Multi-agent empowerment type fan fault causal graph autonomous reasoning method and system

Abstract

This invention belongs to the field of reliability assessment technology for complex equipment operation and maintenance processes, and relates to a multi-agent-enabled autonomous reasoning method and system for wind turbine fault causal graphs. The method includes: acquiring wind turbine operation and maintenance text data; calling a relationship extraction agent to determine the fault operation and maintenance graph based on the wind turbine operation and maintenance text data and the FMEA structure table; acquiring user query commands; analyzing user query commands to determine the problem type and complexity and generating a route schedule; calling a causal reasoning agent to perform evidence retrieval based on the route schedule and generating causal chain reasoning results; calling an evidence review agent to review the rationality and relevance of the causal chain reasoning results based on preset review rules; if successful, outputting the causal chain reasoning results; if unsuccessful, returning to re-call the causal reasoning agent until a preset maximum exit mechanism is met. This solves the practical engineering problems of severe illusion propagation in existing single-agent systems and low matching rates for complex queries in traditional text knowledge bases.

Description

Technical Field

[0001] This invention belongs to the field of reliability assessment technology for complex equipment operation and maintenance processes, specifically involving a multi-agent-enabled autonomous reasoning method and system for wind turbine fault cause-effect graph. Background Technology

[0002] Wind turbines, as complex equipment, have diverse subsystems and components with strong interrelationships and varied failure mechanisms, posing a significant challenge to rapid and accurate fault location and maintenance. Therefore, developing fault causal tracing and intelligent fault maintenance decision-making methods is of great engineering significance. Traditional manual fault handling decision-making methods are not only inefficient but also highly dependent on the professional level and field experience of maintenance personnel, making them prone to subjective judgment bias and decision-making errors. This can lead to untimely fault handling, unreasonable solutions, and even increased fault losses and downtime risks. In recent years, with the rapid development of large-scale model technology, building a fault maintenance knowledge base based on wind turbine operation and maintenance knowledge and combining it with large-scale models to achieve intelligent fault maintenance decision-making has become the mainstream technical approach. However, this method still has the following significant shortcomings: Single-agent architectures struggle to handle complex multi-objective reasoning tasks. Existing methods rely on a single agent to perform multi-stage complex causal reasoning processes, including problem understanding, knowledge retrieval, answer review, and result integration, through prompt word engineering. However, the objectives differ at each stage, and single-agent architectures lack clear division of labor and collaboration mechanisms for such multi-objective reasoning tasks. If an error occurs in a single stage, it can propagate along the task flow to subsequent stages, creating a chain reaction of errors that ultimately render the decision completely unreliable, failing to meet the reliability and robustness requirements of industrial-grade complex reasoning tasks. Therefore, a multi-agent architecture is needed, based on task division strategies and swarm intelligence, to achieve complex reasoning tasks requiring multi-objective and multi-agent collaboration.

[0003] Existing knowledge retrieval solutions are mainly based on text vector knowledge bases, retrieving possible answers through semantic similarity. Each time, they can only recall a few semantically similar text blocks based on the query vector, lacking a global understanding of the entire document set or multi-source knowledge system. However, complex queries in industrial settings often involve cross-document and cross-module knowledge integration, requiring models to extract key clues from massive amounts of information and logically connect them. Existing local retrieval mechanisms not only fail to meet the need for such global knowledge summarization and induction, but also easily miss key information or produce factual illusion errors in complex tasks requiring multi-step reasoning, leading to low reasoning credibility and poor engineering applicability. There are gaps in coverage for complex scenarios, necessitating the development of retrieval solutions adapted to such complex reasoning tasks. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-agent-enabled wind turbine fault causal graph autonomous reasoning method and system to solve the practical engineering problems of severe illusion transmission in single-agent systems and low matching rate of complex queries in traditional text knowledge bases in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a multi-agent-enabled autonomous reasoning method for wind turbine fault causal graph analysis, comprising: The relationship extraction agent is invoked to obtain wind turbine operation and maintenance text data. The operation and maintenance text data is segmented to obtain structured text blocks. The wind turbine fault operation and maintenance graph framework is constructed by combining the entity types and relationship types preset in the FMEA structure table. The wind turbine fault operation and maintenance graph is determined based on the wind turbine fault operation and maintenance graph framework and the structured text blocks. Obtain user query commands, analyze user query commands to determine the question type and complexity, and generate route scheduling based on question type and complexity; The causal reasoning agent is invoked according to the routing schedule. The causal reasoning agent performs evidence retrieval based on the user's query command and generates causal chain reasoning results that include fault symptoms, potential causes, impact consequences and maintenance suggestions. The evidence review agent is invoked to review the rationality and relevance of the causal chain reasoning results based on preset review rules. If the review is successful, the causal chain reasoning results are output. If the review fails, the process returns to invoke the causal reasoning agent again to retrieve evidence until the preset maximum exit mechanism is met.

[0006] Optionally, the step of acquiring wind turbine operation and maintenance text data and performing block processing on the operation and maintenance text data to obtain structured text blocks includes: Acquire wind turbine operation and maintenance text data, and set the basic block length and overlapping sampling length based on the block parameter preset strategy of length constraint and semantic integrity, according to the text length distribution, model input upper limit and operation and maintenance semantic integrity. The step size is determined based on the basic block length and the overlapping sampling length. The basic text blocks are obtained by using the continuous sampling method of overlapping sliding windows. An adaptive window fine-tuning method based on semantic boundary detection is used to perform boundary detection on each basic text block, resulting in a semantically complete structured text block.

[0007] Optionally, the call relationship extraction agent constructs a wind turbine fault operation and maintenance graph framework by combining the entity types and relationship types preset in the FMEA structure table, and determines the wind turbine fault operation and maintenance graph based on the wind turbine fault operation and maintenance graph framework and structured text blocks, specifically including the following steps: A framework for constructing a wind turbine fault operation and maintenance graph is built by combining the entity types and relationship types preset in the FMEA structure table; System prompt word templates are generated based on the wind turbine fault operation and maintenance graph structure framework; based on the system prompt word templates, the entity type and relation type of the structured text block are extracted using a large language model to obtain the structured extraction results, and the structure extraction results are converted into data structures that the graph can write to form candidate knowledge units; Candidate knowledge units are standardized into graph triples. A similarity-based entity ablation method is used to align and merge the entity types of the graph triples and the wind turbine fault operation and maintenance graph framework to obtain entity vectors. The similarity between entity vectors is calculated based on cosine distance. A similarity threshold is used to filter highly similar pairs. If the similarity between entities is not less than a preset value, disambiguation and merging are performed on the entity vectors to obtain a consistent fault operation and maintenance map.

[0008] Optionally, the causal reasoning agent performs evidence retrieval based on the user's query command, generating a causal chain reasoning result that includes fault symptoms, potential causes, impact consequences, and maintenance suggestions, specifically including: Obtain the user's query command and vectorize it to obtain the query vector. The semantic similarity between the query vector and the entity vectors in the fault operation and maintenance graph is calculated, and several entity types with high similarity are identified as candidate entities. Knowledge expansion is performed based on candidate entities, and candidate relations and candidate text units are constructed. The candidate relations are all edges with candidate entities as source nodes or target nodes; the candidate text units are structured text blocks containing at least one candidate entity. Under the limited constraints of the LLM context window, candidate relations and candidate text units are sorted and filtered. Candidate entities, sorted and filtered candidate relationships, and candidate text units are concatenated into a structured context, which is then submitted to the LLM along with the user's question to generate causal inference results.

[0009] Optionally, the sorting rules for sorting and filtering candidate relations and candidate text units under the constraint of a limited LLM context window are as follows: Candidate relationships are divided into intra-network relationships and extra-network relationships; Within a network, relationships are sorted in descending order by combinatorial degree, which is the sum of the degree of the source node and the degree of the target node; outside a network, relationships are sorted in descending order by the number of network entities connected to the relationship, with relationships within a network taking precedence over relationships outside a network. For candidate text units, sort them in descending order according to the highest semantic similarity among the candidate entities associated with the text unit. If the highest similarity is the same, sort them in descending order according to the number of relations associated with the text unit.

[0010] Optionally, the reasonableness and relevance of the causal chain reasoning result are reviewed based on preset review rules. If the review passes, the causal chain reasoning result is output; if it fails, the process returns to call the causal reasoning agent again to retrieve evidence until the preset maximum exit mechanism is met, expressed by the following formula:

[0011] in, This indicates the final output of the fault causal chain retrieval results that meet the review requirements; This represents the candidate result of the fault causal chain generated by the causal reasoning agent in round t; This represents the candidate results of the fault causal chain generated in the previous round; This represents the evidence validity review function, used to verify the validity of candidate results based on rules and experience; Represents a set of physical rules; This represents a set of expert experience. This represents the causal chain reconstruction function, used to correct results that fail the review and generate new candidate results; Indicates the current iteration round; This indicates the maximum number of iterations.

[0012] In a second aspect, the present invention provides a multi-agent-enabled wind turbine fault cause-effect graph autonomous reasoning system, comprising: The central routing decision-making agent is used to obtain user query instructions, analyze user query instructions to determine the problem type and complexity, generate route scheduling based on the problem type and complexity, call causal reasoning agents and evidence review agents based on the route scheduling, and generate wind turbine fault operation and maintenance map by calling relationship extraction agents. A relation extraction agent is used to acquire wind turbine operation and maintenance text data, and to process the operation and maintenance text data into blocks to obtain structured text blocks. A wind turbine fault operation and maintenance graph framework is constructed by combining the entity types and relation types preset in the FMEA structure table. The wind turbine fault operation and maintenance graph is determined based on the wind turbine fault operation and maintenance graph framework and the structured text blocks. The causal reasoning agent is used to retrieve evidence according to user query commands, generate causal chain reasoning results including fault symptoms, potential causes, impact consequences and maintenance suggestions, and feed the causal chain reasoning results back to the central routing decision agent. The evidence review agent is used to review the rationality and relevance of the causal chain reasoning results based on preset review rules. If it passes, it outputs the causal chain reasoning results; if it fails, it feeds back to the central routing decision agent, which then calls the causal reasoning agent to retrieve evidence again until the preset maximum exit mechanism is met.

[0013] Optionally, the evidence review agent reviews the reasonableness and relevance of the causal chain reasoning results based on preset review rules. If the review passes, the causal chain reasoning results are output; if not, the agent returns to re-invoke the causal reasoning agent to retrieve evidence until a preset maximum exit mechanism is met, expressed by the following formula:

[0014] in, This indicates the final output of the fault causal chain retrieval results that meet the review requirements; This represents the candidate result of the fault causal chain generated by the causal reasoning agent in round t; This represents the candidate results of the fault causal chain generated in the previous round; This represents the evidence validity review function, used to verify the validity of candidate results based on rules and experience; Represents a set of physical rules; This represents a set of expert experience. This represents the causal chain reconstruction function, used to correct results that fail the review and generate new candidate results; Indicates the current iteration round; This indicates the maximum number of iterations.

[0015] In a third aspect, the present invention provides an electronic device, characterized in that it includes a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the autonomous reasoning method for multi-agent-enabled wind turbine fault causal graphs as described in any one of the preceding claims.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the autonomous reasoning method for multi-agent-enabled wind turbine fault causal graphs as described in any one of the preceding claims.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This application discloses a multi-agent-enabled autonomous reasoning method for wind turbine fault causal graphs. Through a causal reasoning agent retrieving evidence based on user query commands, it generates a complete causal chain reasoning result containing fault symptoms, potential causes, impact consequences, and maintenance suggestions. This effectively solves the problems of existing methods' inability to handle complex faults and insufficient causal reasoning capabilities, achieving an upgrade from single fault diagnosis to multi-factor causal chain tracing. By analyzing the question type and complexity of user query commands to generate routing scheduling, the appropriate agent can be dynamically invoked according to task requirements. This avoids the shortcomings of traditional static knowledge bases, which cannot adapt to diverse queries and cannot dynamically expand their knowledge structure, significantly improving the system's adaptability to changing operation and maintenance scenarios. An evidence review agent is introduced to review the rationality and relevance of the causal chain reasoning results based on preset review rules. If the results fail, a retry mechanism is automatically triggered until the maximum exit condition is met, thus ensuring the accuracy and credibility of the output results and overcoming the problems of difficult traceability and poor interpretability of conventional black-box reasoning results. Multi-agent collaboration (causal reasoning agent, evidence review agent, and central routing decision-making agent), combined with wind turbine fault operation and maintenance maps and structured textual evidence, enables accurate matching and complete causal tracing for complex query tasks. This provides operation and maintenance personnel with clear and traceable decision-making basis, significantly improving the automation, intelligence, and decision reliability of wind turbine fault operation and maintenance. Through a pre-set maximum exit mechanism, iteration can be reasonably terminated when the reasoning result fails review multiple times, avoiding infinite loops or invalid outputs. Simultaneously, the ability to re-retrieve and reason is retained, enabling the system to stably output valid results even in complex data environments.

[0018] This application discloses a wind turbine fault causal reasoning system based on collaborative multi-agent systems and fault maps. It acquires wind turbine operation monitoring data and maintenance text data, and constructs a wind turbine fault maintenance map based on relation extraction agents. Causal reasoning agents retrieve evidence from the fault map and structured text, generating causal chain reasoning results including fault symptoms, potential causes, impact consequences, and maintenance suggestions. An evidence review agent reviews the reliability of the output results, and a central routing decision-making agent uniformly schedules the multiple agents to ensure the rationality and interpretability of the reasoning results. This invention, through multi-agent collaborative operation and maintenance, achieves accurate matching of complex query tasks and complete causal chain reasoning and tracing, significantly improving the intelligence level and decision reliability of wind turbine fault operation and maintenance. Attached Figure Description

[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the block processing flow of operation and maintenance text data in an embodiment of the present invention. Figure 3 A flowchart illustrating the construction of a knowledge graph for wind turbine operation and maintenance faults in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the generation process of causal chain reasoning results in an embodiment of the present invention. Figure 5 This is a schematic diagram of the central routing decision-making intelligent agent scheduling process according to an embodiment of the present invention; Figure 6 This is a system structure block diagram according to an embodiment of the present invention; Figure 7 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0021] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0022] This invention addresses the shortcomings of existing wind turbine fault diagnosis methods, such as reliance on static knowledge bases, difficulty in handling complex faults, inability to dynamically expand knowledge structures, and insufficient causal reasoning capabilities. It proposes a causal chain reasoning method based on multi-agent collaboration and fault graphs. See also... Figure 1 This invention discloses a multi-agent-enabled autonomous reasoning method for wind turbine fault causal graphs, comprising: The relationship extraction agent is invoked to obtain wind turbine operation and maintenance text data. The operation and maintenance text data is segmented to obtain structured text blocks. The wind turbine fault operation and maintenance graph framework is constructed by combining the entity types and relationship types preset in the FMEA structure table. The wind turbine fault operation and maintenance graph is determined based on the wind turbine fault operation and maintenance graph framework and the structured text blocks. Obtain user query commands, analyze user query commands to determine the question type and complexity, and generate route scheduling based on question type and complexity; The causal reasoning agent is invoked according to the routing schedule. The causal reasoning agent performs evidence retrieval based on the user's query command and generates causal chain reasoning results that include fault symptoms, potential causes, impact consequences and maintenance suggestions. The evidence review agent is invoked to review the rationality and relevance of the causal chain reasoning results based on preset review rules. If the review is successful, the causal chain reasoning results are output. If the review fails, the process returns to invoke the causal reasoning agent again to retrieve evidence until the preset maximum exit mechanism is met.

[0023] The method of this invention generates a complete causal chain that includes fault symptoms, potential causes, impact consequences, and maintenance suggestions by jointly retrieving evidence from graphs and structured text blocks based on user query instructions. This effectively handles complex faults. At the same time, it reviews the rationality and relevance of the reasoning results based on preset review rules. If the results fail, the search is repeated until the exit mechanism is met. This significantly improves the reliability, interpretability, and accurate matching ability of the reasoning results for complex queries, and realizes the dynamic updating and causal tracing of wind turbine fault operation and maintenance knowledge.

[0024] In some embodiments, the causal reasoning agent performs evidence retrieval based on user query instructions, generating a causal chain reasoning result that includes fault symptoms, potential causes, impact consequences, and maintenance suggestions, specifically including: Obtain the user's query command and vectorize it to obtain the query vector. The semantic similarity between the query vector and the entity vectors in the fault operation and maintenance graph is calculated, and several entity types with high similarity are identified as candidate entities. Knowledge expansion is performed based on candidate entities, and candidate relations and candidate text units are constructed. The candidate relations are all edges with candidate entities as source nodes or target nodes; the candidate text units are structured text blocks containing at least one candidate entity. Under the limited constraints of the LLM context window, candidate relations and candidate text units are sorted and filtered. Candidate entities, sorted and filtered candidate relationships, and candidate text units are concatenated into a structured context, which is then submitted to the LLM along with the user's question to generate causal inference results.

[0025] In some embodiments, see Figure 2 The step of acquiring wind turbine operation and maintenance text data, and then segmenting the operation and maintenance text data into blocks to obtain structured text blocks, includes: Acquire wind turbine operation and maintenance text data, and set the basic block length and overlapping sampling length based on the block parameter preset strategy of length constraint and semantic integrity, according to the text length distribution, model input upper limit and operation and maintenance semantic integrity. The step size is determined based on the basic block length and the overlapping sampling length. The basic text blocks are obtained by using the continuous sampling method of overlapping sliding windows. An adaptive window fine-tuning method based on semantic boundary detection is used to perform boundary detection on each basic text block, resulting in a semantically complete structured text block.

[0026] In some embodiments, see Figure 3 The call relationship extraction agent, combined with the entity types and relationship types preset in the FMEA structure table, constructs a wind turbine fault operation and maintenance graph framework. Based on the wind turbine fault operation and maintenance graph framework and structured text blocks, the wind turbine fault operation and maintenance graph is determined, specifically including the following steps: Obtain wind turbine operation and maintenance text data, and process the operation and maintenance text data into blocks to obtain structured text blocks; A framework for constructing a wind turbine fault operation and maintenance graph is built by combining the entity types and relationship types preset in the FMEA structure table; System prompt word templates are generated based on the wind turbine fault operation and maintenance graph structure framework; based on the system prompt word templates, the entity type and relation type of the structured text block are extracted using a large language model to obtain the structured extraction results, and the structure extraction results are converted into data structures that the graph can write to form candidate knowledge units; Candidate knowledge units are standardized into graph triples. A similarity-based entity ablation method is used to align and merge the entity types of the graph triples and the wind turbine fault operation and maintenance graph framework to obtain entity vectors. The similarity between entity vectors is calculated based on cosine distance. A similarity threshold is used to filter highly similar pairs. If the similarity between entities is not less than a preset value, disambiguation and merging are performed on the entity vectors to obtain a consistent fault operation and maintenance map.

[0027] In some embodiments, see Figure 4 The process of obtaining the user's query command, retrieving evidence from the fault operation and maintenance map and structured text blocks based on the query command, and generating a causal chain inference result containing fault symptoms, potential causes, impact consequences, and maintenance suggestions includes the following steps: Obtain the user's query command and vectorize it to obtain the query vector. The semantic similarity between the query vector and the entity vectors in the fault operation and maintenance graph is calculated, and several entity types with high similarity are identified as candidate entities. Knowledge expansion is performed based on candidate entities, and candidate relations and candidate text units are constructed. The candidate relations are all edges with candidate entities as source nodes or target nodes; the candidate text units are structured text blocks containing at least one candidate entity. Under the limited constraints of the LLM context window, candidate relations and candidate text units are sorted and filtered. Candidate entities, sorted and filtered candidate relationships, and candidate text units are concatenated into a structured context, which is then submitted to the LLM along with the user's question to generate causal inference results.

[0028] Preferably, the sorting rules for sorting and filtering candidate relations and candidate text units under the constraint of a limited LLM context window are as follows: Candidate relationships are divided into intra-network relationships and extra-network relationships. Within a network, relationships are sorted in descending order by combinatorial degree, which is the sum of the degree of the source node and the degree of the target node; outside a network, relationships are sorted in descending order by the number of network entities connected to the relationship, with relationships within a network taking precedence over relationships outside a network. For candidate text units, sort them in descending order according to the highest semantic similarity among the candidate entities associated with the text unit. If the highest similarity is the same, sort them in descending order according to the number of relations associated with the text unit.

[0029] In some embodiments, the reasonableness and relevance of the causal chain inference results are reviewed based on preset review rules. If the review passes, the causal chain inference results are output; if not, the process returns to re-searching for evidence in the fault operation and maintenance graph and structured text blocks until the maximum exit mechanism is met, as expressed by the following formula:

[0030] in, This indicates the final output of the fault causal chain retrieval results that meet the review requirements; This represents the candidate result of the fault causal chain generated by the causal reasoning agent in round t; This represents the candidate results of the fault causal chain generated in the previous round; This represents the evidence validity review function, used to verify the validity of candidate results based on rules and experience; Represents a set of physical rules; This represents a set of expert experience. This represents the causal chain reconstruction function, used to correct results that fail the review and generate new candidate results; Indicates the current iteration round; This indicates the maximum number of iterations.

[0031] Example 1 A collaborative multi-agent and fault map-based causal reasoning method for wind turbine faults includes: S1: Obtain wind turbine operation and maintenance text data, and perform structural parsing and block processing on the wind turbine operation and maintenance text data to obtain structured text blocks carrying title levels and source identifiers; S101: Adopting a block parameter preset strategy based on length constraints and semantic integrity, the basic block length is set according to the text length distribution, the upper limit of model input, and the semantic integrity of operation and maintenance. With overlap sampling length To ensure that critical operational events are not truncated and that complete event descriptions are covered, the following constraints apply:

[0032] in, This sets the maximum length limit for the model input.

[0033] S102: Employs a continuous sampling strategy based on overlapping sliding windows with a fixed step size. Starting from the beginning of the text, a sliding window is used to ensure that each new block overlaps with the previous block by a specified length, achieving continuous coverage sampling and avoiding semantic breaks in operation and maintenance. The i-th text block... It can be represented as:

[0034] in, And satisfy , This represents the total length of the text.

[0035] S103: An adaptive window fine-tuning strategy based on semantic boundary detection is adopted to perform boundary detection on each piece of text to determine whether it is truncated in the middle of semantic units such as key terms, fault descriptions, and maintenance actions. If the window is truncated within a semantic unit, then boundary fine-tuning is performed to adjust the window's ending position. satisfy:

[0036] in, For the allowed fine-tuning window, this mechanism allows for fine-tuning of truncated blocks located within semantic unit boundaries (such as punctuation, line breaks, keyword endings, etc.), ensuring semantic integrity within the block and reducing semantic breaks caused by block segmentation.

[0037] S2: Based on the relation extraction agent, entities and relations are extracted from the structured text block. The entity types and relation types preset in the FMEA structure table are combined to construct the wind turbine fault operation and maintenance map, and entity alignment and consistency verification are completed. S201: Construct a wind turbine fault operation and maintenance graph structure framework by combining the entity types and relationship types preset in the FMEA structure table; generate system prompt word templates based on the wind turbine fault operation and maintenance graph structure framework; the relationship extraction agent performs wind turbine fault operation and maintenance entity and relationship extraction on each text block obtained above based on the system prompt word template, outputs structured extraction results according to the entity tags and relationship tags, and converts the extraction results into a graph-writable data structure to form candidate knowledge units; S202: Standardize the candidate knowledge unit into a graph triplet form, wherein the triplet includes at least "head entity - relation - tail entity", denoted as... , of which Entity set For each set of relations, a source identifier, header level, and text block position index are appended to each triple to enable source tracing; S203: A similarity-based entity ablation method is used to align and merge entities, and text embedding is performed on the entity name and its contextual description to obtain a vector representation. And calculate the similarity between entities based on cosine distance:

[0038] S204: Use a similarity threshold to filter highly similar pairs, if For entities and Perform disambiguation and merging to obtain a consistent fault operation and maintenance map.

[0039] More preferably, a prompt word template for the relation extraction intelligent agent system is constructed, in which entity tags and relation tags are defined. The entity includes at least structural component functions, failure modes, failure risks, maintenance methods and their corresponding attribute fields. The relation includes at least "structural component - has function", "function - corresponding failure mode", "failure mode - associated failure risk", "failure risk - corresponding maintenance method", etc., to ensure that the fault map business framework is designed based on the wind turbine operation and maintenance FMEA structure table.

[0040] S3: Receive complex queries input by the user, call the preset causal reasoning retrieval agent 4 to retrieve evidence in the fault operation and maintenance map and structured text blocks, and generate causal chain reasoning results containing fault symptoms, potential causes, impact consequences and maintenance suggestions. S301: Represent the user query as a vector, calculate the semantic similarity between the query vector and the description vectors of each entity in the fault operation and maintenance map, and find the N most similar entities as candidate entities; let the user query be... Entities in the knowledge base The description vector is Then the query and entity semantic similarity The calculation is as follows:

[0041] in, For user queries Vector representation of; S302: Construct a candidate entity set based on similarity, and expand knowledge around the candidate entities to build candidate relations and candidate text units, wherein the candidate relations are all edges with candidate entities as source nodes or target nodes, and their definitions are as follows;

[0042] in, Let r be the set of candidate relations. Represents the set of all relations in the graph. Represents the source entity of relation r. Represents the target entity of relation r. The candidate entity set is expressed as follows:

[0043] in, This indicates that the Nth candidate entity with the highest similarity is selected; A candidate text unit is a structured text block containing at least one candidate entity, defined as follows:

[0044] in, For the set of candidate text units, This represents the set of entities that appear in text block t. For structured text blocks.

[0045] S303: Under the constraint of a limited LLM context window, candidate relations and candidate text units are sorted and filtered, prioritizing the retention of key evidence related to causal chain construction; the specific sorting rules are as follows: First, relationships are divided into intra-network relationships and extra-network relationships: Intra-network relationships refer to the relationships connecting two candidate entities, that is... and External relationships refer to the relationships connecting a candidate entity and a non-candidate entity, i.e. or }

[0046] Then, the relationships are sorted. Within the network, the relationships are sorted in descending order by combinatorial degree, which is the sum of the source node degree and the target node degree. The expression is as follows:

[0047] in, Let be the degree of entity e in the graph; In external relationships, they are sorted in descending order by the number of internal entities connected to the relationship (i.e., the number of non-candidate entities connected to the candidate entity), and internal relationships take precedence over external relationships. To sort text units, first sort them in descending order based on the highest semantic similarity among the candidate entities associated with the text unit; if the highest similarity is the same, then sort them in descending order based on the number of relations associated with the text unit.

[0048] S304: Concatenate the entity description with the sorted and filtered candidate relationships and candidate text units to form a structured context, and submit it along with the user's question to the LLM to generate causal inference results. The causal inference results include at least fault symptoms, potential causes, impact consequences, and suggested maintenance methods, and simultaneously output graph evidence and text block source information corresponding to the conclusions. S4: Call the central routing decision agent to schedule the calling order and iterative review of the multi-agent system, review the rationality of the inference results, and restart the retrieval process when the review fails until the exit mechanism is met.

[0049] Evidence rationality review mechanism under conditional iteration: Based on physical rules and expert experience, the rationality and relevance of the search answer are reviewed. If the review fails, the result is returned to the causal retrieval agent for repeated retrieval and reasoning until the maximum exit mechanism is met. The review output satisfies the following expression:

[0050] in, This indicates the final output of the fault causal chain retrieval results that meet the review requirements; This represents the candidate result of the fault causal chain generated by the causal reasoning agent in round t; This represents the candidate results of the fault causal chain generated in the previous round; This represents the evidence validity review function, used to verify the validity of candidate results based on rules and experience; Represents a set of physical rules; This represents a set of expert experience. This represents the causal chain reconstruction function, used to correct results that fail the review and generate new candidate results; Indicates the current iteration round; Indicates the maximum number of iterations; Furthermore, the approved answers are comprehensively integrated, the logical hierarchy is sorted out, overlapping content is eliminated, and related explanations are added to form a standardized, complete, and directly referable standardized wind turbine fault repair solution; Finally, the solution and evidence tracing information are returned to the user to complete the routing decision and collaboration workflow.

[0051] Example 2 like Figure 6 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a multi-agent-enabled wind turbine fault cause-effect graph autonomous reasoning system, comprising: The central routing decision-making agent is used to obtain user query instructions, analyze user query instructions to determine the problem type and complexity, generate route scheduling based on the problem type and complexity, call causal reasoning agents and evidence review agents based on the route scheduling, and generate wind turbine fault operation and maintenance map by calling relationship extraction agents. A relation extraction agent is used to acquire wind turbine operation and maintenance text data, and to process the operation and maintenance text data into blocks to obtain structured text blocks. A wind turbine fault operation and maintenance graph framework is constructed by combining the entity types and relation types preset in the FMEA structure table. The wind turbine fault operation and maintenance graph is determined based on the wind turbine fault operation and maintenance graph framework and the structured text blocks. The causal reasoning agent is used to retrieve evidence according to user query commands, generate causal chain reasoning results including fault symptoms, potential causes, impact consequences and maintenance suggestions, and feed the causal chain reasoning results back to the central routing decision agent. The evidence review agent is used to review the rationality and relevance of the causal chain reasoning results based on preset review rules. If it passes, it outputs the causal chain reasoning results; if it fails, it feeds back to the central routing decision agent, which then calls the causal reasoning agent to retrieve evidence again until the preset maximum exit mechanism is met.

[0052] In some embodiments, see Figure 5 A causal reasoning system for wind turbine failures based on collaborative multi-agent systems and fault maps is described. The evidence review agent reviews the rationality and relevance of the causal chain reasoning results based on preset review rules. If the review passes, the causal chain reasoning result is output; otherwise, the system returns to call the causal reasoning agent again for evidence retrieval until a preset maximum exit mechanism is met. This is expressed by the following formula:

[0053] in, This indicates the final output of the fault causal chain retrieval results that meet the review requirements; This represents the candidate result of the fault causal chain generated by the causal reasoning agent in round t; This represents the candidate results of the fault causal chain generated in the previous round; This represents the evidence validity review function, used to verify the validity of candidate results based on rules and experience; Represents a set of physical rules; This represents a set of expert experience. This represents the causal chain reconstruction function, used to correct results that fail the review and generate new candidate results; Indicates the current iteration round; This indicates the maximum number of iterations.

[0054] In some embodiments, the central routing decision agent is used to supervise the collaborative process of the wind turbine fault causal reasoning system with the collaborative multi-agent and fault map, execute the routing scheduling logic process, and uniformly define the input and output fields, calling order, status identifier and exception handling rules of each agent.

[0055] See Figure 5 The system constructs a prompt word template for the central routing decision-making intelligent agent system, clearly defining the core functions of the central routing decision-making intelligent agent as supervising the entire multi-agent collaborative process and executing the routing scheduling logic process. It also uniformly defines the input and output fields, calling order, status identifiers and exception handling rules of each intelligent agent, providing a standardized basis for subsequent routing decisions and process control. Furthermore, through the user input acquisition and route scheduling steps, user input is first acquired, and a routing strategy is generated based on the question type and complexity. The specific routing strategy generation expression is as follows:

[0056] Where R is the routing policy generated by the routing agent. Indicates the routing function; This is a function to determine the question type; Q represents the user's question. Furthermore, through routing scheduling and workflow orchestration steps, the query is first routed to the causal reasoning agent to obtain candidate entities, candidate relationships and candidate text units, and form candidate answers; Furthermore, candidate answers are scheduled to the evidence review agent through routing rules to review the rationality and reliability of the answers and obtain fault operation and maintenance solutions under structural standardization. Finally, the final operation and maintenance plan and fault cause tracing information obtained are returned to the user, completing the routing decision and collaborative workflow and autonomous fault operation and maintenance handling.

[0057] Example 3 like Figure 7 As shown, the present invention also provides an electronic device 100 for implementing a wind turbine fault causal reasoning method based on a collaborative multi-agent and fault map. The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0058] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of a collaborative multi-agent and fault map-based wind turbine fault causal reasoning method by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0059] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0060] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0061] The memory 101 in the electronic device 100 stores multiple instructions to implement a causal reasoning method for wind turbine faults based on collaborative multi-agent and fault map data. The processor 102 can execute multiple instructions to achieve the following: Obtain wind turbine operation and maintenance text data, and process the operation and maintenance text data into blocks to obtain structured text blocks; A wind turbine fault operation and maintenance graph framework is constructed by combining the entity types and relationship types preset in the FMEA structure table. The entity types and relationship types of the structured text blocks are extracted based on the wind turbine fault operation and maintenance graph framework. The wind turbine fault operation and maintenance graph is determined according to the entity types and relationship types of the structured text blocks and the wind turbine fault operation and maintenance graph framework. Obtain the user's query command, retrieve evidence from the fault operation and maintenance map and structured text blocks according to the query command, and generate causal chain reasoning results containing fault symptoms, potential causes, impact consequences and maintenance suggestions; The causal chain reasoning results are reviewed for reasonableness and relevance based on preset review rules. If the review is successful, the causal chain reasoning results are output. If the review is unsuccessful, the system returns to search for evidence in the fault operation and maintenance graph and structured text blocks until the preset maximum exit mechanism is met.

[0062] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0063] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0065] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0067] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A multi-agent empowered fan fault causal graph autonomous reasoning method, characterized in that, include: The intelligent agent is invoked to obtain wind turbine operation and maintenance text data, and the operation and maintenance text data is segmented to obtain structured text blocks; A wind turbine fault operation and maintenance map framework is constructed by combining the entity types and relationship types preset in the FMEA structure table. The wind turbine fault operation and maintenance map is determined based on the wind turbine fault operation and maintenance map framework and structured text blocks. Obtain user query commands, analyze user query commands to determine the question type and complexity, and generate route scheduling based on question type and complexity; The causal reasoning agent is invoked according to the routing schedule. The causal reasoning agent performs evidence retrieval based on the user's query command and generates causal chain reasoning results that include fault symptoms, potential causes, impact consequences and maintenance suggestions. The evidence review agent is invoked to review the rationality and relevance of the causal chain reasoning results based on preset review rules. If the review is successful, the causal chain reasoning results are output. If the review fails, the process returns to invoke the causal reasoning agent again to retrieve evidence until the preset maximum exit mechanism is met.

2. The multi-agent empowered fan fault causal graph autonomous reasoning method according to claim 1, characterized in that, The process of acquiring wind turbine operation and maintenance text data involves segmenting the text data into blocks to obtain structured text blocks, including: Acquire wind turbine operation and maintenance text data, and set the basic block length and overlapping sampling length based on the block parameter preset strategy of length constraint and semantic integrity, according to the text length distribution, model input upper limit and operation and maintenance semantic integrity. The step size is determined based on the basic block length and the overlapping sampling length. The basic text blocks are obtained by using the continuous sampling method of overlapping sliding windows. An adaptive window fine-tuning method based on semantic boundary detection is used to perform boundary detection on each basic text block, resulting in a semantically complete structured text block.

3. The multi-agent empowered fan fault causal graph autonomous reasoning method according to claim 1, characterized in that, The call relationship extraction agent, combined with the entity types and relationship types preset in the FMEA structure table, constructs a wind turbine fault operation and maintenance graph framework. Based on the wind turbine fault operation and maintenance graph framework and structured text blocks, the wind turbine fault operation and maintenance graph is determined, specifically including the following steps: A framework for constructing a wind turbine fault operation and maintenance graph is built by combining the entity types and relationship types preset in the FMEA structure table; System prompt word templates are generated based on the wind turbine fault operation and maintenance graph structure framework; based on the system prompt word templates, the entity type and relation type of the structured text block are extracted using a large language model to obtain the structured extraction results, and the structure extraction results are converted into data structures that the graph can write to form candidate knowledge units; Candidate knowledge units are standardized into graph triples. A similarity-based entity ablation method is used to align and merge the entity types of the graph triples and the wind turbine fault operation and maintenance graph framework to obtain entity vectors. The similarity between entity vectors is calculated based on cosine distance. A similarity threshold is used to filter highly similar pairs. If the similarity between entities is not less than a preset value, disambiguation and merging are performed on the entity vectors to obtain a consistent fault operation and maintenance map.

4. The multi-agent empowered fan fault causal graph autonomous reasoning method according to claim 1, characterized in that, The causal reasoning agent retrieves evidence based on the user's query command and generates a causal chain reasoning result that includes fault symptoms, potential causes, impact consequences, and maintenance suggestions, specifically including: Obtain the user's query command and vectorize it to obtain the query vector. The semantic similarity between the query vector and the entity vectors in the fault operation and maintenance graph is calculated, and several entity types with high similarity are identified as candidate entities. Knowledge expansion is performed based on candidate entities, and candidate relations and candidate text units are constructed. The candidate relations are all edges with candidate entities as source nodes or target nodes; the candidate text units are structured text blocks containing at least one candidate entity. Under the limited constraints of the LLM context window, candidate relations and candidate text units are sorted and filtered. Candidate entities, sorted and filtered candidate relationships, and candidate text units are concatenated into a structured context, which is then submitted to the LLM along with the user's question to generate causal inference results.

5. The multi-agent enabled fan fault causal graph autonomous reasoning method of claim 4, wherein, The sorting rules for ranking and filtering candidate relations and candidate text units under the limited constraints of the LLM context window are as follows: Candidate relationships are divided into intra-network relationships and extra-network relationships; Within a network, relationships are sorted in descending order by combinatorial degree, which is the sum of the degree of the source node and the degree of the target node; outside a network, relationships are sorted in descending order by the number of network entities connected to the relationship, with relationships within a network taking precedence over relationships outside a network. For candidate text units, sort them in descending order according to the highest semantic similarity among the candidate entities associated with the text unit. If the highest similarity is the same, sort them in descending order according to the number of relations associated with the text unit.

6. The multi-agent-enabled wind turbine fault cause-effect graph autonomous reasoning method according to claim 1, characterized in that, The reasonableness and relevance of the causal chain reasoning results are reviewed based on preset review rules. If the review passes, the causal chain reasoning results are output; otherwise, the process returns to call the causal reasoning agent again to retrieve evidence until the preset maximum exit mechanism is met. This is expressed by the following formula: in, This indicates the final output of the fault causal chain retrieval results that meet the review requirements; This represents the candidate result of the fault causal chain generated by the causal reasoning agent in round t; This represents the candidate results of the fault causal chain generated in the previous round; This represents the evidence validity review function, used to verify the validity of candidate results based on rules and experience; Represents a set of physical rules; This represents a set of expert experience. This represents the causal chain reconstruction function, used to correct results that fail the review and generate new candidate results; Indicates the current iteration round; This indicates the maximum number of iterations.

7. A multi-agent-enabled wind turbine fault cause-effect graph autonomous reasoning system, characterized in that, include: The central routing decision-making agent is used to obtain user query instructions, analyze user query instructions to determine the problem type and complexity, and generate route scheduling based on the problem type and complexity. The causal reasoning agent and the evidence review agent are invoked according to the routing schedule; And extract the call relationship to generate a wind turbine fault operation and maintenance map; A relation extraction agent is used to acquire wind turbine operation and maintenance text data, and to segment the operation and maintenance text data into blocks to obtain structured text blocks. A wind turbine fault operation and maintenance map framework is constructed by combining the entity types and relationship types preset in the FMEA structure table. The wind turbine fault operation and maintenance map is determined based on the wind turbine fault operation and maintenance map framework and structured text blocks. The causal reasoning agent is used to retrieve evidence according to user query commands, generate causal chain reasoning results including fault symptoms, potential causes, impact consequences and maintenance suggestions, and feed the causal chain reasoning results back to the central routing decision agent. The evidence review agent is used to review the rationality and relevance of the causal chain reasoning results based on preset review rules. If it passes, it outputs the causal chain reasoning results; if it fails, it feeds back to the central routing decision agent, which then calls the causal reasoning agent to retrieve evidence again until the preset maximum exit mechanism is met.

8. A multi-agent empowered fan fault causal graph atlas autonomous reasoning system, characterized in that, The evidence review agent performs a rationality and relevance review on the causal chain reasoning results based on preset review rules. If the review is successful, the causal chain reasoning results are output; otherwise, the process returns to call the causal reasoning agent again to retrieve evidence, until the preset maximum exit mechanism is met, as expressed by the following formula: in, This indicates the final output of the fault causal chain retrieval results that meet the review requirements; This represents the candidate result of the fault causal chain generated by the causal reasoning agent in round t; This represents the candidate results of the fault causal chain generated in the previous round; This represents the evidence validity review function, used to verify the validity of candidate results based on rules and experience; Represents a set of physical rules; This represents a set of expert experience. This represents the causal chain reconstruction function, used to correct results that fail the review and generate new candidate results; Indicates the current iteration round; This indicates the maximum number of iterations.

9. An electronic device, comprising: It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement a multi-agent-enabled wind turbine fault causal graph autonomous reasoning method as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements a multi-agent-enabled wind turbine fault causal graph autonomous reasoning method as described in any one of claims 1 to 6.

Images