilitating efficient computer identification, querying, management, and analysis. In knowledge graphs, structured storage primarily uses triples of entities, relations, and entities as basic units. Each entity has a defined type, and relations have defined semantics.

[0052] Traditional search relies on direct matching of clause text with questions, while knowledge graphs can uncover implicit relationships through entity relations, providing domain knowledge anchors for the large language model 111 and reducing semantic understanding bias. For example, for "GIS equipment leak detection," the relationship between "GIS equipment, core components, SF6 gas chamber" in the knowledge graph can help the large language model 111 accurately locate the core object of "leak detection" as the SF6 gas chamber, rather than the equipment as a whole, making subsequent clause matching more accurate.

[0053] The power industry standards and technologies are constantly being updated. Automated data acquisition tools can regularly scrape newly released standards and technical specifications from professional websites and update the knowledge graph through an entity expansion mechanism to ensure that the system knowledge is synchronized with the latest industry developments, thus solving the problem that static files in traditional storage module 113 are difficult to update in a timely manner.

[0054] The knowledge graph provided in this embodiment offers professional knowledge support for the semantic expansion of the large language model 111, making the pseudo-document content more relevant to the domain. During the initial retrieval, the query module 112 can filter relevant clauses based on entity relationships in the knowledge graph, such as prioritizing clauses directly related to the question entity, thus improving the accuracy of the initial retrieval. As a domain knowledge base, the knowledge graph reduces erroneous associations in the large language model 111, making the relevance judgment during secondary ranking more consistent with professional logic.

[0055] As an optional solution, storage module 113 further includes: a cleaning submodule 121, used to extract text and clean data from the acquired standard file with a preset format, removing irrelevant format characters and retaining the text content; a regular expression matching submodule 122, connected to the cleaning submodule 121, used to identify the clause number for each line of text by regular expression pattern matching, obtaining structured line data with number; a document tree construction submodule 123, connected to the regular expression matching submodule 122, used to construct a multi-branch document tree based on the structured line data and using a stack structure according to the hierarchical relationship of the standard file, wherein the multi-branch document tree includes multiple nodes, each node storing at least the clause number, clause content, and node hierarchical path; and a persistent storage submodule 124, connected to the document tree construction submodule 123, used to persistently store the multi-branch document tree, wherein the multi-branch document tree is used for retrieval and backtracking.

[0056] The four sub-modules of storage module 113—cleaning sub-module 121, regular expression matching sub-module 122, document tree construction sub-module 123, and persistent storage sub-module—are connected in sequence to form a complete standard file structured processing pipeline. The core objective is to transform raw standard files, such as PDF (Portable Document Format) and Word (text-based document) formats, into structured data that computers can efficiently understand and retrieve, laying the foundation for subsequent querying, reordering, and other processes.

[0057] The cleaning submodule 121 acquires standard documents in preset formats, including PDF standards for the power industry and Word-based technical specifications. The cleaning submodule 121 extracts the original text content from non-text formats, removing irrelevant formatting characters, redundant blank lines, and garbled characters, retaining only meaningful text content such as clause descriptions and technical parameters. This results in clean, plain text content arranged in the line order of the original document, clearing away formatting obstacles for subsequent structured processing.

[0058] The regular expression matching submodule 122 is connected to the cleaning submodule 121. It uses regular expression pattern matching to identify the clause number corresponding to each line of text after cleaning, and obtains structured line data with numbering. This binds each line of text to the corresponding clause code, thus solving the problem of corresponding text content with clause identity.

[0059] Regular expressions are rule expressions used to match, find, and replace specific combinations of characters in text. They construct flexible search patterns through a series of predefined characters and syntax, and are widely used in text processing, data validation, information extraction, and other scenarios.

[0060] The document tree construction submodule 123 constructs a multi-branch document tree based on the hierarchical features of the clause numbers from the structured row data after regular expression matching, using a stack structure. The stack in the document tree construction submodule 123 records the hierarchical path of the currently processed clauses. Each clause corresponds to a node in the tree, and the node stores key information, including the clause number, clause content, and node hierarchical path. The hierarchical relationship between nodes is determined by the number hierarchy, ultimately forming a multi-branch document tree that clearly presents the chapter, section, clause, and clause hierarchical structure of the standard document, allowing fragmented clause content to form a logical whole.

[0061] The persistent storage submodule persistently stores the structure and node information of the multi-branch document tree in the database. Storage methods can include using a relational database to store node attributes, with an additional field recording the parent node name to reflect hierarchy; or directly storing the tree structure in a document database. The subsequent query module 112 can perform precise queries based on node number, hierarchical path, or content. It can also quickly trace the parent or child clauses of a clause through the hierarchical path to understand the contextual relationships between clauses.

[0062] The four sub-modules of the storage module 113 provided in this embodiment transform standard file structures that cannot be directly recognized by computers into hierarchical, indexed, and associative structured data through the processes of cleaning, identification, structuring, and storage. This allows the subsequent query module 112 to accurately locate a specific level of clauses. Furthermore, when processing clauses, the large language model 111 can understand the context of the clauses by combining the hierarchical path, thus significantly improving the efficiency and accuracy of professional retrieval.

[0063] As an optional solution, the cloud server 102 also includes a backtracking and rearrangement module 116, which is connected to the query module 112 and is used to reorder the relevant terms in the preliminary search results according to the topic description and the supervision question statement through the large language model 111 to obtain the final search terms results.

[0064] The backtracking and reordering module 116 is connected to the query module 112. It accepts the preliminary search results output by the query module 112. The large language model 111 analyzes the deep semantic relationship between the topic description of the standard document to which the clause belongs and the user's original supervision question statement. This includes not only keyword matching but also semantic intent and the degree of fit of professional scenarios. The backtracking and reordering module 116 does not replace the existing sorting process but forms a multi-level sorting funnel, working synergistically with the query module 112 and the large language model 111 to improve the query effect.

[0065] The initial reordering of the query module 112 is based on the preliminary association between the topic description and the question, which solves the problem of broad-based filtering; the backtracking reordering module 116, on the other hand, solves the problem of precise sorting through the deep understanding of the large language model 111.

[0066] For example, the preliminary results may include clauses related to the topic of "transformer insulation testing", such as "operation steps for transformer insulation testing", "insulation resistance qualification value requirements" and "testing instrument calibration specifications". The backtracking and rearrangement module 116 will prioritize the "qualification value requirements" clause based on the user's intent regarding "qualification value".

[0067] The Big Language Model 111 not only provides semantic analysis capabilities, but also enhances the understanding of professional scenarios by combining entity relationships in the knowledge graph. For example, when a user asks "What should I do if the leakage rate of GIS equipment exceeds the standard?", the model can identify that the core intent is "the handling measures after exceeding the standard", rather than "the leakage rate detection method", and thus prioritize the relevant "fault handling clauses".

[0068] The backtracking and reordering module 116 provided in this embodiment uses deep semantic analysis to accurately locate the core needs of user questions, avoiding irrelevant answers. Through the multi-dimensional semantic understanding of the large language model 111, it comprehensively judges the fit between the terms and these dimensions, improving the accuracy of responses to complex questions. As user feedback accumulates from system use, such as user click preferences for certain types of questions, the judgment criteria for backtracking and reordering can be further optimized by fine-tuning the large language model 111, so that the sorting results gradually adapt to the usage habits of actual business scenarios.

[0069] According to another aspect of the present invention, a data processing method for a technical monitoring system for power equipment is also provided, such as... Figure 2 As shown, the method includes the following steps: Step S201, receiving a supervision problem statement sent by the user terminal 101.

[0070] Step S202: Based on the semantically enhanced text of the supervised problem statement, the corresponding preliminary search results are retrieved from the pre-stored standard document library through a query algorithm. The preliminary search results include a preset number of relevant clauses selected from the relevance sequence, and the relevance sequence includes multiple relevant clauses sorted by relevance.

[0071] Step S203: Based on the preliminary search results, determine the subject description of each relevant clause in its respective standard document.

[0072] In step S204, the relevant clauses in the preliminary search results are reordered according to the topic description and the supervision question statement through the large language model 111 to obtain the final search clause results.

[0073] Step S205: Based on the search results and the supervision question statement, the large language model 111 generates relevant clauses that match the supervision question statement.

[0074] User terminal 101 is used to accept the original supervision questions submitted by the user as the starting point for system processing and to obtain the user's core needs.

[0075] The system semantically enhances user input for supervision by expanding it into text containing specialized terminology and relevant scenarios using a large language model. Query algorithms, such as semantic vector matching and keyword weighting, are then used to retrieve relevant clauses from a pre-stored standard document library, generating a relevance sequence where all relevant clauses are sorted by their matching degree. A predetermined number of the most similar clauses are selected from this sequence as preliminary search results. Semantic enhancement addresses the issues of colloquialism and ambiguity in user expressions, while preliminary filtering narrows the search scope and improves efficiency.

[0076] Determine the subject description of the relevant clauses. For each clause in the preliminary search results, locate the subject of the standard document to which it belongs. For example, if a clause comes from "DL / T596-2021 Preventive Testing Procedures for Power Equipment", its subject description is the items, methods, and limit requirements for preventive testing of power equipment. This provides a basis for determining the subject dimension for subsequent rearrangement and avoids bias caused by relying solely on surface text matching.

[0077] The topic description and the user's original supervision question are input into the large language model 111. The model performs deep semantic analysis to perform a secondary sorting of the terms in the initial search results, resulting in the final search terms. This solves the problem of topic relevance but mismatch between the specific intent and the target topic.

[0078] The large language model 111 combines the final search results and the supervision question statements to generate natural language answers. It translates technical terms into easily understandable content that fits the user's questions, while retaining the standard basis, thus balancing professionalism and readability.

[0079] The data processing method of the power equipment technical supervision system provided in this embodiment forms a complete closed loop of user intent understanding, professional knowledge matching, and natural language output through a five-step process: semantic enhancement, preliminary retrieval, topic localization, intelligent reordering, and result generation. The two-level sorting of algorithm initial screening and semantic reordering by the large language model ensures that the results are matched precisely from relevant to accurate. Relying on a structured standard document library and topic descriptions, it avoids non-professional interpretations, and the final generated natural language answer lowers the threshold for users to understand professional terms.

[0080] As an optional approach, before retrieving the corresponding preliminary search results from a pre-stored standard document library based on the semantically enhanced text of the supervision question statement using a query algorithm, the method further includes: generating first prompt words based on the supervision question statement, the format requirements of the standard documents, and the professional field they belong to, and inputting the first prompt words into the large language model 111; using the large language model 111 based on a preset enhanced knowledge graph to semantically expand the supervision question statement, obtaining pseudo-document content that better fits the professional semantics; and concatenating the supervision question statement and the pseudo-document content to generate semantically enhanced text.

[0081] Power equipment supervision has a strict system of professional terminology and standard expression habits. Based on three elements, including the expression of the supervision problem, the standard document format requirements, and the relevant professional field, the first prompt word is generated. The first prompt word and the knowledge graph jointly constrain the large language model 111, providing clear expansion boundaries and professional guidance, and avoiding colloquial or cross-domain erroneous expressions.

[0082] The design prompt could be as follows: "You are an expert in the power industry, specializing in analyzing texts related to power safety, equipment maintenance, and technical supervision. Your task is to extract all power-related entities from the input text and output them in JSON format. Extracted entities include: any technical terms that may be related to power; equipment: such as 'transformer,' 'circuit breaker,' 'surge arrester,' etc.; power system components: such as 'busbar,' 'cable,' 'grounding device,' etc.; work types: such as 'maintenance,' 'testing,' 'acceptance,' etc.; safety measures: such as 'work order,' 'risk assessment,' 'isolation measures,' etc.; accident types: such as 'short circuit fault,' 'overvoltage,' 'misoperation,' etc.' Please note: Do not return entire sentences, only the extracted entity words; if you are unsure whether it is a power entity, please return relevant terms if possible."

[0083] Pseudo-documents are generated by semantic augmentation using enhanced knowledge graphs, ensuring that terminology and parameter ranges align with industry realities and reducing errors in the large language model. The pseudo-documents contain professional terminology, standard numbers, and other information that closely match the structured clauses in the pre-stored standard document library, enabling subsequent query algorithms to more accurately match relevant clauses and reduce missed detections.

[0084] The user's original supervisory question statement is concatenated with the expanded pseudo-document content to form the final semantically enhanced text. This preserves the user's original intent for easy tracing during subsequent matching, while also improving the matching accuracy with standard document clauses through professionally expanded content. The semantically enhanced text serves as input to the query algorithm, and its professionalism directly impacts the quality of the initial search results. The expanded content, constrained by the knowledge graph, provides query module 112 with richer search anchors, such as specific standard numbers and detection scenarios, making the initial search results more focused on core needs.

[0085] As an optional approach, the large language model 111 is used to semantically expand the supervision question statement based on a pre-defined augmented knowledge graph to obtain pseudo-document content that better fits the professional semantics. This includes: extracting entities from the supervision question statement using the large language model 111 based on examples of various entities to obtain the statement entities; finding adjacent entities adjacent to the statement entities based on the augmented knowledge graph; and semantically expanding the pseudo-document content based on the adjacent entities using the large language model 111.

[0086] The Large Language Model 111 uses entity examples from the power equipment domain, such as predefined equipment, parameter, and operation entity samples, as references to identify and extract expressive entities from users' supervised question statements. Extracting expressive entities transforms fuzzy natural language questions into professional concept anchors that can be recognized by the knowledge graph, laying the foundation for subsequent association expansion.

[0087] Based on the structured relationships between entities in the enhanced knowledge graph, such as inclusion, association, restriction, and reference, we find adjacent entities that are directly connected to the entity being described. Through the association network of the knowledge graph, we mine highly relevant professional concepts that are not explicitly mentioned in the user's question, providing content material for semantic expansion and avoiding deviation of the expanded content from professional logic.

[0088] The Large Language Model 111 takes representational entities as its core and combines adjacent entities and their relationships to semantically expand the original question, generating pseudo-document content containing professional details. Generating pseudo-document content based on adjacent entities transforms isolated questions into professional text containing entity relationships, standard references, and scenario details, preserving the user's core needs while supplementing them with professional semantics that match the terms of standard documents.

[0089] Traditional semantic expansion may generate irrelevant content due to errors in the large language model 111. However, by limiting the expansion scope through adjacent entities, it is ensured that the content always revolves around the professional logic of power equipment supervision. The information contained in the adjacent entities, such as standard numbers, testing items, and parameter limits, highly matches the core elements of the clauses in the pre-stored standard document library, providing accurate matching anchors for subsequent query algorithms. The entity relationships in the knowledge graph reflect the professional cognitive logic of the power industry. The pseudo-document content generated based on this is more in line with the knowledge structure of front-line personnel, facilitating the understanding and application of subsequent clauses.

[0090] As an optional approach, before semantically expanding the supervised question representation using a large language model 111 based on a pre-built enhanced knowledge graph to obtain pseudo-document content that better fits the professional semantics, the method also includes: obtaining a standard file in plain text format; extracting entities from the standard file using the large language model 111 based on examples of various entities to obtain an entity set; using the entities in the entity set as seeds, collecting data from professional information websites using automated collection tools to expand the entity set; and storing the entities and entity relationships in the entity set in a structured manner in a pre-built knowledge graph to obtain an enhanced knowledge graph.

[0091] Collect relevant standard documents for technical supervision of power equipment, such as national standard GB and industry standard DL / T, and convert them into plain text format. This means removing non-text elements such as layout marks and images from PDF and Word formats to provide a clean text data source for subsequent entity extraction and avoid format interference.

[0092] The Big Language Model 111 uses entity examples from the power industry as a reference to automatically identify entities and extract core concepts from plain text standard documents, forming an initial entity set as the basic node of the knowledge graph.

[0093] Using an initial entity set as a seed, relevant data is collected from professional information websites in the power industry, such as the State Grid Standard Platform, the Power Equipment Testing Database, and the Industry Journal Database, through automated data collection tools such as targeted web crawlers and API integration. This enriches the types and quantities of entities, breaks through the limitations of a single standard document, expands the scope of knowledge coverage, and makes the entity set more comprehensive.

[0094] The expanded entity set and the relationships between entities are stored in a pre-built knowledge graph framework in the form of entity-relationship-entity triples, forming an enhanced knowledge graph. It supports structured representation of entity attributes and multi-level relationships, facilitating rapid subsequent querying of entity associations.

[0095] The entities and relationships in the enhanced knowledge graph provided in this embodiment offer clear professional boundaries for the subsequent semantic expansion of the large language model 111, ensuring that the expanded content conforms to industry logic. Compared to general knowledge graphs, the enhanced knowledge graph constructed by this method focuses on the power equipment supervision scenario, containing a large amount of subdivided domain knowledge, enabling the system to understand more complex professional issues. Through automated data collection tools, newly released standards or technical specifications can be periodically crawled from professional websites, continuously expanding the entity set and relationship network, ensuring that the knowledge graph is synchronized with the latest industry developments.

[0096] As an optional approach, before retrieving preliminary search results from a pre-stored standard document library based on the semantically enhanced text of the supervised problem statement using a query algorithm, the method further includes: obtaining standard documents in a preset format; performing text extraction and data cleaning on the standard documents, removing irrelevant format characters and retaining the text content; identifying the clause number for each line of text using regular expression pattern matching to obtain structured line data with numbers; constructing a multi-branch document tree based on the structured line data using a stack structure according to the organizational hierarchy of the standard documents, wherein the multi-branch document tree includes multiple nodes, each node storing at least the clause number, clause content, and node hierarchy path; and persistently storing the multi-branch document tree, which is used for retrieval and backtracking.

[0097] Collect relevant standard documents for technical supervision of power equipment, such as national standards and industry specifications in PDF and Word formats. These documents must conform to preset format specifications, such as including clear chapter, section, clause, and paragraph numbering. Obtaining standard documents in preset formats provides a unified source of raw data for subsequent structured processing, ensuring the standardization of the processing flow.

[0098] Extract plain text content from preset format files, such as parsing text from PDFs and removing formatting marks from Word documents; filter out irrelevant formatting characters, such as page breaks, special symbols, redundant blank lines, and garbled characters, retaining only the text content with actual meaning, such as clause descriptions and technical parameters, eliminating formatting interference, and obtaining clean text material, laying the foundation for subsequent numbering recognition.

[0099] After cleaning the plain text content, regular expression pattern matching is used to identify the clause number corresponding to each line of text, resulting in structured line data with numbering. This binds each line of text to the corresponding clause code, solving the problem of matching text content with clause identity.

[0100] Regular expressions are rule expressions used to match, find, and replace specific combinations of characters in text. They construct flexible search patterns through a series of predefined characters and syntax, and are widely used in text processing, data validation, information extraction, and other scenarios.

[0101] The structured row data after regular expression matching is used to construct a multi-branch document tree based on the hierarchical features of the clause numbers. The stack is used to record the hierarchical path of the currently processed clauses. Each clause corresponds to a node in the tree, and the node stores key information, including the clause number, clause content, and node hierarchical path. The hierarchical relationship between nodes is determined by the number hierarchy, and finally a multi-branch document tree is formed, which clearly presents the chapter, section, clause, and clause hierarchical structure of the standard document, and makes the fragmented clause content form a logical whole.

[0102] The structure and node information of the multi-branch document tree are persistently stored in a database. Storage methods can include using a relational database to store node attributes and an additional field to record the parent node name to reflect hierarchy; or directly storing the tree structure in a document database. Subsequent query module 112 can perform precise queries based on node numbers, hierarchical paths, or content. It can also quickly trace back to the parent or child clauses of a specific clause through the hierarchical path to understand the contextual relationships between clauses.

[0103] The process described in this embodiment is the core of building a standard document library. The output structured document tree provides a guarantee for accurate retrieval, semantic matching, and result backtracking in subsequent retrieval steps. Clause numbers and hierarchical paths enable the query algorithm to perform targeted matching, avoiding the inefficiency of full-text fuzzy search. The cleaned plain text content and structured numbers allow the semantically enhanced text to more accurately match relevant clauses. The standard document structure, which cannot be directly recognized by computers, is transformed into hierarchical, indexed, and associative structured data, allowing the subsequent query module 112 to accurately locate clauses at a specific level. Furthermore, the large language model 111, when processing clauses, can understand the context of the clauses by combining the hierarchical path, achieving a significant improvement in the efficiency and accuracy of professional retrieval.

[0104] As an optional approach, based on the preliminary search results, the topic description of each relevant clause in the corresponding standard document is determined, including: obtaining the corresponding chapter number in the corresponding standard document based on the preliminary search results; obtaining the node number of the target level by combining the corresponding chapter number with the corresponding multi-branch document tree; traversing all parent nodes above the node number of the target level in the multi-branch document tree of the corresponding standard document; and concatenating the semantic content corresponding to the parent nodes in hierarchical order to construct the topic description of the node number in the standard document.

[0105] From the preliminary search results, the complete number of each relevant clause in the corresponding standard document, i.e., the chapter number, is extracted. Using the chapter number as the anchor point, the specific position of the clause in the standard document structure is located, providing a starting point for subsequent hierarchical backtracking.

[0106] By combining the multi-branch document tree corresponding to the standard document, the structure of the multi-branch document tree reflects the hierarchical relationship of chapters, sections, articles, and clauses. According to preset rules, such as tracing back to the section level or the chapter level, the target level to be traced is determined, and the corresponding node number is found. The setting of the target level is usually related to the granularity of the standard document. For example, power standards often use chapters or sections as thematic units to ensure that the thematic description is neither too broad nor too fragmented. Defining the granularity range of the thematic description avoids the problems of thematic fragmentation caused by tracing back too deeply or thematic ambiguity caused by tracing back too shallowly.

[0107] Starting from the target level node number, traverse upwards through the multi-branched document tree, visiting all its parent nodes, including higher-level nodes, until the root node. For example, the parent nodes of the target level node "3.2" include the chapter-level node "3" and the root node, which is the document title "DL / T596-2021 Preventive Testing Procedures for Power Equipment". The hierarchical path stored in each node of the multi-branched document tree allows direct tracing of parent node relationships, eliminating the need for recalculation when traversing all parent nodes. Collect multi-level contextual information, such as the macro and meso categories to which the clauses belong, to provide complete hierarchical material for the topic description.

[0108] Articles are typically arranged in descending order of hierarchy, following the order of root node, chapter, section, and target level. This approach demonstrates logical progression while maintaining readability. The hierarchical position of each clause is then transformed into semantic topic tags, intuitively reflecting the professional field and sub-category to which the clause belongs.

[0109] The process described in this embodiment associates scattered clauses with their respective macro-themes, enabling the theme description to accurately reflect the professional scope of the clauses. Compared to single clause content, the theme description, through the splicing of content from multiple levels of parent nodes, can reflect the macro-level positioning and meso-level subdivision of the clauses in the standard document, providing a clear thematic basis for subsequent rearrangement. In the power sector, there is a significant reuse of terminology; the theme description, through hierarchical context, can effectively distinguish these terms and avoid semantic confusion. The hierarchical concepts contained in the theme description can also be associated with entities in the enhanced knowledge graph, further strengthening the system's understanding of the professional attributes of the clauses.

[0110] As an optional approach, based on the search terms results and the supervision question statement, the large language model 111 generates relevant terms matching the supervision question statement, including: using an enhanced knowledge graph as a supplementary knowledge base; generating second prompt words based on the supplementary knowledge base, the search terms results, and the supervision question statement; inputting the second prompt words into the large language model 111, and outputting the corresponding relevant terms, the question answer, and supplementary information for the relevant terms.

[0111] The supplementary knowledge base provided in the above embodiments can include entity association information, definitions of professional terms, and reference relationships between standard documents. The supplementary knowledge base provides structured professional knowledge support for the model, thereby reducing erroneous information in the generated content and ensuring the professionalism and accuracy of the answers.

[0112] By combining supplementary knowledge base, search results, and supervisory question statements to generate second prompt words, the large language model can be guided to generate content within the boundaries of professional knowledge, while clarifying the output format and information dimensions to ensure that the results are both comprehensive and standardized.

[0113] Relevant clauses are directly quoted from the original text of the search results to ensure the authority of the basis; the answers to questions use natural language to interpret the matching relationship between the clauses and the questions, improving readability and lowering the understanding threshold for frontline staff; supplementary information on relevant clauses is supplemented with professional background based on knowledge graphs, expanding the practical value of the answers and helping users fully grasp the application scenarios of the clauses. The model prioritizes ensuring the accuracy of the original clause text, and then combines the related information from the knowledge graph for explanation and supplementation, avoiding free interpretation that deviates from professional knowledge.

[0114] The final output stage provided in this embodiment follows the precise clauses after backtracking and rearranging, transforming the structured search results into information that can be directly used by humans; it leverages an enhanced knowledge graph to compensate for the information limitations of a single clause, improving the completeness of the answer; and it meets the needs of different users through the output of related clauses, question answers, and supplementary information related to the clauses. The output maintains the professionalism of power standards while possessing the efficiency of intelligent question answering, perfectly adapting to the practical needs of technical supervision scenarios.

[0115] Embodiments of the present invention also provide a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of the present invention.

[0116] Embodiments of the present invention also provide a computer program product, including a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform the method of an embodiment of the present invention.

[0117] An embodiment of the present invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method of the embodiment of the present invention.

[0118] refer to Figure 3 The present invention will now describe a structural block diagram of an electronic device that can serve as an embodiment of the present invention, serving as an example of a hardware device applicable to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0119] like Figure 3 As shown, the electronic device includes a computing unit 301, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. The RAM 303 may also store various programs and data required for the operation of the electronic device. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0120] Multiple components in the electronic device are connected to I / O interface 305, including: input unit 306, output unit 307, storage unit 308, and communication unit 309. Input unit 306 can be any type of device capable of inputting information into the electronic device. Input unit 306 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of the electronic device. Output unit 307 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 308 may include, but is not limited to, disks and optical discs. Communication unit 309 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, and / or wireless communication transceivers, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0121] The computing unit 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, CPUs, graphics processing units (GPUs), various special-purpose artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processors, controllers, microcontrollers, etc. The computing unit 301 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of the present invention can be implemented as computer programs tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on an electronic device via ROM 302 and / or communication unit 309. In some embodiments, the computing unit 301 can be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0122] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0123] In the context of embodiments of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, or infrared systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0124] It should be noted that the term "comprising" and its variations used in the embodiments of this invention are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The modifications of "one" and "a plurality" mentioned in the embodiments of this invention are illustrative and not restrictive, and those skilled in the art should understand that unless explicitly indicated otherwise in the context, they should be understood as "one or more".

[0125] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0126] The steps described in the method embodiments provided by the present invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of protection of the present invention is not limited in this respect.

[0127] The term "embodiment" in this specification refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily imply the same embodiment, nor does it imply independence or alternativeity from other embodiments. The various embodiments in this specification are described in a related manner, with reference to each other for similar or identical parts. In particular, for apparatus, device, and system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, and relevant details are referred to in the description of the method embodiments.

[0128] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A technical supervision system for power equipment, characterized in that, include: User terminal, the user terminal being used to receive a supervised problem statement input by a user; A cloud server, communicatively connected to the user terminal, includes a large language model, a query module, a storage module, and a rewriting module. The storage module stores standard files, which include relevant clauses. The rewriting module includes a first call port connected to the large language model. The rewriting module is used to input the supervision problem statement into the large language model, receive semantic expansion of the supervision problem statement based on the large language model to obtain pseudo-document content that better fits professional semantics, and concatenate the supervision problem statement and the pseudo-document content to generate semantically enhanced text. The query module is connected to the storage module and the large language model. It is used to retrieve preliminary search results from a pre-stored standard document library based on the semantically enhanced text of the supervision question statement using a query algorithm. The preliminary search results include a preset number of relevant clauses selected from a relevance sequence, which includes multiple relevant clauses sorted by relevance. The module also reorders the relevant clauses in the preliminary search results based on the topic description of the standard document to which the relevant clauses belong and the relevance priority of the supervision question statement. The large language model is used to perform a second rearrangement of the clauses after the first rearrangement based on the deep semantic relevance of the topic description and the supervision question statement to obtain the final search clause results; and to generate a natural language answer by combining the final search clause results and the supervision question statement.

2. The technical supervision system for power equipment according to claim 1, characterized in that, The cloud server further includes: a knowledge graph creation module, used to extract entities from the standard file based on examples of various entities using a large language model to obtain an entity set; using entities in the entity set as seeds, using automated collection tools to collect data on professional information websites to expand the entity set; and storing the entities and entity relationships in the entity set in a pre-built knowledge graph in a structured manner to obtain an enhanced knowledge graph.

3. The technical supervision system for power equipment according to claim 1, characterized in that, The storage module further includes a cleaning submodule, which is used to extract text and clean data from the acquired standard files in a preset format, removing irrelevant format characters and retaining the text content. The regular expression matching submodule, connected to the cleaning submodule, is used to identify the clause number of each line of text by regular expression pattern matching, and obtain structured line data with number; The document tree construction submodule, connected to the regular expression matching submodule, is used to construct a multi-branch document tree based on the structured line data, using a stack structure according to the hierarchical relationship of the standard file. The multi-branch document tree includes multiple nodes, and each node stores at least the clause number, clause content, and node hierarchical path. The persistent storage submodule, connected to the document tree construction submodule, is used to persistently store the multi-branch document tree, wherein the multi-branch document tree is used for retrieval and backtracking.

4. The technical supervision system for power equipment according to claim 1, characterized in that, The cloud server further includes a backtracking and rearrangement module, which is connected to the query module and is used to reorder the relevant terms in the preliminary search results based on the topic description and the supervision question statement using a large language model, so as to obtain the final search terms results.

5. A data processing method for a technical supervision system for power equipment, characterized in that, include: Receive the supervision question statement sent by the user terminal; Based on the description of the supervision problem, the format requirements of the standard document, and the relevant professional field, a first prompt word is generated and input into the large language model; By using a large language model based on a pre-set enhanced knowledge graph, the semantics of the supervision problem statement are semantically expanded to obtain pseudo-document content that is more in line with professional semantics. The supervision problem statement and the pseudo-document content are concatenated to generate semantically enhanced text; Based on the semantically enhanced text of the supervision problem statement, a query algorithm is used to retrieve corresponding preliminary search results from a pre-stored standard document library. The preliminary search results include a preset number of relevant clauses selected from a relevance sequence, and the relevance sequence includes multiple relevant clauses sorted by relevance. Based on the preliminary search results, determine the subject description of each relevant clause in its respective standard document; Based on the subject descriptions of the standard documents to which the relevant clauses in the preliminary search results belong and the relevant priorities of the statements of the supervision issues, the relevant clauses in the preliminary search results are rearranged. Using a large language model, based on the deep semantic relevance between the topic description and the supervision question statement, the clauses after the first rearrangement are rearranged a second time to obtain the final search clause results; Based on the search terms and the supervision question statement, a natural language answer to the supervision question statement is generated by a large language model.

6. The method according to claim 5, characterized in that, By using a large language model based on a pre-defined enhanced knowledge graph, the semantics of the supervised problem statement are semantically expanded to obtain pseudo-document content that better fits the professional semantics, including: Based on examples of various entities, the large language model is used to extract entities from the supervision problem representation to obtain the representation entities; Based on the enhanced knowledge graph, find the adjacent entities that are adjacent to the expressed entity; The pseudo-document content is obtained by semantically expanding the adjacent entities using a large language model.

7. The method according to claim 5, characterized in that, Before semantically expanding the supervised problem statement using a large language model based on a pre-set enhanced knowledge graph to obtain pseudo-document content that better fits professional semantics, the method further includes: Obtain a standard file in plain text format; The standard document is used to extract entities based on examples of various entities using a large language model, resulting in an entity set. Using entities in the entity set as seeds, data is collected from professional information websites using automated data collection tools to expand the entity set; The entities and their relationships in the entity set are stored in a structured manner in a pre-constructed knowledge graph to obtain the enhanced knowledge graph.

8. The method according to claim 5, characterized in that, Before retrieving the corresponding preliminary search results from a pre-stored standard document library based on the semantically enhanced text of the supervision problem statement using a query algorithm, the method further includes: Obtain a standard file in a preset format; The standard file is subjected to text extraction and data cleaning to remove irrelevant formatting characters and retain the text content; For each line of text, the clause number is identified by regular expression pattern matching, resulting in structured line data with numbering. Based on the structured line data, a multi-branch document tree is constructed using a stack structure according to the hierarchical relationship of the standard file. The multi-branch document tree includes multiple nodes, and each node stores at least the clause number, clause content, and node hierarchical path. The multi-branch document tree is persistently stored, and is used for retrieval and backtracking.

9. The method according to claim 5, characterized in that, Based on the preliminary search results, the subject descriptions of each relevant clause in its respective standard document are determined, including: Based on the preliminary search results, obtain the corresponding chapter number in the relevant standard file; Based on the corresponding chapter number and the corresponding multi-branch document tree, obtain the node number of the target level to be traced back; In the multi-branch document tree of the standard document, traverse all parent nodes above the node number of the target level; The semantic content corresponding to the parent node is concatenated in hierarchical order to construct the topic description of the node number in the standard file.

10. The method according to claim 5, characterized in that, Based on the search terms and the supervision question statement, a natural language answer to the supervision question statement is generated by a large language model, including: Enhance the knowledge graph as a supplementary knowledge base; Based on the supplementary knowledge base, the search terms results, and the supervision question statement, a second prompt word is generated; Input the second prompt word into the large language model, and output the corresponding relevant clauses, question answers, and supplementary information for the relevant clauses.

11. An electronic device, comprising: A processor and a memory storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 5 to 10.

Images