Wind turbine generator operation and maintenance knowledge base construction method based on large model and mechanism self-learning

Abstract

The invention discloses a wind turbine generator operation and maintenance knowledge base construction method based on a large model and mechanism self-learning. The wind turbine generator operation and maintenance knowledge base construction method comprises the steps of wind turbine generator operation and maintenance domain knowledge Schema definition and large model cue word template design used for wind turbine generator operation and maintenance knowledge extraction; obtaining operation and maintenance multi-modal data of the wind turbine generator, performing preprocessing, and performing knowledge extraction through a large model based on a designed cue word template; a dynamic knowledge association and wind turbine generator operation and maintenance knowledge base fault mechanism self-learning updating mechanism is established, operation and maintenance data and a knowledge graph are associated in real time, and the knowledge base is automatically learned and updated through an exception triggering mechanism; and constructing and storing a wind turbine generator operation and maintenance knowledge graph based on a knowledge extraction result, generating a semantic association sub-graph through clustering, generating a sub-graph clustering report, and realizing efficient knowledge retrieval. Based on the above content, the wind turbine generator operation and maintenance knowledge base which is efficient, accurate and updated in real time is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning, and specifically relates to a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning. Background Art

[0002] In recent years, China's wind power energy industry has developed rapidly, with the installed capacity continuously expanding, and has become a leader in the global wind power field. With the increase in the scale of wind farms and the improvement of the complexity of wind turbines, the traditional operation and maintenance mode is facing huge challenges. During the operation of wind turbines, a large amount of data will be generated, including real-time operation data collected by sensors, equipment maintenance records, fault reports, expert diagnosis opinions, etc. These data include both structured numerical information and unstructured text descriptions. How to extract effective knowledge from these multi-source heterogeneous data and construct an intelligent operation and maintenance knowledge base has become a key issue in improving the efficiency of wind power operation and maintenance.

[0003] Currently, the construction of knowledge bases in the wind power industry mainly relies on manual collation and expert experience summary. This method is inefficient and it is difficult to ensure the integrity and accuracy of knowledge. Unstructured data (such as maintenance reports, fault cases, etc.) often contains a large number of professional terms and complex semantics, and traditional information extraction techniques are difficult to accurately identify the entities and relationships therein. In addition, existing knowledge bases are usually static and cannot be dynamically associated with the data of the SCADA (Supervisory Control and Data Acquisition) system for the real-time operation of wind turbines, resulting in lagging knowledge updates and making it difficult to provide timely and effective support for fault diagnosis.

[0004] At the technical level, although large language models in the field of natural language processing (such as BERT, GPT, etc.) perform well in general text understanding, due to the lack of professional knowledge in the wind power field, they are prone to deviations when dealing with industry terms and complex fault descriptions. At the same time, the data types involved in wind power operation and maintenance are diverse, including numerical sensor data, text-based maintenance records, etc. How to achieve the unified representation and efficient retrieval of these multi-modal data is also an important challenge currently faced.

[0005] In the prior art, although rule-based knowledge extraction methods can handle some structured data, their flexibility and adaptability are significantly insufficient when faced with complex unstructured texts. And general natural language processing tools have low accuracy in entity recognition and relationship extraction tasks in the wind power field and are difficult to meet the actual application requirements. Although statically constructed knowledge graphs can store historical fault cases, due to the lack of dynamic association with real-time operation data, they cannot adapt to the real-time changes in the operating state of the wind turbines, resulting in a significant reduction in the practical value of the knowledge base.

[0006] In view of the above problems, the present invention proposes a method for constructing an operation and maintenance knowledge base of wind turbine units that integrates large language models and prompt learning, aiming to solve the problems of low efficiency, insufficient accuracy, and poor real-time performance in the construction of existing knowledge bases through domain-adaptive knowledge extraction and knowledge base dynamic association technologies, and provide more reliable knowledge support for intelligent wind power operation and maintenance. Summary of the Invention

[0007] The present invention provides a method for constructing an operation and maintenance knowledge base of wind turbine units based on large models and mechanism self-learning, so as to solve the problems of low efficiency in knowledge extraction relying on manual annotation, disconnection between static knowledge bases and dynamic operating conditions data, difficulty in fusing multi-modal heterogeneous data, and insufficient domain semantic adaptability of general models in the prior art.

[0008] According to a first aspect, in one embodiment, a method for constructing an operation and maintenance knowledge base of wind turbine units based on large models and mechanism self-learning is provided, and the method includes:

[0009] Defining the knowledge Schema of the wind turbine operation and maintenance field, and designing the large model prompt word template for wind turbine operation and maintenance knowledge extraction;

[0010] Obtaining multi-modal data of wind turbine operation and maintenance and performing preprocessing, and performing knowledge extraction through the large model based on the designed prompt word template, and converting the structured data and unstructured data knowledge into a unified vectorized representation;

[0011] Establishing a dynamic knowledge association and a self-learning update mechanism for the fault mechanism of the wind turbine operation and maintenance knowledge base, realizing automatic learning and updating of the knowledge base by real-time associating operation and maintenance data with the knowledge graph through an exception trigger mechanism;

[0012] Constructing and storing the wind turbine operation and maintenance knowledge graph based on the knowledge extraction results, generating semantic association subgraphs through clustering, and generating subgraph clustering reports to achieve efficient knowledge retrieval.

[0013] Furthermore, the definition of the knowledge Schema of the wind turbine operation and maintenance field specifically includes:

[0014] Defining entity categories according to the usage scenarios of wind turbine fault diagnosis, operation and maintenance, collecting and classifying terms in the wind power operation and maintenance field, and extracting terms from the operation and maintenance manuals, SCADA system logs, and historical work order records provided by wind power equipment manufacturers;

[0015] Among them, the entity and relationship definitions in the domain knowledge Schema are respectively:

[0016] Entity definition: including fault types, equipment components, and repair plans;

[0017] Relationship definition: including causal relationship, repair relationship, and association relationship.

[0018] Furthermore, the design of the large model prompt word template for wind turbine operation and maintenance knowledge extraction specifically includes:

[0019] Define the task objective, entity extraction implementation process, entity extraction examples, and support dynamic types for domain adaptation;

[0020] Among them, the task objective includes: guiding the large model to accurately extract predefined specific types of entities and their relationships from multimodal data texts; standardizing the output content structure to support the automatic construction of knowledge graphs;

[0021] The extraction process of entities and relationships includes:

[0022] Entity recognition: Extract the entity name, match the entity name with the preset entity type, and generate an entity description containing attributes and activities;

[0023] Relationship extraction: Establish entity pairs based on the extracted entities, provide relationship explanations, and assign relationship strength weights;

[0024] Formatted output: Use a list of tuples with fixed delimiters to strictly distinguish entity and relationship records.

[0025] Furthermore, obtain wind turbine operation and maintenance multimodal data and perform preprocessing, specifically including:

[0026] Multimodal data processing: If the input data is unstructured image data, use OCR image recognition technology to convert the text materials related to wind power operation and maintenance recorded in the image into text documents; if the input data is structured data including various sensor data of the SCADA system, clean and align the data, including removing outliers and unifying the acquisition times of different sensors to the same time reference.

[0027] Furthermore, perform knowledge extraction through the large model based on the designed prompt word template, and convert the knowledge of structured data and unstructured data into a unified vectorized representation, specifically including:

[0028] Document chunking processing and semantic coherence: Set a sliding window to cut the text document, and set the default window size and overlap rate to ensure semantic coherence;

[0029] Knowledge extraction execution: Input the cut and chunked text into the large model in the field of wind turbine operation and maintenance, and the large model outputs entity pair triples according to the prompt word template;

[0030] Multi-round query to extract entities and relationships: Set the number of multi-round queries, and use the large model to perform multi-round knowledge extraction to ensure entity and relationship extraction;

[0031] Vector Embedding: Use vectorization models including BGE-M3, Sentence-BERT, and node2vec to perform vector embedding on the extracted structured and unstructured data knowledge to generate numerical vectors.

[0032] Furthermore, establish a dynamic knowledge association and a self-learning update mechanism for the fault mechanism of the wind turbine operation and maintenance knowledge base, and realize real-time association of operation and maintenance data with the knowledge graph, and automatically learn and update the knowledge base through an exception trigger mechanism, specifically including:

[0033] Real-time Data Parsing: Receive SCADA system sensor data in real-time and parse the key fields, including device ID, parameter type, value, and timestamp;

[0034] Bidirectionally Bind the SCADA System Alarm Log with the Knowledge Base: Match and bind the alarm log content with the fault entities in the knowledge base;

[0035] Knowledge Base Self-Learning Update Mechanism: If the alarm log fails to match the nodes in the knowledge base, trigger the self-learning process of the fault mechanism, generate and add new fault mechanism entities and relationships, and manually review the knowledge base update;

[0036] Among them, the knowledge base update strategy is:

[0037] Add New Fault Cases: Automatically generate fault triples, including phenomena, causes, and solutions.

[0038] Furthermore, construct and store the wind turbine operation and maintenance knowledge graph based on the knowledge extraction results, generate semantic association subgraphs through clustering, and generate subgraph clustering reports to achieve efficient knowledge retrieval, specifically including:

[0039] Construct knowledge graph nodes and edges based on the extracted entities and relationships, and store them in the Neo4j graph database;

[0040] Use the Leiden algorithm to perform intelligent clustering analysis on the nodes in the knowledge base, and automatically divide the nodes with strong semantic associations into different functional subgraphs;

[0041] Combined with the natural language generation ability of the large model, transform the subgraph module into a dynamic index structure for the retrieval-enhanced generation system, that is, the subgraph clustering report, to support efficient knowledge retrieval and context enhancement in the fault diagnosis scenario.

[0042] Furthermore, the subgraph clustering report refers to a comprehensive document generated by the knowledge graph through algorithm clustering and large model analysis, describing the specific functional subgraphs in the knowledge spectrum, and the content includes:

[0043] Subgraph Theme: Summarize the core semantic categories of the subgraph;

[0044] Core Entities: List the most important nodes and their attributes within the sub - graph;

[0045] Typical Relationships: Describe the most representative association paths within the sub - graph;

[0046] Operation and Maintenance Suggestions: Targeted maintenance solutions given based on the characteristics of the sub - graph.

[0047] According to the second aspect, in one embodiment, a wind turbine operation and maintenance knowledge base construction system based on large - model and mechanism self - learning is provided. The system includes:

[0048] Schema Definition and Prompt Design Module, which is used for defining the knowledge Schema in the field of wind turbine operation and maintenance, and designing the large - model prompt templates for extracting wind turbine operation and maintenance knowledge;

[0049] Multi - source Data Pre - processing and Knowledge Extraction Module, which is used to obtain multi - modal data of wind turbine operation and maintenance and perform pre - processing, and extract knowledge through the large - model based on the designed prompt templates, and convert the structured and unstructured data knowledge into a unified vectorized representation;

[0050] Dynamic Knowledge Association and Knowledge Base Self - learning Update Module, which is used to establish a dynamic knowledge association and a wind turbine operation and maintenance knowledge base fault mechanism self - learning update mechanism, and realize automatic learning and updating of the knowledge base by real - time associating operation and maintenance data with the knowledge graph through an anomaly trigger mechanism;

[0051] Knowledge Graph Storage and Sub - graph Clustering Module, which is used to construct and store the wind turbine operation and maintenance knowledge graph based on the knowledge extraction results, generate semantic - associated sub - graphs through clustering, and generate a sub - graph clustering report to achieve efficient knowledge retrieval.

[0052] According to the third aspect, in one embodiment, an electronic device is provided. The device includes: a processor and a memory;

[0053] The memory is used to store one or more program instructions;

[0054] The processor is used to run one or more program instructions to execute the steps of a method for constructing a wind turbine operation and maintenance knowledge base based on large - model and mechanism self - learning as described in any one of the above.

[0055] According to the fourth aspect, in one embodiment, a computer - readable storage medium is provided. A computer program is stored on the computer - readable storage medium, and when the computer program is executed by a processor, the steps of a method for constructing a wind turbine operation and maintenance knowledge base based on large - model and mechanism self - learning as described in any one of the above are realized.

[0056] The present invention provides a method for constructing an operation and maintenance knowledge base of wind turbines based on large models and mechanism self-learning. By integrating large language models (LLMs) with prompt learning techniques and combining multi-modal data processing and dynamic knowledge association mechanisms, an efficient, accurate, and real-time updated operation and maintenance knowledge base of wind turbines is constructed. It has the following

[0057] Beneficial effects:

[0058] (1) Improvement in the efficiency of domain adaptation knowledge extraction: Through domain-oriented prompt learning strategies, the ability of large language models to extract entities and relationships from unstructured wind power texts is optimized, significantly reducing the dependence on manual annotation. Experiments show that the knowledge extraction efficiency is increased by 40%-50% compared with traditional methods, and the entity recognition accuracy rate reaches more than 90%, effectively covering complex semantic scenarios.

[0059] (2) Dynamic knowledge association and real-time update capabilities: Establish a dynamic binding mechanism between SCADA real-time data and the knowledge graph to achieve synchronous update of the knowledge base and the device operation status. When the sensor detects an anomaly, the system can automatically trigger the correction or addition of fault cases in the knowledge base, and the response delay is reduced to the second level, solving the problem of lag in traditional static knowledge bases.

[0060] (3) Multi-modal data fusion and unified representation: Adopt BGE-M3, Sentence-BERT, and node2vec models to uniformly vectorize and represent structured sensor data and unstructured texts, breaking through the heterogeneous data barrier. This technology improves the cross-modal knowledge retrieval efficiency by 30%, supports joint reasoning of multi-source data in fault diagnosis, and increases the diagnostic coverage rate in complex scenarios by 25%.

[0061] (4) Knowledge graph self-learning and intelligent expansion: Combining the Leiden algorithm with a dynamic new entity review mechanism, the knowledge base can automatically optimize the topological structure according to new fault cases and generate a semantic subgraph clustering report. In practical applications, the construction and operation and maintenance of the knowledge base assisted by large models can significantly reduce the manual maintenance cost.

[0062] (5) Enhancement of operation and maintenance decision-making support capabilities: Through the collaboration of the knowledge graph and large models, the system can provide accurate fault cause analysis and recommended maintenance solutions for operation and maintenance personnel. The generated subgraph clustering report supports intelligent retrieval and semantic display, improving the fault handling efficiency and the interpretability of the solutions.

[0063] Through the above technical means, the present invention realizes the intelligence, dynamics, and domain adaptability of the construction of the wind power operation and maintenance knowledge base, provides an efficient and reliable technical support for fault diagnosis, equipment maintenance, and decision-making optimization, and helps the intelligent transformation of the wind power industry. Description of the Drawings

[0064] Figure 1The overall flowchart of a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning provided by an embodiment of the present invention;

[0065] Figure 2 The specific implementation flowchart of a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning provided by an embodiment of the present invention;

[0066] Figure 3 The flowchart of the Schema definition and prompt word design steps in a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning provided by an embodiment of the present invention;

[0067] Figure 4 The flowchart of the multi-source data preprocessing and knowledge extraction steps in a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning provided by an embodiment of the present invention;

[0068] Figure 5 The flowchart of the dynamic knowledge association and knowledge base self-learning update steps in a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning provided by an embodiment of the present invention;

[0069] Figure 6 The flowchart of the knowledge graph storage and sub-graph clustering steps in a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning provided by an embodiment of the present invention;

[0070] Figure 7 The schematic diagram of the logical structure of a system for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning provided by an embodiment of the present invention. Detailed implementation manners

[0071] The present invention will be further described in detail below in conjunction with the accompanying drawings through specific implementation manners. Similar elements in different implementation manners are labeled with related similar element numbers. In the following implementation manners, many detailed descriptions are provided to enable a better understanding of the present invention. However, those skilled in the art can easily recognize that some of the features can be omitted in different situations, or can be replaced by other elements, materials, or methods. In some cases, some operations related to the present invention are not shown or described in the specification to avoid overwhelming the core part of the present invention with excessive descriptions. For those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations based on the descriptions in the specification and the general technical knowledge in the art.

[0072] In addition, the features, operations, or characteristics described in the specification can be combined in any appropriate manner to form various embodiments. At the same time, the steps or actions in the method description can also be reordered or adjusted in a manner obvious to those skilled in the art. Therefore, the various sequences in the specification and drawings are only for clearly describing a certain embodiment and do not mean a necessary sequence, unless it is stated otherwise that a certain sequence must be followed.

[0073] As Figure 1 shown, a method for constructing an operation and maintenance knowledge base of a wind turbine based on large models and mechanism self-learning provided by the first embodiment of the present invention. By designing a domain-oriented prompt learning strategy, optimizing the entity and relationship extraction ability of the large language model for unstructured text, and achieving automated high-precision knowledge extraction; establishing a dynamic binding mechanism between SCADA real-time data and the knowledge graph to solve the defect that the static knowledge base is difficult to respond to the real-time state changes of equipment; proposing a unified vector representation method for multi-modal data based on BGE-M3 and Sentence-BERT to break through the heterogeneity barrier between structured data (such as sensor time series, equipment parameters) and unstructured text (such as operation and maintenance manuals, fault reports), and realizing efficient fusion and reasoning of cross-modal knowledge. Through the above technical means, the present invention constructs an operation and maintenance knowledge base of wind turbines that is dynamically updated, domain-adapted, and supports multi-modal collaboration, provides accurate and real-time structured knowledge support for fault diagnosis, and significantly improves the reliability and adaptability of the wind power intelligent operation and maintenance system driven by both knowledge and data. The following will be described in detail in conjunction with Figure 1 and Figure 2 for detailed description.

[0074] As Figure 1 shown, in step S100, the definition of the wind turbine operation and maintenance domain knowledge Schema and the design of the large model prompt word template for wind turbine operation and maintenance knowledge extraction.

[0075] As Figure 3 shown, in this step, entity types, relationship types, etc. are defined according to the expert knowledge in the wind power operation and maintenance field, and relevant prompt words for entity extraction of the large model are designed.

[0076] The above steps specifically include:

[0077] S110, structured definition of the wind power fault diagnosis Schema: Define entity categories according to the usage scenarios of wind turbine fault diagnosis operation and maintenance, collect terms in the wind power operation and maintenance field and classify them, and extract terms from operation and maintenance manuals, SCADA system logs, and historical work order records provided by wind power equipment manufacturers.

[0078] Among them, the domain knowledge Schema is used for domain fine-tuning of the wind power operation and maintenance large model, and the entity and relationship definitions in the Schema are respectively:

[0079] Entity definition: Fault type (fault code, symptom description, emergency level), equipment component (component name, model, installation location), repair plan (plan steps, applicable faults, man-hour estimate);

[0080] Relationship definition: Causal relationship (fault A causes fault B), repair relationship (component C is repaired through plan D), association relationship (sensor E detects component F).

[0081] S120, Design of entity and relationship extraction prompt template: Give the task objective, entity extraction implementation process, entity extraction examples, and support dynamic types for domain adaptation.

[0082] Among them, the task objective given in the prompt is: Guide the large language model to accurately extract specific types of entities and their relationships (entity types can be configured) defined in step S110 from multi-modal text; standardize the output content structure to support the automatic construction of knowledge graphs.

[0083] Among them, the extraction process of entities and relationships is as follows:

[0084] 1) Entity recognition. Extract entity names, match the entity names with preset entity types (which can be dynamically configured through parameters to construct domain-specific entity types in the wind power field), and generate entity descriptions containing attributes and activities.

[0085] 2) Relationship extraction. Based on the extracted entities, establish entity pairs (source and target), provide relationship explanations (which need to be based on text evidence), and assign a relationship strength weight of 0 - 10 points.

[0086] 3) Formatted output. Use a list of tuples with fixed delimiters to strictly distinguish entity and relationship records.

[0087] As Figure 1 shown, in step S200, obtain multi-modal data of wind turbine operation and maintenance, preprocess it, and perform knowledge extraction through a large model based on the designed prompt template, converting structured and unstructured data knowledge into a unified vectorized representation.

[0088] As Figure 4 shown, in this step, preprocess the structured data (such as SCADA sensor data) and unstructured data (such as fault reports) of wind power operation and maintenance, use a domain-optimized large model to extract entities (fault types, equipment components, etc.) and relationships (causal, repair, etc.), and output standardized triples for knowledge graph construction.

[0089] The above steps specifically include:

[0090] S210, Multi-modal data processing:

[0091] If the input data is unstructured data, such as PDF images, the OCR image recognition technology is used to convert the text materials related to wind power operation and maintenance, such as operation and maintenance manuals and wind turbine instructions recorded in PDF and pictures, into text documents; if the input data is structured data, such as sensor data transmitted by the SCADA system, the data is cleaned and aligned, the outliers in the SCADA system data are removed, and the acquisition times of different sensors integrated in the SCADA system are unified to the same time base (UTC+8).

[0092] S220, Blocking and Semantic Coherence: Set a sliding window of a given text token size to cut the text document. The default window size is set to 512 tokens, and the overlap rate is 20% to ensure semantic coherence.

[0093] S230, Knowledge Extraction Execution: Input the segmented text into a large model in the wind power field (Deepseek fine-tuned model), and the large model outputs entity pair triples according to the prompt template (such as "relationship" <|> "gearbox vibration anomaly" <|> "bearing wear" <|> "bearing wear may cause gearbox vibration anomaly" <|>).

[0094] S240, Multi-round Query to Extract Entities and Relationships to Prevent Omissions of Entities and Relationships: Set the number of multi-round queries, and use the large model for multi-round knowledge extraction to ensure the extraction of entities and relationships.

[0095] S250, Vector Embedding: Use vector embedding models such as BGE-M3, Sentence-BERT, and node2vec to perform vector embedding on the structured and unstructured data knowledge extracted above to generate numerical vectors.

[0096] As Figure 1 shown, in step S300, a dynamic knowledge association and a self-learning update mechanism for the fault mechanism of the wind turbine operation and maintenance knowledge base are established to realize automatic learning and updating of the knowledge base by real-time associating operation and maintenance data with the knowledge graph through an anomaly trigger mechanism.

[0097] As Figure 5 shown, in this step, the SCADA data is real-time associated with the knowledge graph, and the fault triples (phenomenon - cause - solution) are automatically updated through an anomaly trigger mechanism, and the knowledge base is continuously optimized in combination with manual review.

[0098] The above steps specifically include:

[0099] S310, Real-time Data Parsing: Real-time receive SCADA system sensor data (transmitted in JSON format), and parse the key fields: device ID, parameter type (temperature, vibration, etc.), value, timestamp;

[0100] S320: Bidirectionally bind the alarm logs of SCADA with the knowledge base, and match and bind the alarm log content (such as "generator temperature exceeding the limit") with the fault entities in the knowledge base (such as "generator overheating").

[0101] S330: Knowledge base self-learning update mechanism. If the alarm log and the nodes in the knowledge base fail to match in step S320, trigger the process of adding new entities (requiring manual review).

[0102] Among them, the knowledge base update strategy is as follows:

[0103] Add new fault cases: Automatically generate triples (phenomenon -> cause -> solution). For example, "temperature continuously rising and accompanied by abnormal noise" → output "cooling system blocked → clean the heat dissipation channel".

[0104] As Figure 1 shown, in step S400, construct and store the wind turbine operation and maintenance knowledge graph based on the knowledge extraction results, generate semantic association subgraphs through clustering, and generate subgraph clustering reports to achieve efficient knowledge retrieval.

[0105] As Figure 6 shown, this step constructs the nodes and edges of the knowledge graph based on the entities and relationships extracted in step S200, and stores them in the Neo4j graph database; generates semantic association subgraphs through clustering using the Leiden algorithm, and constructs a retrievable index (i.e., subgraph clustering report) for the RAG system to achieve efficient knowledge retrieval.

[0106] The above steps specifically include:

[0107] S410: Based on the storage architecture design of the Neo4j graph database, structurally model the nodes (such as fault types, equipment components, repair plans, etc.) and their associated relationships in the wind turbine operation and maintenance knowledge graph, and construct a complete knowledge graph storage system.

[0108] S420: Based on the constructed knowledge graph, use the Leiden algorithm to perform intelligent clustering analysis on the nodes in the knowledge base, and automatically divide the nodes with strong semantic associations into different functional subgraphs.

[0109] S430: Combine the natural language generation ability of the large language model to convert the subgraph module into a dynamic index structure (i.e., subgraph clustering report) for the retrieval-augmented generation (RAG) system, supporting efficient knowledge retrieval and context enhancement in the fault diagnosis scenario.

[0110] Among them, the subgraph clustering report refers to a comprehensive document generated by the knowledge graph through algorithm clustering and large model analysis, describing specific functional subgraphs in the knowledge spectrum, and its content includes:

[0111] 1. Sub - graph Theme: Summarize the core semantic category of the sub - graph (e.g., "Fault diagnosis related to gearboxes");

[0112] 2. Core Entities: List the most important nodes and their attributes within the sub - graph;

[0113] 3. Typical Relationships: Describe the most representative association paths within the sub - graph;

[0114] 4. Operation and Maintenance Suggestions: Targeted maintenance solutions based on the characteristics of the sub - graph.

[0115] According to the above content:

[0116] (1) This embodiment proposes an adaptive prompt learning and large - model optimization technology in the field of wind turbine fault diagnosis. It designs a dynamically configurable prompt - word template based on professional terms in the wind power operation and maintenance field, and guides the large - language model to accurately extract entities and relationships of fault phenomena, causes, and repair solutions through a multi - round interrogation mechanism, solving the semantic understanding deviation problem of general models for complex wind power fault descriptions (such as multi - level causal chains and compound faults).

[0117] (2) This embodiment proposes a unified vectorization representation and fusion technology for multi - modal heterogeneous data, covering cross - modal embedding methods based on BGE - M3, Sentence - BERT, and node2vec, to achieve vector - space alignment of structured sensor data (such as SCADA system time - series data) and unstructured text (such as fault reports), in order to solve the problem that it is difficult to jointly analyze and diagnose multi - source heterogeneous wind power operation data and operation and maintenance knowledge.

[0118] (3) This embodiment proposes a method for constructing a dynamic index of a knowledge graph for a RAG system. Based on the knowledge - graph sub - graph clustering technology of the Leiden algorithm, it automatically organizes entities and relationships in the wind power operation and maintenance field into semantically associated sub - graph modules, and transforms them into a retrievable index structure (sub - graph clustering report) dedicated to the RAG system, realizing efficient recall of knowledge and context enhancement.

[0119] (4) This embodiment proposes a dynamic self - learning technology of a knowledge graph driven by the real - time working conditions of wind turbines. By parsing the monitoring parameters of wind turbines in the SCADA system (such as generator winding temperature, gearbox oil pressure, blade vibration amplitude), it matches real - time alarm logs with fault nodes in the knowledge graph. If the match fails, it automatically generates wind power fault triples, self - learns and updates the content of the knowledge base, and combines an artificial review mechanism to ensure that the newly added cases comply with wind power safety specifications, realizing the synchronous response of the knowledge base content and the operating state of the unit.

[0120] Corresponding to the above - disclosed method for constructing an operation and maintenance knowledge base of a wind turbine based on a large - model and mechanism self - learning, an embodiment of the present invention also discloses a system for constructing an operation and maintenance knowledge base of a wind turbine based on a large - model and mechanism self - learning, as Figure 7 shown, which specifically includes:

[0121] A Schema definition and prompt - word design module, which is used for defining the knowledge Schema in the field of wind turbine operation and maintenance, and designing the large - model prompt - word template for extracting wind turbine operation and maintenance knowledge;

[0122] A multi - source data pre - processing and knowledge extraction module, which is used to obtain multi - modal data of wind turbine operation and maintenance and perform pre - processing, and extract knowledge through a large - model based on the designed prompt - word template, and convert the structured data and unstructured data knowledge into a unified vectorized representation;

[0123] A dynamic knowledge association and knowledge - base self - learning update module, which is used to establish a dynamic knowledge association and a fault mechanism self - learning update mechanism for the wind turbine operation and maintenance knowledge base, and realize automatic learning and updating of the knowledge base by real - time associating operation and maintenance data with the knowledge graph through an abnormal trigger mechanism;

[0124] A knowledge - graph storage and sub - graph clustering module, which is used to construct and store the wind turbine operation and maintenance knowledge graph based on the knowledge extraction results, generate semantic - associated sub - graphs through clustering, and generate a sub - graph clustering report to achieve efficient knowledge retrieval.

[0125] It should be noted that for the detailed description of a system for constructing an operation and maintenance knowledge base of a wind turbine based on a large - model and mechanism self - learning provided by an embodiment of the present invention, reference can be made to the relevant description of a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large - model and mechanism self - learning provided by an embodiment of the present invention, which will not be elaborated here.

[0126] In addition, an embodiment of the present invention also provides an electronic device, the device includes: a processor and a memory; the memory is used to store one or more program instructions; the processor is used to run one or more program instructions to execute the steps of a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large - model and mechanism self - learning as described in any one of the above.

[0127] It should be noted that for the detailed description of an electronic device provided by an embodiment of the present invention, reference can be made to the relevant description of a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large - model and mechanism self - learning provided by an embodiment of the present application, which will not be elaborated here.

[0128] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning as described in any one of the above are implemented.

[0129] It should be noted that for a detailed description of a computer-readable storage medium provided in an embodiment of the present invention, reference may be made to the relevant description of a method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning provided in an embodiment of the present application, which will not be elaborated here.

[0130] Those skilled in the art can understand that all or part of the functions of the above methods can be implemented in a hardware manner or in a computer program manner. When all or part of the functions in the above embodiments are implemented in a computer program manner, the program can be stored in a computer-readable storage medium, and the storage medium may include: read-only memory, random access memory, magnetic disk, optical disk, hard disk, etc., and the above functions are implemented by a computer executing the program. For example, the program is stored in the memory of the device, and when the processor executes the program in the memory, the above all or part of the functions can be implemented. In addition, when all or part of the functions in the above embodiments are implemented in a computer program manner, the program can also be stored in a storage medium such as a server, another computer, magnetic disk, optical disk, flash drive or mobile hard disk, and is saved to the memory of the local device by downloading or copying, or the system of the local device is updated, and when the processor executes the program in the memory, the above all or part of the functions in the embodiments can be implemented.

[0131] The above uses specific examples to elaborate on the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the technical field to which the present invention pertains, several simple deductions, deformations or substitutions can be made according to the idea of the present invention.

Claims

1. A method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning, characterized in that The method includes: Defining the knowledge Schema in the field of wind turbine operation and maintenance, and designing the prompt word template for knowledge extraction of wind turbine operation and maintenance; Obtaining multi-modal data of wind turbine operation and maintenance and preprocessing it, and performing knowledge extraction through a large model based on the designed prompt word template, converting structured data and unstructured data knowledge into a unified vector representation; Establishing a dynamic knowledge association and a self-learning update mechanism for the fault mechanism of the wind turbine operation and maintenance knowledge base, realizing real-time association of operation and maintenance data with the knowledge graph, and automatically learning and updating the knowledge base through an exception trigger mechanism; Constructing and storing the wind turbine operation and maintenance knowledge graph based on the knowledge extraction results, generating semantic association subgraphs through clustering, and generating subgraph clustering reports to achieve efficient knowledge retrieval.

2. The method for constructing an operation and maintenance knowledge base of a wind turbine based on large models and mechanism self-learning according to claim 1, characterized in that Defining the knowledge Schema in the field of wind turbine operation and maintenance, specifically including: Defining entity categories according to the usage scenarios of wind turbine fault diagnosis, operation and maintenance, collecting and classifying terms in the field of wind power operation and maintenance, and extracting terms from operation and maintenance manuals, SCADA system logs, and historical work order records provided by wind power equipment manufacturers; Among them, the entity and relationship definitions in the domain knowledge Schema are respectively: Entity definition: including fault type, equipment component, repair plan; Relationship definition: including causal relationship, repair relationship, association relationship.

3. The method for constructing an operation and maintenance knowledge base of a wind turbine based on large models and mechanism self-learning according to claim 1, wherein, Designing the prompt word template for knowledge extraction of wind turbine operation and maintenance, specifically including: Defining the task objective, the implementation process of entity extraction, entity extraction examples, and supporting dynamic types for domain adaptation; Among them, the task objective includes: guiding the large model to accurately extract predefined specific types of entities and their relationships from multi-modal data texts; standardizing the output content structure to support the automatic construction of the knowledge graph; The extraction process of entities and relationships includes: Entity recognition: extracting entity names, matching the entity names with predefined entity types, and generating entity descriptions containing attributes and activities; Relationship extraction: establishing entity pairs based on the extracted entities, providing relationship explanations, and assigning relationship strength weights; Formatted output: using a list of tuples with fixed delimiters to strictly distinguish entity and relationship records.

4. The method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning according to claim 1, characterized in that, Obtaining multi-modal data of wind turbine operation and maintenance and preprocessing it, specifically including: Multi-modal data processing: if the input data is unstructured image data, using OCR image recognition technology to convert the text materials related to wind power operation and maintenance recorded in the image into text documents; if the input data is structured data including various sensor data of the SCADA system, cleaning and aligning the data, including removing outliers and unifying the acquisition time of different sensors to the same time base.

5. A method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning, characterized in that, Performing knowledge extraction through a large model based on the designed prompt word template, converting structured data and unstructured data knowledge into a unified vector representation, specifically including: Document chunking processing and semantic coherence: setting a sliding window to cut the text document, setting the default window size and overlap rate to ensure semantic coherence; Knowledge extraction execution: inputting the cut and chunked text into the large model in the field of wind turbine operation and maintenance, and the large model outputs entity pair triples according to the prompt word template; Multi-round query for entity and relationship extraction: Set the number of multi-round queries, and use the large model for multi-round knowledge extraction to ensure entity and relationship extraction; Vector embedding: Use vectorization models including BGE-M3, Sentence-BERT, and node2vec to perform vector embedding on the extracted structured and unstructured data knowledge to generate numerical vectors.

6. The method for constructing an operation and maintenance knowledge base of a wind turbine based on large models and mechanism self-learning according to claim 1, characterized in that Establish a dynamic knowledge association and a self-learning update mechanism for the fault mechanism of the wind turbine operation and maintenance knowledge base, and realize real-time association of operation and maintenance data with the knowledge graph, and automatically learn and update the knowledge base through an exception trigger mechanism, specifically including: Real-time data parsing: Receive SCADA system sensor data in real time and parse key fields, including device ID, parameter type, value, and timestamp; Bidirectionally bind the SCADA system alarm log with the knowledge base: Match and bind the alarm log content with the fault entities in the knowledge base; Knowledge base self-learning update mechanism: If the alarm log fails to match the nodes in the knowledge base, trigger the self-learning process of the fault mechanism, generate and add new fault mechanism entities and relationships, and manually review the knowledge base update; Among them, the knowledge base update strategy is: Add new fault cases: Automatically generate fault triples, including phenomena, causes, and solutions.

7. A method for constructing an operation and maintenance knowledge base of a wind turbine based on a large model and mechanism self-learning, characterized in that, Construct and store the wind turbine operation and maintenance knowledge graph based on the knowledge extraction results, generate semantic association subgraphs through clustering, and generate subgraph clustering reports to achieve efficient knowledge retrieval, specifically including: Construct knowledge graph nodes and edges based on the extracted entities and relationships, and store them in the Neo4j graph database; Use the Leiden algorithm to perform intelligent clustering analysis on the nodes in the knowledge base, and automatically divide the nodes with strong semantic associations into different functional subgraphs; Combined with the natural language generation ability of the large model, transform the subgraph module into a dynamic index structure for the retrieval-enhanced generation system, that is, the subgraph clustering report, to support efficient knowledge retrieval and context enhancement in the fault diagnosis scenario.

8. A method for constructing an operation and maintenance knowledge base of a wind turbine based on large models and mechanism self-learning, as claimed in claim 7, wherein The subgraph clustering report refers to a comprehensive document generated by the knowledge graph through algorithm clustering and large model analysis, which describes specific functional subgraphs in the knowledge spectrum, and the content includes: Subgraph theme: Summarize the core semantic categories of the subgraph; Core entities: List the most important nodes and their attributes within the subgraph; Typical relationships: Describe the most representative association paths within the subgraph; Operation and maintenance suggestions: Targeted maintenance plans based on subgraph features.

9. A wind turbine operation and maintenance knowledge base construction system based on large models and mechanism self-learning, characterized in that, The system includes: Schema definition and prompt word design module, used for knowledge Schema definition in the field of wind turbine operation and maintenance, and design of large model prompt word templates for wind turbine operation and maintenance knowledge extraction; Multi-source data preprocessing and knowledge extraction module, used to obtain multi-modal data of wind turbine operation and maintenance and perform preprocessing, and perform knowledge extraction through the large model based on the designed prompt word templates, and convert structured and unstructured data knowledge into a unified vectorized representation; Dynamic knowledge association and knowledge base self-learning update module, used to establish a dynamic knowledge association and a self-learning update mechanism for the fault mechanism of the wind turbine operation and maintenance knowledge base, and realize real-time association of operation and maintenance data with the knowledge graph, and automatically learn and update the knowledge base through an exception trigger mechanism; The knowledge graph storage and subgraph clustering module is used to construct and store the wind turbine operation and maintenance knowledge graph based on the knowledge extraction results, generate semantic association subgraphs through clustering, and generate subgraph clustering reports to achieve efficient knowledge retrieval.

10. An electronic device, characterized in that, The device includes: a processor and a memory; The memory is used to store one or more program instructions; The processor is used to run one or more program instructions to execute the steps of a method for constructing a wind turbine operation and maintenance knowledge base based on large models and mechanism self-learning as described in any one of claims 1 to 8.

Images