Question answering method, device and equipment based on knowledge graph in wind power field and medium

Abstract

The invention discloses a question answering method and device based on a knowledge graph in the wind power field, equipment and a medium, and relates to the technical field of machine learning, and the method comprises the steps: receiving input operation and maintenance question information, and carrying out the semantic analysis, and obtaining an operation and maintenance object and a question intention; retrieving corresponding knowledge information in a preset wind power field knowledge graph based on the operation and maintenance object and the question intention, and combining with the operation and maintenance question information to input into a pre-trained large language model; and outputting corresponding operation and maintenance answering information according to the knowledge information and the operation and maintenance question information through a large language model. According to the method, scattered and heterogeneous wind power field knowledge is integrated, the illusion risk caused by lack of field knowledge of a large language model is effectively inhibited, and the accuracy and reliability of operation and maintenance answering information are improved, so that excessive dependence on expert experience is reduced, the knowledge acquisition process of operation and maintenance personnel is simplified, the field operation and maintenance efficiency is remarkably improved, and the operation and maintenance efficiency is improved. And downtime and cost caused by response delay are reduced.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to a question-answering method, apparatus, equipment and medium based on a knowledge graph in the field of wind power. Background Technology

[0002] In the operation and maintenance of wind farms, maintenance personnel need to handle complex knowledge covering multiple fields such as mechanics, electrical systems, and hydraulics. Currently, this knowledge exists mostly in the form of technical manuals, fault reports, and expert experience, forming scattered and heterogeneous knowledge silos. This situation often leads to two problems: firstly, maintenance personnel find it difficult to quickly and accurately locate the information needed to solve specific faults from massive amounts of data, as traditional retrieval methods are inefficient and prone to missing key information; secondly, on-site problems heavily rely on a few experts for remote or on-site diagnosis, resulting in long response delays and high costs.

[0003] While there are existing auxiliary tools based on general search engines or document management systems, they lack an understanding of the semantics of the wind power field, cannot effectively connect interdisciplinary knowledge, and have limited intelligence. In recent years, large language models have demonstrated powerful natural language processing capabilities, but in the wind power field, the content they generate carries the risk of factual illusion and is difficult to directly apply to wind farm operation and maintenance decisions where accuracy is extremely important. Summary of the Invention

[0004] The main purpose of this application is to provide a question-answering method, device, equipment and medium based on knowledge graphs in the wind power field, which aims to solve the technical problem of excessive illusion risk when using large language models in the current wind farm operation and maintenance process.

[0005] To achieve the above objectives, this application proposes a question-answering method based on a knowledge graph in the wind power field. The question-answering method based on the knowledge graph in the wind power field includes: Receive input operation and maintenance questions, and perform semantic parsing on the operation and maintenance questions to obtain the operation and maintenance objects and the questioning intent in the operation and maintenance questions; Based on the operation and maintenance object and the questioning intent, the corresponding knowledge information is retrieved from the preset wind power field knowledge graph, and the knowledge information is combined with the operation and maintenance question information and input into the pre-trained large language model. Based on the knowledge information and the operation and maintenance question information, the large language model outputs operation and maintenance answer information corresponding to the operation and maintenance question information.

[0006] In one embodiment, the preset wind power domain knowledge graph records entity nodes, attribute nodes, and the relationship paths between the entity nodes. The step of retrieving corresponding knowledge information from the preset wind power domain knowledge graph based on the operation and maintenance object and the question intent includes: The large language model is used to determine at least one entity corresponding to the operation and maintenance object, as well as the query direction corresponding to the question intent; Based on the entity and the query direction, a graph traversal is performed in the preset wind power domain knowledge graph to retrieve the target entity node, target attribute node and target relationship path; The target entity node, the target attribute node, and the target relationship path are used as the knowledge information.

[0007] In one embodiment, the question-answering method based on a knowledge graph in the wind power field further includes: Extract entities, attributes, and relationships between entities from multiple wind power data sources; The entities are aligned and merged to obtain structured data, wherein the structured data includes relation triples and / or attribute pairs. The relation triples record a first entity, a second entity, and the relationship between the first entity and the second entity. The attribute pairs record a third entity, the attributes of the third entity, and the values ​​of the attributes of the third entity. The structured data is stored in a graph database to construct the preset wind power knowledge graph.

[0008] In one embodiment, the step of combining the knowledge information with the operation and maintenance question information and inputting it into a pre-trained large language model includes: The knowledge information is converted into contextual description information in natural language format; Based on the contextual description information and the operation and maintenance question information, prompt words are generated and input into a pre-trained large language model.

[0009] In one embodiment, the step of outputting operation and maintenance answer information corresponding to the operation and maintenance question information based on the knowledge information and the operation and maintenance question information using the large language model includes: Based on the knowledge information, the large language model generates the first candidate answer information corresponding to the operation and maintenance question; The large language model generates a second candidate answer to the operation and maintenance question based on its internal knowledge information, wherein the internal knowledge information is determined by the model parameters obtained through training of the large language model. The first candidate answer information and the second candidate answer information are compared, and the operation and maintenance answer information corresponding to the operation and maintenance question information is output according to the comparison result.

[0010] In one embodiment, the step of comparing the first candidate answer information and the second candidate answer information, and outputting the operation and maintenance answer information corresponding to the operation and maintenance question information based on the comparison result, includes: Based on the target entity node, target attribute node and target relationship path in the knowledge information, verify the consistency of the statements on the same entity, the same attribute and / or the same relationship in the second candidate solution information; If the consistency verification passes, the first candidate solution information and the second candidate solution information are integrated to generate and output the operation and maintenance solution information corresponding to the operation and maintenance question information. If the consistency verification fails, the second candidate answer information is corrected based on the knowledge information, and the first candidate answer information and the corrected second candidate answer information are integrated to generate and output the operation and maintenance answer information corresponding to the operation and maintenance question information.

[0011] In one embodiment, the step of correcting the second candidate solution information based on the knowledge information includes: Based on the target entity node, target attribute node, and / or target relationship path corresponding to the entity, attribute, and / or relationship that failed the consistency verification in the knowledge information, a corrected text fragment is generated. The entities, attributes, and / or relationships that failed the consistency verification in the second candidate solution information are replaced with the corrected text fragments to obtain the corrected second candidate solution information.

[0012] Furthermore, to achieve the above objectives, this application also proposes a question-answering device based on a knowledge graph in the wind power field, wherein the question-answering device based on the knowledge graph in the wind power field includes: The parsing module is used to receive input operation and maintenance questions and perform semantic parsing on the operation and maintenance questions to obtain the operation and maintenance objects and the questioning intent in the operation and maintenance questions. The retrieval module is used to retrieve corresponding knowledge information from a preset wind power knowledge graph based on the operation and maintenance object and the question intent, and to combine the knowledge information with the operation and maintenance question information and input it into a pre-trained large language model. The output module is used to output operation and maintenance answer information corresponding to the operation and maintenance question information based on the knowledge information and the operation and maintenance question information through the large language model.

[0013] In addition, to achieve the above objectives, this application also proposes an electronic device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the question-answering method based on a knowledge graph in the wind power field as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the question-answering method based on the knowledge graph in the wind power field as described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: This application integrates dispersed and heterogeneous wind power knowledge by introducing a pre-defined knowledge graph of the wind power field as a structured knowledge source. This enables operation and maintenance personnel to quickly and accurately retrieve relevant information based on the operation and maintenance object and the intent of the question, directly solving the problems of low efficiency and easy omission of key content in traditional retrieval methods. Simultaneously, the retrieved knowledge information is combined with a large language model, using the knowledge information as factual basis to guide the large language model in generating answers. This effectively suppresses the risk of illusion caused by the large language model's lack of domain knowledge, improves the accuracy and reliability of operation and maintenance answers, and reduces over-reliance on expert experience. Finally, the operation and maintenance answers are output through natural language interaction, simplifying the knowledge acquisition process for operation and maintenance personnel, significantly improving on-site operation and maintenance efficiency, and reducing downtime and costs caused by response delays. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the question-answering method based on a knowledge graph in the wind power field presented in this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the question-answering method based on a knowledge graph in the wind power field presented in this application. Figure 3 A simplified flowchart illustrating the question-answering method based on a knowledge graph in the wind power field provided in Embodiment 2 of this application; Figure 4 This is a schematic diagram of the module structure of the question-answering method device based on a knowledge graph in the wind power field according to an embodiment of this application; Figure 5 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the question-and-answer method based on the knowledge graph in the wind power field in the embodiments of this application.

[0019] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] The main solution of this application embodiment is as follows: receiving input operation and maintenance (O&M) questions, performing semantic parsing on the O&M questions to obtain the O&M object and question intent in the O&M questions; retrieving corresponding knowledge information from a preset wind power domain knowledge graph based on the O&M object and the question intent, and combining the knowledge information with the O&M questions and inputting it into a pre-trained large language model; and outputting O&M answer information corresponding to the O&M questions based on the knowledge information and the O&M questions through the large language model.

[0023] While existing auxiliary tools based on general search engines or document management systems exist, they lack an understanding of the semantics of the wind power field, cannot effectively connect interdisciplinary knowledge, and have limited intelligence. In recent years, large language models have demonstrated powerful natural language processing capabilities, but in the wind power field, the content they generate carries the risk of factual illusion and is difficult to directly apply to wind farm operation and maintenance decisions where accuracy is extremely important.

[0024] This application provides a solution that integrates dispersed and heterogeneous wind power knowledge by introducing a pre-defined knowledge graph of the wind power field as a structured knowledge source. This enables operation and maintenance personnel to quickly and accurately retrieve relevant information based on the operation and maintenance object and the intent of the question, directly solving the problems of low efficiency and easy omission of key content in traditional retrieval methods. Simultaneously, the retrieved knowledge information is combined with a large language model, using the knowledge information as factual basis to guide the large language model in generating answers. This effectively suppresses the risk of illusion caused by the large language model's lack of domain knowledge, improves the accuracy and reliability of operation and maintenance answers, and reduces over-reliance on expert experience. Finally, the operation and maintenance answers are output through natural language interaction, simplifying the knowledge acquisition process for operation and maintenance personnel, significantly improving on-site operation and maintenance efficiency, and reducing downtime and costs caused by response delays.

[0025] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or question-and-answer system capable of performing the above functions. The following description uses a question-and-answer system as an example to illustrate this embodiment and the subsequent embodiments.

[0026] Based on this, embodiments of this application provide a question-answering method based on a knowledge graph in the wind power field, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the question-answering method based on a knowledge graph in the wind power field according to this application.

[0027] In this embodiment, the question-answering method based on a knowledge graph in the wind power field includes steps S10 to S30: Step S10: Receive the input operation and maintenance question information, and perform semantic parsing on the operation and maintenance question information to obtain the operation and maintenance object and the question intent in the operation and maintenance question information; It should be noted that the operation and maintenance question information is the text content that wind farm operation and maintenance personnel input in natural language in the interactive interface to request specific operation and maintenance knowledge or solve specific operation and maintenance problems. This information is the original input and trigger signal for the entire question and answer process.

[0028] Among them, the operation and maintenance object refers to the target thing in the wind power system that the system identifies after semantic understanding and analysis of the operation and maintenance question information. It is usually an entity that exists in the knowledge graph, such as specific equipment, components, faults, parameters or operation items. The question intent refers to the user's core demand and question type determined by the system through semantic parsing of the operation and maintenance question information. It defines the logical direction of information retrieval and answer organization, such as whether to query the cause of the fault, obtain the solution, confirm the operation steps, or inquire about the status information.

[0029] Step S20: Based on the operation and maintenance object and the questioning intent, retrieve the corresponding knowledge information in the preset wind power field knowledge graph, and combine the knowledge information with the operation and maintenance question information and input it into the pre-trained large language model; It should be noted that the pre-built wind power knowledge graph refers to a structured data set that is pre-constructed and stored in a graph database, representing the professional knowledge system of the wind power field in a network structure. Nodes represent entities or attributes, and edges represent various semantic relationships or attribute associations between entities, forming the data foundation for the system's intelligent retrieval. Knowledge information, on the other hand, refers to one or more relevant structured knowledge subgraphs or their textual representations extracted by the system after performing graph traversal and retrieval operations on the pre-built wind power knowledge graph based on the identified operation and maintenance object and the intent of the question. This information serves as the accurate factual basis for generating the answer.

[0030] The pre-trained large language model is a basic artificial intelligence model that has been trained on a large-scale general corpus and has the ability to understand natural language and generate text. In this embodiment, the model is configured to receive prompts that incorporate knowledge information and generate natural language text that conforms to the facts of the domain.

[0031] Step S30: Based on the knowledge information and the operation and maintenance question information, the large language model outputs operation and maintenance answer information corresponding to the operation and maintenance question information.

[0032] It should be noted that the operation and maintenance answer information is a structured or unstructured text response that is directly addressed to the user's operation and maintenance questions after the pre-trained large language model integrates the input knowledge information and the original operation and maintenance questions, and is presented to the user to complete this knowledge question and answer interaction.

[0033] Understandably, existing wind farm operation and maintenance suffers from problems such as knowledge fragmentation, retrieval difficulties, and the potential for factual illusions when directly applying large language models. Therefore, this embodiment first performs semantic parsing on the operation and maintenance questions in natural language form to extract the operation and maintenance objects and the questioning intent. Then, based on the parsing results, it searches a specially constructed pre-built wind power domain knowledge graph to obtain accurate structured knowledge information. Subsequently, this knowledge information and the operation and maintenance questions are input into a pre-trained large language model. Finally, the large language model merges the two contents to generate operation and maintenance answers. This effectively avoids the serious problems of low efficiency, omission of key information, and unreliable content generated by large language models due to the lack of domain factual constraints in traditional retrieval methods. It achieves effective integration and accurate retrieval of fragmented and heterogeneous wind power domain knowledge, significantly improving the accuracy and reliability of the question-and-answer system's answers. At the same time, natural language interaction greatly reduces the threshold and time cost for operation and maintenance personnel to acquire professional knowledge, thereby effectively reducing reliance on expert experience and improving the overall efficiency of on-site operation and maintenance.

[0034] For example, maintenance personnel input maintenance questions about wind turbine malfunctions through a natural language interface on a terminal device. The question-and-answer system running on the terminal device processes the maintenance question using an integrated semantic parsing module, identifying key maintenance objects (such as "gearbox oil temperature too high") and the question's intent (such as "inquire about possible causes"). The system then connects to a pre-built and stored wind power domain knowledge graph in a graph database, performing a graph query based on the identified maintenance objects and question intent to retrieve entities, attributes, and relationship paths associated with "gearbox," "oil temperature too high," and "fault cause" as knowledge information. The system assembles the retrieved knowledge information and the original maintenance question into a structured prompt, which is then input into a pre-trained large language model. This large language model uses the knowledge information in the prompt as factual constraints, combined with its natural language understanding and generation capabilities, to generate an accurate and readable maintenance answer to the question, which is then presented to the maintenance personnel through the interactive interface.

[0035] This embodiment provides a question-answering method based on a knowledge graph in the wind power field. By introducing a pre-set knowledge graph in the wind power field as a structured knowledge source, it integrates scattered and heterogeneous knowledge in the wind power field, enabling operation and maintenance personnel to quickly and accurately retrieve relevant information based on the operation and maintenance object and the intent of the question. This directly solves the problems of low efficiency and easy omission of key content in traditional retrieval methods. At the same time, the retrieved knowledge information is combined with a large language model. The knowledge information is used as factual basis to guide the large language model to generate answers, effectively suppressing the illusion risk caused by the large language model's lack of domain knowledge, improving the accuracy and reliability of operation and maintenance answers, and thus reducing the over-reliance on expert experience. Finally, the operation and maintenance answers are output through natural language interaction, simplifying the process of operation and maintenance personnel acquiring knowledge, significantly improving on-site operation and maintenance efficiency, and reducing downtime and costs caused by response delays.

[0036] In one feasible implementation, the preset wind power knowledge graph records entity nodes, attribute nodes, and the relationship paths between the entity nodes. Step S20, which involves retrieving corresponding knowledge information from the preset wind power knowledge graph based on the operation and maintenance object and the question intent, may include steps S21-S23: Step S21: Determine at least one entity corresponding to the operation and maintenance object and the query direction corresponding to the question intent through the large language model; It should be noted that an entity is a basic unit that represents a specific thing, concept, or element in the wind power field, obtained by semantic parsing and mapping the operation and maintenance objects in the operation and maintenance query information and placing them in the preset wind power field knowledge graph. As the core component of the graph node, it is used to represent specific field elements such as equipment, components, fault types, or technical terms.

[0037] The query direction is based on the analysis and classification of the question intent in the operation and maintenance questions. It is the logical orientation or semantic goal for information retrieval and exploration in the preset wind power knowledge graph. The graph traversal process is guided to focus on specific types of knowledge associations, such as causal inference, structural composition, attribute query or solution matching.

[0038] Step S22: Based on the entity and the query direction, perform graph traversal in the preset wind power domain knowledge graph to retrieve the target entity node, target attribute node, and target relationship path; It should be noted that the target entity node is a graph node in the preset wind power knowledge graph that is located through graph traversal based on the identified entities and the query direction. This node is the graph node corresponding to one or more entities that are directly related to the current operation and maintenance question. These nodes constitute the core fact carrier required for answer generation.

[0039] Target attribute nodes are graph nodes that are further retrieved during graph traversal based on the query direction. They are used to describe or modify the specific characteristics, state, or values ​​of target entity nodes and provide supplementary information on entity-related detailed parameters or conditional constraints.

[0040] The target relationship path is a sequence of relation edges retrieved through graph traversal in the pre-defined wind power domain knowledge graph. These edges connect target entity nodes and / or target attribute nodes and express specific semantic relationships between them, forming a structured knowledge chain that reveals the logical connections between entities and attributes, such as membership, causality, or interaction.

[0041] Step S23: The target entity node, the target attribute node, and the target relationship path are used as the knowledge information.

[0042] Understandably, relying solely on semantic parsing results for graph retrieval may not accurately and completely locate all structured knowledge related to complex operation and maintenance questions. Therefore, this implementation method first uses a large language model to deeply understand the parsed operation and maintenance objects and question intent to determine at least one corresponding entity and a clear query direction. Then, based on this entity and query direction, a directed graph traversal is performed in a preset wind power domain knowledge graph to retrieve directly related target entity nodes, target attribute nodes describing their characteristics, and target relationship paths connecting them. This complete knowledge structure is used as knowledge information, thereby avoiding the problems of inaccurate retrieval results and isolated and fragmented knowledge fragments caused by simple keyword matching or shallow parsing. It achieves a systematic acquisition of deep-level, multi-related knowledge in the knowledge graph, providing logically rigorous and informationally complete structured factual evidence for subsequent answer generation, thus significantly improving the depth, accuracy, and reliability of the final operation and maintenance answer information.

[0043] For example, after the system receives the maintenance question and completes semantic parsing to obtain the maintenance object and the question intent, it first uses a pre-trained large language model to perform deep semantic understanding and refinement on the parsed maintenance object and question intent. This determines at least one specific entity corresponding to the maintenance object in a pre-defined wind power domain knowledge graph and clarifies the specific query direction corresponding to the question intent. The system then initiates a graph traversal retrieval process in the pre-defined wind power domain knowledge graph based on the determined entity and query direction. Using a graph query language or algorithm, it performs multi-hop retrieval along relevant relation edges, ultimately retrieving target entity nodes directly or indirectly associated with the initial entity, target attribute nodes describing the state or parameters of these target entity nodes, and target relation paths connecting these nodes and expressing their semantic associations. The system integrates these retrieved target entity nodes, target attribute nodes, and target relation paths into structured knowledge information, and then inputs it along with the original maintenance question information into the large language model to generate the final maintenance answer information.

[0044] In this implementation, a large language model is first used to deeply understand the maintenance object and the question intent obtained after semantic parsing, in order to determine the specific entity corresponding to the maintenance object in the knowledge graph and the query direction pointed to by the question intent. Then, a directed graph traversal is performed in the preset wind power domain knowledge graph based on the entity and the query direction, thereby retrieving the target entity node, target attribute node and target relationship path closely related to the maintenance question as knowledge information. This avoids the problems of one-sided, isolated or insufficiently related search results caused by the lack of fine analysis of user intent and in-depth exploration of the structural relationship of the knowledge graph in traditional retrieval methods. It realizes the efficient and accurate extraction of logically coherent and informationally complete structured knowledge fragments from the knowledge graph, laying the foundation for the subsequent generation of accurate and reliable maintenance answer information, and significantly improving the intelligence and practicality of the question answering system.

[0045] In one feasible implementation, the step of combining the knowledge information with the operation and maintenance question information and inputting it into the pre-trained large language model in step S20 may include steps S24-S25: Step S24: Convert the knowledge information into contextual description information in natural language format; It should be noted that the contextual description information is a coherent text fragment that conforms to the syntax and semantics of natural language, which is formed by reorganizing and expressing the retrieved knowledge information through set transformation rules or templates. This description information is intended to serve as background knowledge context that can be understood and processed by the large language model, and together with the original operation and maintenance question information, it constitutes a complete prompt to guide the model to generate accurate answers.

[0046] Step S25: Based on the context description information and the operation and maintenance question information, construct prompt words and input the prompt words into the pre-trained large language model.

[0047] Understandably, since knowledge information retrieved directly from knowledge graphs is often difficult for pre-trained large language models to understand and utilize effectively, this implementation further converts the knowledge information into contextual description information that conforms to the syntax and semantics of natural language. Then, based on this contextual description information and the original operation and maintenance question information, a structured prompt word is formed and input into the large language model. This avoids the problems of logical confusion, insufficient information integration, or factual bias in the generated answer caused by the model's inability to directly parse and associate discrete data. It achieves the effective injection of precise domain knowledge in the model's optimized natural language form, significantly improving the depth of understanding and accuracy of application of professional knowledge by the large language model, thereby ensuring that the final generated operation and maintenance answer information has both factual correctness and linguistic fluency.

[0048] For example, after the system retrieves structured knowledge information based on a pre-defined wind power domain knowledge graph, a dedicated natural language generation module organizes the entities, attributes, and relationships within this knowledge information into coherent and fluent natural language paragraphs according to predefined conversion rules and templates, thereby generating contextual description information. Subsequently, the system concatenates and formats this contextual description information with the received original operation and maintenance (O&M) questions to form a structured prompt. Finally, the system uses this prompt as input and submits it to a pre-trained large language model for processing, guiding the model to generate corresponding O&M answers based on the provided precise domain knowledge context.

[0049] In one feasible implementation, step S30 may include steps S31 to S33: Step S31: Based on the knowledge information, the large language model generates the first candidate answer information corresponding to the operation and maintenance question information; It should be noted that the first candidate answer information refers to the preliminary answer text generated by the pre-trained large language model based on the knowledge information retrieved and input from the preset wind power domain knowledge graph, after understanding and processing, for the operation and maintenance questions.

[0050] Furthermore, during the generation of the first candidate solution information, the large language model does not utilize internal knowledge information to ensure that the first candidate solution information is generated solely based on knowledge information derived from the knowledge graph of the wind power field, thereby ensuring the accuracy of the first candidate solution information.

[0051] Step S32: Generate a second candidate answer information corresponding to the operation and maintenance question information based on the internal knowledge information of the large language model, wherein the internal knowledge information is determined by the model parameters obtained by the large language model through training; It should be noted that the internal knowledge information refers to the general knowledge, language patterns, and reasoning abilities that the pre-trained large language model learns from a large-scale general corpus and internalizes into its model parameters during the training phase. This knowledge is independent of the specific domain knowledge information of the current input, that is, the knowledge information retrieved from the pre-set wind power domain knowledge graph. The model parameters are all the internal weights and values ​​used to store the learned language rules and knowledge, adjusted and determined by the pre-trained large language model through optimization algorithms during training. These parameters determine the specific behavior of the model in processing input and generating output.

[0052] The second candidate answer information refers to another preliminary answer text generated by the pre-trained large language model based on its internal knowledge information, that is, not constrained by the input knowledge information, but based on the general knowledge contained in its own parameters, for the same operation and maintenance question information.

[0053] Step S33: Compare the first candidate answer information and the second candidate answer information, and output the operation and maintenance answer information corresponding to the operation and maintenance question information according to the comparison result.

[0054] It should be noted that the comparison result is a conclusion drawn by the system after comparing and analyzing the content of the first candidate answer and the second candidate answer, regarding the degree of difference between the two in terms of factual statements, logical consistency, or completeness.

[0055] Understandably, relying solely on external knowledge retrieved from the knowledge graph to input into a large language model for answer generation carries the risk that the model may overlook some input information, have knowledge biases in its inherent parameters, or have incomplete knowledge content covered by the knowledge graph when dealing with complex problems. On the other hand, relying solely on the model's internal knowledge information to generate answers carries the inherent risk of illusion. Therefore, this implementation first generates a first candidate answer based on the input knowledge information using a large language model, and simultaneously generates a second candidate answer based solely on the internal knowledge information contained in the model parameters obtained from its internal training using the same model. The two are then compared to avoid the problem of one-sided or inaccurate answers caused by insufficient utilization of input knowledge by the model or factual errors in its internal knowledge. This approach achieves cross-validation and complementary fusion of results generated from internal and external knowledge sources, thereby further enhancing the robustness, completeness, and factual accuracy of the final operational answer information.

[0056] For example, after receiving maintenance (O&M) questions and performing semantic parsing to obtain the O&M object and the question intent, the system retrieves corresponding knowledge information from a pre-defined wind power knowledge graph based on the O&M object and the question intent. Then, the knowledge information and the O&M question information are input into a pre-trained large language model. This large language model first generates a first candidate answer to the O&M question based on the input knowledge information, and simultaneously generates a second candidate answer to the same O&M question based on its internal knowledge information corresponding to the model parameters obtained through training. The system then compares the content of the first and second candidate answers, obtains the comparison result, and makes a decision based on this result, outputting the final O&M answer.

[0057] In the specific implementation process, step S33 may also include steps S331 to S333: Step S331: Based on the target entity node, target attribute node and target relationship path in the knowledge information, verify the consistency of the statements on the same entity, the same attribute and / or the same relationship in the second candidate answer information; It should be noted that consistency means that when comparing the statements about specific entities, attributes or relationships in the wind power field in the second candidate solution information with the objectively represented factual content of the target entity nodes, target attribute nodes and target relationship paths in the knowledge information retrieved from the preset wind power field knowledge graph and used as the benchmark, the consistency should be accurate and consistent in terms of factual assertions, logical relationships and numerical attributes.

[0058] For example, the entities, attributes, and / or relationships in the second candidate answer information are mapped and aligned with the target entity nodes, target attribute nodes, and / or target relationship paths in the knowledge information to determine the standard expression in the knowledge information, wherein the standard expression includes target entity nodes, target attribute nodes, and / or target relationship paths; the text similarity between the corresponding statement in the second candidate answer information and the standard expression is calculated, wherein the statement includes entities, attributes, and / or relationships; if the text similarity is greater than or equal to a preset first threshold, the consistency verification is deemed to have passed; if the text similarity is lower than the first threshold but higher than a preset second threshold, the pre-trained large language model is used to perform semantic consistency analysis on the statement and the standard expression; if the semantic consistency analysis result of the large language model is semantically consistent, the consistency verification is deemed to have passed; if the semantic consistency analysis result of the large language model is semantically inconsistent, the consistency verification is deemed to have failed, wherein the condition for the semantic consistency analysis result to be semantically consistent is that each statement in the second candidate answer information after mapping and alignment is determined to be semantically consistent. This example employs a dual-filtering verification mechanism. First, it performs a coarse screening using rapid text similarity, and then initiates a complex but precise large-scale model semantic analysis for cases with ambiguous boundaries. This verification mechanism balances efficiency and accuracy, effectively solving the technical problem of different expressions but the same semantics.

[0059] Step S332: If the consistency verification passes, the first candidate solution information and the second candidate solution information are integrated to generate and output the operation and maintenance solution information corresponding to the operation and maintenance question information. Step S333: If the consistency verification fails, the second candidate answer information is corrected based on the knowledge information, and the first candidate answer information and the corrected second candidate answer information are integrated to generate and output the operation and maintenance answer information corresponding to the operation and maintenance question information.

[0060] Understandably, since simple information comparison may not be able to effectively identify and correct specific factual errors or contradictions in the second candidate solution information that contradict external knowledge sources, this implementation process further verifies the consistency of statements about the same entity, attribute, or relationship in the second candidate solution information by using the target entity node, target attribute node, and target relationship path in the knowledge information as a benchmark. Based on the verification results, information integration or the generation of corrective text fragments based on knowledge information is performed to specifically correct the second candidate solution information. This effectively avoids the problem of factual contradictions or reduced credibility in the generated operation and maintenance solution information caused by the inherent bias or error of the knowledge information within the large language model directly contaminating the final output. On the basis of cross-validation, this further ensures and strengthens the strict factual alignment between the final output answer and the structured domain knowledge graph, thereby making the operation and maintenance solution information more accurate.

[0061] For example, if the maintenance question is: "What are the possible causes of abnormal noise in the gearbox?", the knowledge information retrieved from the knowledge graph includes the target entity node "gearbox". The second candidate answer might incorrectly reference the entity, for example, stating: "Wearing generator bearings may cause abnormal noise". In this case, the system performs consistency verification based on the target entity node "gearbox" in the knowledge information, finding that the "generator" stated in the second candidate answer is inconsistent with the benchmark entity "gearbox". The verification fails, and the system generates a corrected text fragment based on the "gearbox" node in the knowledge information, for example: "Wearing bearings or gears inside the gearbox may cause abnormal noise", and replaces the original incorrect statement with this fragment. The corrected second candidate answer is then integrated with the first candidate answer and output.

[0062] For example, if the maintenance question is: "Is the current gearbox oil temperature normal?", the knowledge information retrieved from the knowledge graph includes the target entity node "gearbox" and its target attribute node "oil temperature: 85°C" (representing the current real-time data). A second candidate answer might be based on outdated internal knowledge stating: "Gearbox oil temperature is usually below 70°C, and the current temperature should be normal." The system performs consistency verification based on the target entity node "gearbox" and the target attribute node "oil temperature: 85°C" in the knowledge information. It finds that the statement about the attribute value "below 70°C" in the second candidate answer is inconsistent with the benchmark attribute value "85°C". The verification fails, and the system generates a corrected text fragment based on the knowledge information, such as: "The current real-time oil temperature is 85°C, higher than the normal range," and corrects the second candidate answer accordingly. Finally, it integrates the corrected text with the first candidate answer and outputs the final answer containing the accurate value.

[0063] Furthermore, the step of correcting the second candidate solution information based on the knowledge information may also include steps S301-S302: Step S301: Generate a corrected text fragment based on the target entity node, target attribute node, and / or target relationship path corresponding to the entity, attribute, and / or relationship that failed the consistency verification in the knowledge information; It should be noted that the corrected text fragment is a correctly described text that conforms to natural language standards, generated by the system based on accurate factual information contained in the target entity node, target attribute node, and / or target relationship path retrieved from the preset wind power domain knowledge graph as an authoritative benchmark when the statement of a specific entity, attribute, or relationship in the second candidate solution information fails the consistency verification. This text fragment is used to directly replace the corresponding part of the second candidate solution information that contains factual errors or contradictions, so as to ensure that its content is fully aligned with the structured domain knowledge.

[0064] Step S302: Replace the entities, attributes and / or relationships that failed the consistency verification in the second candidate solution information with the corrected text fragments to obtain the corrected second candidate solution information.

[0065] Understandably, since inappropriate correction methods may lead to an uncontrollable correction process or introduce new ambiguities, by first generating a corrected text fragment that accurately describes the facts based on the target nodes and paths directly corresponding to the entities, attributes, and / or relationships that failed verification in the knowledge information, and then precisely replacing the corresponding erroneous content in the second candidate answer information with this text fragment, we can avoid the problem that the correction process may rely on vague semantic adjustments or general rewriting, which could lead to deviations from authoritative knowledge or damage to the overall coherence of the original answer. This achieves accurate, efficient, and traceable local correction of inconsistent content, ensuring that the final integrated operation and maintenance answer information maintains good linguistic integrity and logical consistency while strictly conforming to the facts of the domain knowledge graph.

[0066] For example, if the second candidate answer states, "...the inverter's cooling system should be checked," but consistency verification reveals that the correct target entity node retrieved from the knowledge graph that is directly related to the maintenance question is "gearbox" rather than "inverter," then the system will generate a corrected text fragment "gearbox" based on the facts (such as identifier and type) carried by the "target entity node" in the knowledge information, i.e., the "gearbox" node. This fragment will then directly replace the incorrect entity "inverter" in the second candidate answer, thereby correcting the original sentence to "...the gearbox's cooling system should be checked."

[0067] For example, if the second candidate solution states: "...the gearbox pressure sensor reading is abnormal...", but consistency verification reveals that the "target relationship path" connected to "gearbox" in the knowledge graph indicates that it contains "temperature sensor" rather than "pressure sensor", then the system will simultaneously generate a composite corrected text fragment "gearbox temperature sensor" based on the "target entity node" (gearbox) and the related "target relationship path" (such as "containment" relationship), and use this fragment to replace the erroneous statement "gearbox pressure sensor" in the second candidate solution, thereby correcting the original sentence to "...the gearbox temperature sensor reading is abnormal...".

[0068] In this embodiment, the same pre-trained large language model generates first candidate answer information based on the input knowledge information and second candidate answer information based on the internal knowledge information corresponding to the model parameters obtained from its training. The two are then compared in content. This avoids the problem that the final answer generated may be one-sided, inaccurate, or unreliable because the large language model may not fully comply with the input external knowledge or its internal parameters may have domain fact bias. It achieves cross-validation and complementarity between external knowledge sources and internal knowledge sources of the model, thereby significantly enhancing the comprehensive accuracy, factual consistency, and decision robustness of the operation and maintenance answer information.

[0069] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The question-answering method based on the knowledge graph of the wind power field further includes steps S01 to S03: Step S01: Extract entities, attributes, and relationships between entities from multiple wind power data sources; It should be noted that data sources in the wind power field are various original materials or data sets that provide professional knowledge and information in the wind power field. These include, but are not limited to, structured, semi-structured, or unstructured documents and data in various forms such as technical manuals, maintenance procedures, fault records, expert experience texts, and equipment lists.

[0070] Step S02: Align and merge the entities to obtain structured data, wherein the structured data includes relation triples and / or attribute pairs. The relation triples record a first entity, a second entity, and the relationship between the first entity and the second entity. The attribute pairs record a third entity, the attributes of the third entity, and the values ​​of the attributes of the third entity. It should be noted that structured data is a data expression form with clear semantics and a unified format formed by standardizing and organizing the raw information extracted from data sources in the wind power field. This data enables knowledge to be efficiently stored, retrieved, and reasoned by computer systems.

[0071] A relation triple is a form of structured data representation, consisting of two related entities and a relation term that precisely describes the specific semantic relationship between them. It is used to characterize various connections between things in the wind power field. An attribute pair is another form of structured data representation, consisting of an entity, an attribute item describing a specific characteristic of that entity, and the specific attribute value corresponding to that attribute item. It is used to characterize the inherent properties or states of entities in the wind power field. The first and second entities together constitute the basic elements of a relation triple, representing two specific wind power field things or concepts connected by a certain semantic relationship. The third entity is the basic element of an attribute pair, specifically referring to the specific wind power field thing or concept described and modified by the attribute pair. It is understood that the first and second entities are different entities, while the third entity can be the same entity as the first entity, the same entity as the second entity, or a different entity from the first and second entities. This embodiment does not make specific limitations in this regard.

[0072] Step S03: Store the structured data in a graph database to construct the preset wind power knowledge graph.

[0073] It should be noted that a graph database is a database system specifically designed for storing, managing, and querying graph structure data (i.e., a data model composed of nodes and edges). It provides underlying technical support for the persistent storage and efficient traversal retrieval of entity, attribute, and relation data in the knowledge graph of the wind power industry.

[0074] Understandably, since the pre-defined wind power knowledge graph itself requires a specific and implementable construction method to ensure the accuracy of its knowledge sources, the standardization of its structure, and the usability of its content, this embodiment first extracts entities, attributes, and relationships from multiple heterogeneous wind power data sources, then aligns and merges the extraction results to eliminate ambiguity and conflicts, forming unified relation triples and attribute pairs of equivalent structured data. Finally, the structured data is stored in a graph database to construct the knowledge graph. This avoids the problems of chaotic, contradictory, or incomplete knowledge graph content that may result from directly using raw data that has not been systematically integrated. It realizes the reliable construction of a high-quality, structured knowledge graph from scattered, multi-source domain knowledge, providing a solid and scalable knowledge foundation for subsequent graph-based accurate retrieval and intelligent question answering.

[0075] For example, the system first collects raw information from multiple wind power data sources, and uses natural language processing technology to perform entity recognition, relation extraction, and attribute extraction to obtain an initial set of entities, attributes, and relations. Then, knowledge fusion technology is used to align and merge entities from different data sources that point to the same real-world entity, eliminating naming ambiguities and description conflicts, forming unified and standardized relation triples (e.g., <gearbox, contains, temperature sensor>) and attribute pairs (e.g., <gearbox, rated oil temperature, 80°C>). Finally, the system imports this structured data in batches into a graph database for storage and indexing, thereby constructing a pre-defined wind power knowledge graph with a complete structure that can be efficiently queried and traversed.

[0076] In this embodiment, entities, attributes, and relationships are extracted from multiple heterogeneous wind power data sources. These extracted results are then aligned and fused to form unified and standardized relation triples and attribute pairs, which are then stored in a graph database to construct a knowledge graph. This approach avoids the problems of knowledge fragmentation, semantic conflicts, and low information retrieval efficiency that may occur when directly using raw, scattered, and heterogeneous data to build a knowledge base. It systematically integrates multi-source, interdisciplinary domain knowledge into a high-quality, structured, and unified knowledge system, thereby providing a reliable, complete, and scalable core knowledge foundation for subsequent precise semantic retrieval and intelligent question answering based on the knowledge graph. This effectively supports the accuracy and practicality of the entire question answering system.

[0077] For example, to help understand the implementation process of the question-answering method based on a knowledge graph in the wind power field obtained by combining this embodiment with the above embodiment one, please refer to... Figure 3 , Figure 3 A simplified flowchart of a question-answering method based on a knowledge graph in the wind power field is provided, specifically: First, the system extracts and integrates knowledge from multiple heterogeneous wind power data sources to construct a structured wind power knowledge graph. When a user inputs an operation and maintenance (O&M) question, the system retrieves relevant knowledge information from the knowledge graph based on this information. This retrieved knowledge information, along with the original O&M question, is then input into a large language model. Finally, the large language model comprehensively processes this information to generate and output accurate and reliable O&M answers, thus completing a full intelligent question-and-answer interaction.

[0078] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the question-and-answer method based on the knowledge graph in the wind power field. Any simple modifications based on this technical concept are within the scope of protection of this application.

[0079] This application also provides a question-answering device based on a knowledge graph in the wind power field. Please refer to [link / reference]. Figure 4The question-answering device based on a knowledge graph in the wind power field includes: The parsing module 10 is used to receive input operation and maintenance question information and perform semantic parsing on the operation and maintenance question information to obtain the operation and maintenance object and the question intent in the operation and maintenance question information; The retrieval module 20 is used to retrieve corresponding knowledge information in a preset wind power field knowledge graph based on the operation and maintenance object and the question intent, and to combine the knowledge information with the operation and maintenance question information and input it into a pre-trained large language model. The output module 30 is used to output maintenance answer information corresponding to the maintenance question information based on the knowledge information and the maintenance question information through the large language model.

[0080] Optionally, the preset wind power knowledge graph records entity nodes, attribute nodes, and the relationship paths between the entity nodes. The retrieval module 20 is further used for: The large language model is used to determine at least one entity corresponding to the operation and maintenance object, as well as the query direction corresponding to the question intent; Based on the entity and the query direction, a graph traversal is performed in the preset wind power domain knowledge graph to retrieve the target entity node, target attribute node and target relationship path; The target entity node, the target attribute node, and the target relationship path are used as the knowledge information.

[0081] Optionally, the question-answering device based on a knowledge graph in the wind power field further includes a construction module 40, which is used for: Extract entities, attributes, and relationships between entities from multiple wind power data sources; The entities are aligned and merged to obtain structured data, wherein the structured data includes relation triples and / or attribute pairs. The relation triples record a first entity, a second entity, and the relationship between the first entity and the second entity. The attribute pairs record a third entity, the attributes of the third entity, and the values ​​of the attributes of the third entity. The structured data is stored in a graph database to construct the preset wind power knowledge graph.

[0082] Optionally, the retrieval module 20 is further configured to: The knowledge information is converted into contextual description information in natural language format; Based on the contextual description information and the operation and maintenance question information, prompt words are generated and input into a pre-trained large language model.

[0083] Optionally, the output module 30 is further configured to: Based on the knowledge information, the large language model generates the first candidate answer information corresponding to the operation and maintenance question; The large language model generates a second candidate answer to the operation and maintenance question based on its internal knowledge information, wherein the internal knowledge information is determined by the model parameters obtained through training of the large language model. The first candidate answer information and the second candidate answer information are compared, and the operation and maintenance answer information corresponding to the operation and maintenance question information is output according to the comparison result.

[0084] Optionally, the output module 30 is further configured to: Based on the target entity node, target attribute node and target relationship path in the knowledge information, verify the consistency of the statements on the same entity, the same attribute and / or the same relationship in the second candidate solution information; If the consistency verification passes, the first candidate solution information and the second candidate solution information are integrated to generate and output the operation and maintenance solution information corresponding to the operation and maintenance question information. If the consistency verification fails, the second candidate answer information is corrected based on the knowledge information, and the first candidate answer information and the corrected second candidate answer information are integrated to generate and output the operation and maintenance answer information corresponding to the operation and maintenance question information.

[0085] Optionally, the output 30 is further used for: Based on the target entity node, target attribute node, and / or target relationship path corresponding to the entity, attribute, and / or relationship that failed the consistency verification in the knowledge information, a corrected text fragment is generated. The entities, attributes, and / or relationships that failed the consistency verification in the second candidate solution information are replaced with the corrected text fragments to obtain the corrected second candidate solution information.

[0086] The question-answering device based on a knowledge graph in the wind power field provided in this application employs the question-answering method based on a knowledge graph in the wind power field described in the above embodiments, which can solve the technical problem of excessively high illusion risks when using large language models in the current wind farm operation and maintenance process. Compared with the prior art, the beneficial effects of the question-answering device based on a knowledge graph in the wind power field provided in this application are the same as the beneficial effects of the question-answering method based on a knowledge graph in the wind power field provided in the above embodiments, and other technical features in the question-answering device based on a knowledge graph in the wind power field are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0087] This application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the question-answering method based on the knowledge graph in the wind power field described in Embodiment 1 above.

[0088] The following is for reference. Figure 5 The diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of this application. The electronic devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0089] like Figure 5 As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. The communication device 1009 allows the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although the diagrams show electronic devices with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented alternatively.

[0090] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0091] The electronic device provided in this application employs the question-answering method based on a knowledge graph in the wind power field as described in the above embodiments, which can solve the technical problem of excessively high illusion risks when using large language models in the current wind farm operation and maintenance process. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the question-answering method based on a knowledge graph in the wind power field provided in the above embodiments, and other technical features of this electronic device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0092] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0093] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0094] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the question-and-answer method based on a knowledge graph in the wind power field in the above embodiments.

[0095] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0096] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0097] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by an electronic device, the electronic device: receives input maintenance and operation (O&M) questions and performs semantic parsing on the O&M questions to obtain the O&M object and the questioning intent in the O&M questions; retrieves corresponding knowledge information in a preset wind power domain knowledge graph based on the O&M object and the questioning intent, and combines the knowledge information with the O&M questions and inputs it into a pre-trained large language model; and outputs O&M answer information corresponding to the O&M questions based on the knowledge information and the O&M questions through the large language model.

[0098] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0099] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0100] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0101] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-described question-answering method based on a knowledge graph in the wind power field. This addresses the technical problem of excessively high illusion risks associated with using large language models in wind farm operation and maintenance. Compared to existing technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the question-answering method based on a knowledge graph in the wind power field provided in the above embodiments, and will not be elaborated upon here.

[0102] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A question-answering method based on a knowledge graph in the wind power field, characterized in that, The question-answering method based on the knowledge graph of the wind power field includes: Receive input operation and maintenance questions, and perform semantic parsing on the operation and maintenance questions to obtain the operation and maintenance objects and the questioning intent in the operation and maintenance questions; Based on the operation and maintenance object and the questioning intent, the corresponding knowledge information is retrieved from the preset wind power field knowledge graph, and the knowledge information is combined with the operation and maintenance question information and input into the pre-trained large language model. Based on the knowledge information and the operation and maintenance question information, the large language model outputs operation and maintenance answer information corresponding to the operation and maintenance question information.

2. The question-answering method based on a knowledge graph in the wind power field as described in claim 1, characterized in that, The preset wind power knowledge graph contains entity nodes, attribute nodes, and the relationship paths between the entity nodes. The step of retrieving corresponding knowledge information from the preset wind power knowledge graph based on the operation and maintenance object and the question intent includes: The large language model is used to determine at least one entity corresponding to the operation and maintenance object, as well as the query direction corresponding to the question intent; Based on the entity and the query direction, a graph traversal is performed in the preset wind power domain knowledge graph to retrieve the target entity node, target attribute node and target relationship path; The target entity node, the target attribute node, and the target relationship path are used as the knowledge information.

3. The question-answering method based on a knowledge graph in the wind power field as described in claim 2, characterized in that, The question-answering method based on the knowledge graph of the wind power field also includes: Extract entities, attributes, and relationships between entities from multiple wind power data sources; The entities are aligned and merged to obtain structured data, wherein the structured data includes relation triples and / or attribute pairs. The relation triples record a first entity, a second entity, and the relationship between the first entity and the second entity. The attribute pairs record a third entity, the attributes of the third entity, and the values ​​of the attributes of the third entity. The structured data is stored in a graph database to construct the preset wind power knowledge graph.

4. The question-answering method based on a knowledge graph in the wind power field as described in claim 1, characterized in that, The step of combining the knowledge information with the operation and maintenance question information and inputting it into the pre-trained large language model includes: The knowledge information is converted into contextual description information in natural language format; Based on the contextual description information and the operation and maintenance question information, prompt words are generated and input into a pre-trained large language model.

5. The question-answering method based on a knowledge graph in the wind power field as described in claim 1, characterized in that, The step of outputting operation and maintenance answer information corresponding to the operation and maintenance question information based on the knowledge information and the operation and maintenance question information through the large language model includes: Based on the knowledge information, the large language model generates the first candidate answer information corresponding to the operation and maintenance question; The large language model generates a second candidate answer to the operation and maintenance question based on its internal knowledge information, wherein the internal knowledge information is determined by the model parameters obtained through training of the large language model. The first candidate answer information and the second candidate answer information are compared, and the operation and maintenance answer information corresponding to the operation and maintenance question information is output according to the comparison result.

6. The question-answering method based on a knowledge graph in the wind power field as described in claim 5, characterized in that, The step of comparing the first candidate answer information and the second candidate answer information, and outputting the operation and maintenance answer information corresponding to the operation and maintenance question information according to the comparison result includes: Based on the target entity node, target attribute node and target relationship path in the knowledge information, verify the consistency of the statements on the same entity, the same attribute and / or the same relationship in the second candidate solution information; If the consistency verification passes, the first candidate solution information and the second candidate solution information are integrated to generate and output the operation and maintenance solution information corresponding to the operation and maintenance question information. If the consistency verification fails, the second candidate answer information is corrected based on the knowledge information, and the first candidate answer information and the corrected second candidate answer information are integrated to generate and output the operation and maintenance answer information corresponding to the operation and maintenance question information.

7. The question-answering method based on a knowledge graph in the wind power field as described in claim 6, characterized in that, The step of correcting the second candidate solution information based on the knowledge information includes: Based on the target entity node, target attribute node, and / or target relationship path corresponding to the entity, attribute, and / or relationship that failed the consistency verification in the knowledge information, a corrected text fragment is generated. The entities, attributes, and / or relationships that failed the consistency verification in the second candidate solution information are replaced with the corrected text fragments to obtain the corrected second candidate solution information.

8. A question-answering device based on a knowledge graph in the wind power field, characterized in that, The question-answering device based on a knowledge graph in the wind power field includes: The parsing module is used to receive input operation and maintenance questions and perform semantic parsing on the operation and maintenance questions to obtain the operation and maintenance objects and the questioning intent in the operation and maintenance questions. The retrieval module is used to retrieve corresponding knowledge information from a preset wind power knowledge graph based on the operation and maintenance object and the question intent, and to combine the knowledge information with the operation and maintenance question information and input it into a pre-trained large language model. The output module is used to output operation and maintenance answer information corresponding to the operation and maintenance question information based on the knowledge information and the operation and maintenance question information through the large language model.

9. An electronic device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the question-answering method based on a knowledge graph in the wind power field as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the question-answering method based on the knowledge graph in the wind power field as described in any one of claims 1 to 7.

Images