Generative model construction method for operation and maintenance of power equipment and related device
By building a generative model based on power equipment operation and inspection, combining LORA fine-tuning module and question-and-answer map, the model adaptability and reliability problems are solved, and efficient power equipment operation and inspection analysis and decision-making support are achieved.
Patent Information
- Application Number
- CN202510470206.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-15
- Publication Date
- 2025-08-01
AI Technical Summary
The existing generative models have poor adaptability and low decision-making reliability in power equipment operation and inspection, making it difficult to adapt to the diversity and complexity in different scenarios.
Combining the generative model and the LORA fine-tuning module, a question-and-answer map is constructed by mining the operation and inspection records of power equipment and standard knowledge, fine-tuning instruction clustering and supervision training are carried out, low-rank matrix is introduced, LORA fine-tuning module is formed and the initial generative model is integrated to achieve scene-based fine-tuning.
It improves the adaptability and decision-making reliability of the generative model, can conduct more accurate operation inspection analysis and prediction of power equipment, and improves the intelligence level and efficiency of power operation and maintenance management.
Smart Images
Figure CN120410495A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power equipment operation and maintenance management, and in particular relates to a generative model construction method for power equipment operation and maintenance and related devices. Background Art
[0002] With the rapid development of power systems, the safe and stable operation of power equipment has become crucial. As a key step in ensuring the proper functioning of power equipment, the efficiency and accuracy of power equipment maintenance directly impact the overall performance and reliability of the power system. Generative models, by learning from large amounts of power equipment maintenance data, can generate models with predictive and diagnostic capabilities, providing intelligent support for power equipment maintenance.
[0003] Currently, existing generative model-based power equipment maintenance methods typically collect historical maintenance data from power equipment and train generative models to predict the equipment's operating status and potential faults. However, power equipment maintenance data is diverse and complex, with significant differences between different devices and scenarios. Existing generative models often struggle to adapt to this diversity and complexity, resulting in reduced predictive and diagnostic accuracy across different scenarios. In summary, given the current technical challenges of poor adaptability and low decision reliability in generative models for power equipment maintenance, there is an urgent need to develop a new, efficient generative model adaptation solution for power equipment maintenance. Summary of the Invention
[0004] The purpose of the present invention is to provide a method and related device for constructing a generative model for power equipment operation and maintenance, so as to solve one or more of the above-mentioned technical problems. In the technical solution disclosed in the present invention, a generative model and a LORA fine-tuning module are combined to perform power operation and maintenance analysis. The LORA fine-tuning module can be used to perform scenario-based fine-tuning constraints on the initial generative model based on power operation and maintenance scenario data to achieve efficient adaptation. This solves the technical problems of poor model adaptability and low decision reliability existing in existing generative models for power equipment operation and maintenance, and achieves the technical effect of improving the model's adaptability and decision reliability.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for constructing a generative model for power equipment operation and maintenance, comprising the following steps: Recall the operation and inspection records of power equipment and integrate them with power standard knowledge to mine and obtain question-answer graphs and fine-tuning instruction sets based on power knowledge; Based on the question-answer graph, an initial generative model is trained using a supervised training method; Cluster the obtained fine-tuning instruction set and set multiple fine-tuning types; based on the set multiple fine-tuning types, introduce an initial low-rank matrix and train a LORA fine-tuning module in a supervised training manner; wherein, the LORA fine-tuning module includes multiple parallel fine-tuning low-rank matrices; Integrate the LORA fine-tuning module into the initial generative model to obtain a final composite generative model; wherein, the composite generative model is used to receive power operation and maintenance scenario data, and the LORA fine-tuning module performs scenario-based fine-tuning on the initial generative model according to the power operation and maintenance scenario data to obtain a fine-tuned generative model, and the fine-tuned generative model performs power operation and maintenance analysis according to the power operation and maintenance scenario data to obtain a power operation and maintenance result.
[0006] A further improvement of the present invention lies in that In the step of calling the power equipment operation and maintenance records and integrating power standard knowledge to mine the Q&A graph and fine-tuning instruction set based on power knowledge, when mining the Q&A graph based on power knowledge, the following processing is performed: Traverse the power equipment operation and maintenance records, perform task homologous clustering, mine the entities and relationships of power operation and maintenance, and construct an operation and maintenance knowledge graph; Based on the Q&A logic, transform the operation and maintenance knowledge graph to obtain an initial Q&A graph; wherein, the Q&A logic is determined based on the characteristics of the operation and maintenance tasks; Traverse the power standard knowledge and compensate the initial Q&A graph to obtain a Q&A graph based on power knowledge.
[0007] A further improvement of the present invention lies in that In the step of calling the power equipment operation and maintenance records and integrating power standard knowledge to mine the Q&A graph and fine-tuning instruction set based on power knowledge, when mining the fine-tuning instruction set, the following processing is performed: Traverse the power equipment operation and maintenance records, and determine the operation and maintenance task characteristics and the tendency characteristics of the operation and maintenance results of each record data; Combine the operation and maintenance task characteristics and the tendency characteristics to generate the fine-tuning instruction set.
[0008] A further improvement of the present invention lies in that The step of introducing an initial low-rank matrix based on the set multiple fine-tuning types and training a LORA fine-tuning module in a supervised training manner specifically includes: Traverse all fine-tuning types to complete the training of the fine-tuning low-rank matrix corresponding to each fine-tuning type; combine all the fine-tuning low-rank matrices in parallel to obtain a LORA fine-tuning module; Among them, when traversing all fine-tuning types, for the nFine-tuning type, based on the instruction cluster corresponding to the fine-tuning type, obtain fine-tuning samples; based on the fine-tuning samples, perform supervised training on the initialized low-rank matrix to determine the n fine-tuned low-rank matrix; The LORA fine-tuning module is provided with multiple triggering rules, and each triggering rule is respectively mapped to each fine-tuned low-rank matrix.
[0009] A further improvement of the present invention lies in that The step of integrating the LORA fine-tuning module into the initial generative model to obtain the final composite generative model includes: Combine the original parameter matrix of the initial generative model with the fine-tuned low-rank matrix of the LORA fine-tuning module to obtain the final composite generative model.
[0010] In a second aspect of the present invention, a generative model construction system for power equipment operation and maintenance inspection is provided, including: A data mining module, configured to call power equipment operation and maintenance inspection records and integrate power standard knowledge, and mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge; A first training module, configured to train an initial generative model in a supervised training manner based on the question-and-answer graph; A second training module, configured to perform clustering processing on the obtained fine-tuning instruction set, set multiple fine-tuning types; based on the set multiple fine-tuning types, introduce an initialized low-rank matrix, and train a LORA fine-tuning module in a supervised training manner; wherein, the LORA fine-tuning module includes a plurality of parallel fine-tuned low-rank matrices; A fusion module, configured to integrate the LORA fine-tuning module into the initial generative model to obtain the final composite generative model; wherein, the composite generative model is configured to receive power operation and maintenance scenario data, the LORA fine-tuning module performs scenario-based fine-tuning on the initial generative model according to the power operation and maintenance scenario data to obtain a fine-tuned generative model, and the fine-tuned generative model performs power operation and maintenance analysis according to the power operation and maintenance scenario data to obtain a power operation and maintenance result.
[0011] A further improvement of the present invention lies in that In the step of calling power equipment operation and maintenance inspection records and integrating power standard knowledge to mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge, when mining the question-and-answer graph based on power knowledge, the following processing is performed: Traverse the power equipment operation and maintenance inspection records, perform task homology clustering, mine the entities and relationships of power operation and maintenance inspection, and construct an operation and maintenance knowledge graph; Based on the Q&A logic, transform the operation and maintenance knowledge graph to obtain an initial Q&A graph; wherein, the Q&A logic is determined based on the characteristics of operation and maintenance tasks. Traverse the electric power standard knowledge and compensate the initial Q&A graph to obtain a Q&A graph based on electric power knowledge.
[0012] A further improvement of the present invention lies in In the step of calling the operation and maintenance records of electric power equipment and integrating the electric power standard knowledge to mine a Q&A graph based on electric power knowledge and a fine-tuning instruction set, when mining the fine-tuning instruction set, the following processing is performed: Traverse the operation and maintenance records of electric power equipment to determine the operation and maintenance task characteristics and the tendency characteristics of the operation and maintenance results of each record data. Combine the operation and maintenance task characteristics and the tendency characteristics to generate the fine-tuning instruction set.
[0013] A further improvement of the present invention lies in The step of introducing an initial low-rank matrix based on a set of multiple fine-tuning types and training to obtain a LORA fine-tuning module in a supervised training manner specifically includes: Traverse all fine-tuning types to complete the training of the fine-tuning low-rank matrix corresponding to each fine-tuning type; juxtapose and combine all fine-tuning low-rank matrices to obtain a LORA fine-tuning module. Among them, when traversing all fine-tuning types, for the n fine-tuning type, based on the instruction cluster corresponding to the fine-tuning type, obtain fine-tuning samples; based on the fine-tuning samples, perform supervised training on the initial low-rank matrix to determine the n fine-tuning low-rank matrix. The LORA fine-tuning module is provided with multiple triggering rules, and each triggering rule is respectively mapped to each fine-tuning low-rank matrix.
[0014] A further improvement of the present invention lies in The step of integrating the LORA fine-tuning module into the initial generative model to obtain a final composite generative model includes: Combine the original parameter matrix of the initial generative model and the fine-tuning low-rank matrix of the LORA fine-tuning module to obtain a final composite generative model.
[0015] In the third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the generative model construction method for operation and maintenance of electric power equipment according to any one of the first aspects of the present invention.
[0016] In a fourth aspect of the present invention, there is provided a non-transitory computer-readable storage medium having stored thereon a computer program, which when executed by a processor, implements the method for constructing a generative model for power equipment operation and maintenance inspection according to any one of the first aspects of the present invention.
[0017] In a fifth aspect of the present invention, there is provided a computer program product, which includes a computer program, the computer program comprising computer program code means stored on a computer-readable medium or carrier wave, the computer program code means being configured to cause a computer or processor to control the execution of the steps of the method for constructing a generative model for power equipment operation and maintenance inspection according to any one of the first aspects of the present invention.
[0018] Compared with the prior art, the present invention has the following beneficial effects: In the technical solution disclosed by the present invention, by invoking the records of power equipment operation and maintenance inspection and integrating power standard knowledge, a question-and-answer graph based on power knowledge and a fine-tuning instruction set are mined. Based on the question-and-answer graph, an initial generative model is trained; the fine-tuning instruction set is clustered, the fine-tuning type is set, and an initial low-rank matrix is introduced, and the LORA fine-tuning module is supervised and trained. The LORA fine-tuning module includes a plurality of parallel fine-tuning low-rank matrices. The LORA fine-tuning module is incorporated into the initial generative model to obtain a final composite generative model; explanatorily, in the actual application process, the composite generative model receives data such as power operation and maintenance scenario data (in an exemplary solution, it may also include specific requirements), combines the LORA fine-tuning module and the initial generative model for scenario-based fine-tuning and power operation and maintenance analysis, and finally generates power operation and maintenance data, achieving the remarkable effect of improving the adaptability and decision-making reliability of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art; obviously, the drawings in the following description are some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0020] Figure 1 It is a schematic flowchart of a method for constructing a generative model for power equipment operation and maintenance inspection in an embodiment of the present invention.
[0021] Figure 2 It is a schematic flowchart of a method for efficiently adapting a generative model for power equipment operation and maintenance inspection in a specific embodiment of the present invention.
[0022] Figure 3 It is a schematic flowchart of mining a question-and-answer graph based on power knowledge in a specific embodiment of the present invention.
[0023] Figure 4 This is a schematic diagram of a generative model construction system for power equipment operation and maintenance inspection in an embodiment of the present invention. Detailed implementation manners
[0024] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention; obviously, the described embodiment technical solutions are part of the embodiments of the present invention, rather than all of the embodiments.
[0025] Based on the technical solutions disclosed in the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.
[0026] Please refer to Figure 1 , a generative model construction method for power equipment operation and maintenance inspection provided by an embodiment of the present invention includes the following steps: Step 1: Call the power equipment operation and maintenance records and integrate the power standard knowledge to mine and obtain a question-and-answer graph based on power knowledge and a fine-tuning instruction set. Step 2: Based on the question-and-answer graph obtained in Step 1, train an initial generative model in a supervised training manner. Step 3: Cluster the fine-tuning instruction set obtained in Step 1 and set multiple fine-tuning types; based on the multiple fine-tuning types and introduce an initial low-rank matrix, and train a LORA fine-tuning module in a supervised training manner; wherein, the LORA fine-tuning module includes a plurality of parallel fine-tuning low-rank matrices. Step 4: Incorporate the LORA fine-tuning module into the initial generative model to obtain a final composite generative model; wherein, the composite generative model is used to receive power operation and maintenance scenario data, the LORA fine-tuning module performs scenario-based fine-tuning on the initial generative model according to the power operation and maintenance scenario data to obtain a fine-tuned generative model, and the fine-tuned generative model performs power operation and maintenance analysis according to the power operation and maintenance scenario data to obtain a power operation and maintenance result.
[0027] In the technical solution disclosed in the embodiment of the present invention, by combining the initial generative model and the LORA fine-tuning module for power operation and maintenance analysis, the initial generative model can be scene-tuned through the LORA fine-tuning module according to the power operation and maintenance requirements to achieve efficient adaptation, improving the adaptability and decision reliability of the model. Among them, by using the rich data in the power equipment operation and maintenance records, combined with the standard knowledge of the power industry, through advanced data mining techniques, a question-and-answer graph covering the core issues in the power field is refined. The fine-tuning instruction set can accurately reflect the specific requirements and scenarios in power operation and maintenance, providing precise guidance for subsequent model training adjustment. Specifically, by way of example, based on the constructed question-and-answer graph, an advanced machine learning method is used to train a generative model capable of generating power-related questions and answers; the fine-tuning instruction set is carefully clustered, and different fine-tuning types are set according to the characteristics and requirements of the instructions; the concept of initializing a low-rank matrix is introduced, and the LORA (Low-Rank Adaptation) fine-tuning module is supervised and trained. This module contains multiple parallel fine-tuning low-rank matrices, and each matrix is optimized for a specific fine-tuning type, improving the flexibility and efficiency of fine-tuning; finally, the trained LORA fine-tuning module is seamlessly integrated into the generative model to form a composite model that not only has general question-and-answer capabilities but also can be fine-tuned for specific scenarios.
[0028] Based on the disclosure of the above embodiment, in the actual application process, the technical solution of the embodiment of the present invention can receive real-time data from the power operation and maintenance scenario and the demand information of the user. Combining the general question-and-answer capabilities of the generative model and the specific scenario fine-tuning capabilities of the LORA fine-tuning module, in-depth analysis and scenario fine-tuning of the power operation and maintenance data are carried out. The generated power operation and maintenance data not only accurately reflects the operating status and potential problems of the equipment but also significantly improves the adaptability and decision reliability of the model, providing strong technical support for the operation and maintenance management of the power industry. In summary, the technical solution of the embodiment of the present invention constructs an efficient and flexible power operation and maintenance analysis system by integrating power equipment operation and maintenance records, power standard knowledge, question-and-answer graph, fine-tuning instruction set, and LORA fine-tuning module, effectively improving the intelligent level and decision-making efficiency of power operation and maintenance.
[0029] On the basis of the above embodiment, in an embodiment of the present invention, it may further include: Step 5, performing visualization processing on the power operation and maintenance results through a visualization terminal to obtain power operation and maintenance visualization results, so as to better guide the operation and maintenance management of power equipment.
[0030] In a specific embodiment of the present invention, to mine a question-and-answer graph based on power knowledge, the following processing is performed: Traverse the power equipment operation and maintenance records, perform task homology clustering, mine the entities and relationships of power operation and maintenance, and construct an operation and maintenance knowledge graph; Based on the question-and-answer logic, transform the operation and maintenance knowledge graph to determine the question-and-answer graph; wherein, the question-and-answer logic is determined based on the characteristics of operation and maintenance tasks. Traverse the power standard knowledge, compensate the question-and-answer graph, and obtain the improved question-and-answer graph.
[0031] In the preferred solution of the embodiment of the present invention, first, by traversing the operation and maintenance records of power equipment and task homologous clustering, mine the entities and relationships of power operation and maintenance, and construct an operation and maintenance knowledge graph; then, based on the question-and-answer logic, convert the operation and maintenance knowledge graph into a question-and-answer graph; finally, traverse the power standard knowledge to compensate and improve the question-and-answer graph; the above series of processing steps make the question-and-answer graph more perfect and accurate, and can provide strong support for the subsequent intelligent analysis of power operation and maintenance.
[0032] In a specific embodiment of the present invention, mine the fine-tuning instruction set and perform the following processing: Traverse the operation and maintenance records of power equipment, and determine the operation and maintenance task characteristics and the tendency characteristics of operation and maintenance results of each record; interpretively, when analyzing the operation and maintenance records, focus on the part of the operation and maintenance results, and extract the tendency characteristics of the results; the tendency characteristics of operation and maintenance results may include whether there are abnormalities or faults in the equipment, the types and severity of abnormalities or faults, whether further processing or maintenance is required, suggestions or solutions for processing or maintenance, etc. These tendency characteristics reflect the possible results and trends after the execution of operation and maintenance tasks, and are crucial for generating a targeted fine-tuning instruction set.
[0033] Generate the fine-tuning instruction set based on the operation and maintenance task characteristics and the tendency characteristics; interpretively, the fine-tuning instruction set contains a series of specific instructions or operation suggestions, aiming to fine-tune or optimize the generative model or related system according to specific operation and maintenance task characteristics and tendency characteristics; the content of the instruction set may include adjusting the parameter settings of the model, adding or modifying the training data of the model, introducing new features or variables, optimizing the algorithm or structure of the model, etc. By generating and executing these fine-tuning instructions, the adaptability and accuracy of the generative model in the field of power operation and maintenance can be further improved, providing more powerful technical support for the operation and maintenance management of the power industry.
[0034] In a specific embodiment of the present invention, the supervised training LORA fine-tuning module performs the following processing: For the first instruction cluster corresponding to the first fine-tuning type, obtain the first fine-tuning sample; in a specific exemplary technical solution, according to the characteristics and requirements of the first instruction cluster, obtain the first fine-tuning sample from the operation and maintenance records of power equipment or other relevant data sources. The first fine-tuning sample should contain the input data and expected output related to the first instruction cluster, and is used to supervise the training of the fine-tuning initialized low-rank matrix. Based on the first fine-tuning sample, perform supervised training on the initialized low-rank matrix to determine the first fine-tuned low-rank matrix; Traverse all fine-tuning types to complete the training of the Nth fine-tuned low-rank matrix; in a specific exemplary technical solution, repeat the above process, for each fine-tuning type, obtain the corresponding fine-tuning sample, and perform supervised training on the initialized low-rank matrix to determine the corresponding fine-tuned low-rank matrix; N is the total number of fine-tuning types; Concatenate the first fine-tuned low-rank matrix to the Nth fine-tuned low-rank matrix as the LORA fine-tuning module.
[0035] In an embodiment of the present invention, all the trained fine-tuned low-rank matrices (from the first fine-tuned low-rank matrix to the Nth fine-tuned low-rank matrix) are concatenated together to form an overall LORA fine-tuning module; wherein, each matrix is optimized for a specific fine-tuning type and can flexibly adapt to different fine-tuning requirements.
[0036] In a possible implementation manner, after completing the training of the Nth fine-tuned low-rank matrix, perform the following processing: Set the trigger rule for the fine-tuned low-rank matrix, wherein the trigger rule is set based on scenario factors; in a specific exemplary technical solution, the trigger rule for the fine-tuned low-rank matrix can be set according to the actual application scenario and requirements, and the trigger rule is used to determine under what conditions or scenarios which fine-tuned low-rank matrix should be used for fine-tuning. The trigger rule can be set based on scenario factors such as device type, fault type, operation and maintenance task type, urgency level, etc. By setting a reasonable trigger rule, it can be ensured that the appropriate fine-tuned low-rank matrix is used at the appropriate time; Establish the mapping relationship between the trigger rule and the N fine-tuned low-rank matrices; in a specific exemplary technical solution, establish the mapping relationship between the trigger rule and the N fine-tuned low-rank matrices. This mapping relationship clarifies the fine-tuned low-rank matrix corresponding to each trigger rule, enabling the corresponding fine-tuned low-rank matrix to be quickly found and used according to the trigger rule in actual applications. By establishing the mapping relationship, the flexibility and usability of the LORA fine-tuning module can be improved, enabling it to better adapt to different application scenarios and requirements.
[0037] In a specific embodiment of the present invention, integrate the LORA fine-tuning module into the generative model and perform the following processing: Determine the original parameter matrix of the generative model, where the original parameter matrix is determined based on the training of the question-and-answer graph. Further, by way of example, when training the generative model based on the question-and-answer graph, the question-and-answer graph provides a rich set of power knowledge question-and-answer pairs as the data source for model training. During the training process, the generative model learns the knowledge and logic in the question-and-answer graph and forms its original parameter matrix, which contains the model's understanding and generation ability for power knowledge questions and answers. Combine the original parameter matrix with the fine-tuning low-rank matrix and use it as the way to integrate the LORA fine-tuning module into the generative model. By way of example, when combining the original parameter matrix with the fine-tuning low-rank matrix, a low-rank update strategy can be adopted, that is, using the fine-tuning low-rank matrix as an increment or correction and applying it to the original parameter matrix. Specifically, the fine-tuning low-rank matrix can be weighted and summed with the original parameter matrix, or the fine-tuning low-rank matrix can be multiplied by a scaling factor and then added to the original parameter matrix, so that the information in the fine-tuning low-rank matrix can be integrated into the original parameter matrix, thereby fine-tuning the output of the generative model. Another possible way is to use the fine-tuning low-rank matrix as a sub-matrix or subspace of the original parameter matrix and integrate it into the original parameter matrix through specific transformation or projection methods, which can more precisely control the impact of the fine-tuning low-rank matrix on the original parameter matrix.
[0038] In a possible implementation, in combination with the generative model for power operation and maintenance analysis and scenario-based fine-tuning, perform the following processing: Identify the operation and maintenance scenario data, trigger the target fine-tuning low-rank matrix by matching the trigger rule; using the target fine-tuning low-rank matrix as the output constraint, the generative model performs operation and maintenance analysis in the power scenario on the operation and maintenance scenario data to determine the power operation and maintenance data.
[0039] In a possible implementation, after generating the power operation and maintenance data, perform the following processing: Determine the preset expansion period, integrate the periodic power operation and maintenance data; based on the power operation and maintenance data, expand the question-and-answer graph to determine the expanded graph; based on the expanded graph, update the generative model.
[0040] Please refer to Figure 2 and Figure 3 In a specific embodiment of the present invention, a method for efficiently adapting a generative model for power equipment operation and maintenance is provided, including the following steps: Step S100: Invoke the operation and maintenance records of power equipment, integrate power standard knowledge, and mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge. Specifically, retrieve and load the historical operation and maintenance data of power equipment from the database, including equipment status, operation steps, and their corresponding execution results. Integrate knowledge such as the standard operation procedures and safety specifications of power equipment into the operation and maintenance records to form a comprehensive knowledge base. Using data mining techniques, identify common problems, fault types, and their solutions for power equipment based on the operation and maintenance records of power equipment and their standard knowledge, and construct a question-and-answer graph (a knowledge structure represented in the form of questions and answers). Analyze specific situations and conditions in the operation and maintenance records of power equipment, and extract an instruction set that can guide the generative model to adjust its output in specific scenarios, that is, the fine-tuning instruction set.
[0041] As Figure 3 shown, in a possible implementation, to mine a question-and-answer graph based on power knowledge, step S100 further specifically includes: Step S110: Traverse the operation and maintenance records of the power equipment, perform task homologous clustering, mine the entities and relationships of power operation and maintenance, and construct an operation and maintenance knowledge graph. Specifically, the operation and maintenance records of power equipment include equipment information, maintenance history, fault records, etc. Using clustering algorithms (such as K-means, hierarchical clustering, etc.), according to features such as task type, equipment type, and fault type in the operation and maintenance records, divide the records into different clusters, and these clusters represent operation and maintenance tasks with similar properties or sources. In each cluster, identify key entities (such as equipment, faults, components, etc.) and the relationships between them (such as the relationship between faults and equipment, the relationship between equipment and maintenance plans, etc.). These entities and relationships constitute the basic elements of the operation and maintenance knowledge graph. Use a graph database or a graph construction tool to represent the mined entities and relationships in the form of a graph to form an operation and maintenance knowledge graph. The nodes in the graph represent entities, and the edges represent relationships.
[0042] Step S120: Based on the question-and-answer logic, transform the operation and maintenance knowledge graph to determine the question-and-answer graph, where the question-and-answer logic is determined based on the characteristics of the operation and maintenance tasks. Specifically, according to the characteristics and requirements of the operation and maintenance tasks, design a set of question-and-answer logic, which defines how to extract information from the operation and maintenance knowledge graph to answer specific questions. Using the defined question-and-answer logic, transform the operation and maintenance knowledge graph into a question-and-answer graph. Each node or edge in the question-and-answer graph corresponds to a possible question or answer, or a mapping relationship from a question to an answer. Optimize the transformed question-and-answer graph, including adjusting the graph structure, adding indexes or caches, etc., to ensure that it can efficiently support question-and-answer operations.
[0043] Step S130: Traverse the power standard knowledge and compensate the Q&A graph. Specifically, compare and integrate the power standard knowledge with the Q&A graph. For the missing or incomplete information in the Q&A graph, use the power standard knowledge to supplement and improve it. This implementation method clusters operation and maintenance tasks with similar properties through task homology clustering, which facilitates knowledge mining and graph construction. The graph conversion based on Q&A logic enables the graph to directly support Q&A operations, facilitating the extraction of information from the graph to answer specific questions. By traversing the power standard knowledge to compensate the Q&A graph, the accuracy and integrity of the graph are ensured, providing sufficient training data for the generative model.
[0044] In a possible implementation, to mine the fine-tuning instruction set, step S100 further includes: Step S140: Traverse the power equipment operation and maintenance records, and determine the operation and maintenance task characteristics and the tendency characteristics of the operation and maintenance results for each record. Specifically, the power equipment operation and maintenance records include fields such as the basic information of the equipment, operation and maintenance time, operation and maintenance personnel, operation and maintenance task description, and operation and maintenance results. Extract key information from the operation and maintenance task description, such as equipment type, fault type, maintenance location, etc., as the characteristics of the operation and maintenance tasks. Analyze the operation and maintenance results to determine whether they contain specific tendency information, such as whether the fault occurs frequently, whether the maintenance is successful, whether components need to be replaced, etc.
[0045] Step S150: Generate the fine-tuning instruction set based on the operation and maintenance task characteristics and the tendency characteristics. Specifically, analyze the correlation between the operation and maintenance task characteristics and the operation and maintenance result tendency characteristics to determine which characteristics have a significant impact on the operation and maintenance results. According to the results of the feature correlation analysis, determine the types of fine-tuning instructions to be generated, such as maintenance suggestions for specific fault types, preventive maintenance plans for frequently faulty equipment, etc. Generate the specific content of the fine-tuning instructions based on the specific values of the operation and maintenance task characteristics and the tendency characteristics, including maintenance steps, required tools, precautions, etc. Optimize the generated fine-tuning instruction set to ensure the accuracy and feasibility of the instructions, including removing redundant instructions, adjusting the order of instructions, adding necessary explanations, etc. Store the generated fine-tuning instruction set in the database and set an update mechanism to ensure that the fine-tuning instruction set can be continuously optimized as the operation and maintenance records are updated. This implementation method generates specific fine-tuning instructions based on the operation and maintenance task characteristics and the operation and maintenance result tendency characteristics, and these instructions reflect the correlation between different operation and maintenance tasks and results, improving the efficiency and accuracy of power equipment operation and maintenance.
[0046] Step S200: Supervise and train the generative model based on the question-and-answer knowledge graph. Specifically, use the questions and answers in the question-and-answer knowledge graph as training data to supervise and train the generative model. Through training, the generative model can accurately understand and generate questions and answers related to the operation and maintenance of power equipment.
[0047] Step S300: Cluster the fine-tuning instruction set, set the fine-tuning types, introduce an initial low-rank matrix, supervise and train the LORA fine-tuning module, and integrate the LORA fine-tuning module into the generative model. The LORA fine-tuning module includes multiple parallel fine-tuning low-rank matrices. Specifically, cluster the fine-tuning instruction set according to similar task characteristics to form different fine-tuning types. Initialize and generate the LORA (Low-Rank Adaptation) fine-tuning module using a low-rank matrix, and train each fine-tuning type to obtain different fine-tuning low-rank matrices. Among them, the LORA fine-tuning module is a lightweight model fine-tuning method that can, without changing most of the model's parameters, fine-tune the model output by adjusting a few low-rank matrices. Finally, combine the LORA fine-tuning module with the generative model so that the generative model can adjust its output according to the fine-tuning instruction set in a specific scenario.
[0048] In a possible implementation, the step of supervising and training the LORA fine-tuning module, step S300 further includes: Step S310: For the first instruction cluster corresponding to the first fine-tuning type, obtain the first fine-tuning samples. Specifically, the fine-tuning type refers to specific adjustment requirements for different power equipment (such as transformers, circuit breakers), different operation and maintenance tasks (such as preventive maintenance, fault troubleshooting), or different operating conditions (such as high-temperature, high-humidity environments). Each fine-tuning type corresponds to a specific set of fine-tuning instructions, which are designed to make the generative model better adapt to a specific scenario. Among them, the first instruction cluster is the set of all fine-tuning instructions related to the first fine-tuning type, which is mined from the operation and maintenance records of power equipment and reflects the standard practices or best practices for handling problems in this specific scenario. Based on the first instruction cluster, prepare the samples for supervising and training the LORA fine-tuning module, including input questions (such as "What is the treatment method for too high transformer oil temperature?") and the corresponding expected outputs (such as a series of standard treatment steps for this question).
[0049] Step S320: Based on the first fine-tuning sample, perform supervised training on the initialized low-rank matrix to determine the first fine-tuned low-rank matrix. Specifically, the core idea of the LORA fine-tuning module is to use a low-rank matrix to reduce the amount of model parameter updates, thereby accelerating the training speed and reducing computational resource consumption. The initialized low-rank matrix is the starting point of the LORA fine-tuning module, which is a matrix with fewer parameters and is used to capture the characteristics of specific scenarios during training. The initialized low-rank matrix is trained using the first fine-tuning sample to minimize the difference between the model output and the expected output, thereby adjusting the parameters of the low-rank matrix to enable it to reflect the characteristics of a specific fine-tuning type.
[0050] Step S330: Traverse the fine-tuning types to complete the training of the Nth fine-tuned low-rank matrix. Specifically, perform iterative processing on all fine-tuning types. For each fine-tuning type, repeat steps S310 and S320 until the fine-tuned low-rank matrices of all types are trained. The Nth fine-tuned low-rank matrix refers to the low-rank matrix obtained by training for the last (or the Nth) fine-tuning type.
[0051] Step S340: Concatenate the first fine-tuned low-rank matrix to the Nth fine-tuned low-rank matrix as the LORA fine-tuning module. Specifically, concatenate all the trained fine-tuned low-rank matrices together to form a complete LORA fine-tuning module. This module can be regarded as a neural network layer with multiple parallel channels, and each channel corresponds to a specific fine-tuning type. When it is necessary to fine-tune the generative model, the corresponding fine-tuned low-rank matrix can be selected for application according to the specific scenario. This implementation method trains an independent low-rank matrix for each fine-tuning type, and the LORA fine-tuning module can flexibly adapt to different operation and inspection scenarios, which enables the generative model to be precisely adjusted for specific scenarios while maintaining generality.
[0052] In a possible implementation, after completing the training of the Nth fine-tuned low-rank matrix, step S300 further includes: Step S350: Set the triggering rules for the fine-tuned low-rank matrix, where the triggering rules are set based on scenario factors. Specifically, in the LORA fine-tuning module, each fine-tuned low-rank matrix corresponds to a specific fine-tuning type, which is used to fine-tune the generative model in a specific scenario. The triggering rules are the conditions or criteria for determining when to apply a specific fine-tuned low-rank matrix, and these rules are set based on scenario factors, that is, they are closely related to the actual scenarios of power equipment operation and maintenance. Specifically, analyze various scenario factors that may be encountered in the process of power equipment operation and maintenance, such as equipment type, operating status, environmental conditions, etc. Based on the analysis results, set one or more triggering conditions for each fine-tuned low-rank matrix. For example, for the fine-tuned low-rank matrix for dealing with transformer failures in a high-temperature environment, the triggering condition can be "the equipment type is a transformer and the environmental temperature is higher than 40°C". Combine the triggering conditions into a complete triggering rule, and these rules can be simple logical judgments (such as "and", "or" relationships), or complex condition combinations.
[0053] Step S360: Establish the mapping relationship between the triggering rules and N fine-tuned low-rank matrices. Specifically, the mapping relationship refers to the process of associating the triggering rules with the corresponding fine-tuned low-rank matrices, so that when a certain triggering rule is met, the corresponding fine-tuned low-rank matrix can be automatically selected and applied. First, create a mapping table to store the corresponding relationship between the triggering rules and the fine-tuned low-rank matrices. This table can be a simple list or a complex database structure. For each fine-tuned low-rank matrix, add its triggering rule as the key and the fine-tuned low-rank matrix itself or its identifier as the value to the mapping table. Finally, verify the correctness of the mapping relationship by simulating different operation and maintenance scenarios. Ensure that in each scenario, the corresponding fine-tuned low-rank matrix can be correctly triggered and applied. This implementation method realizes the automatic selection and application of the fine-tuned low-rank matrix by setting the triggering rules and establishing the mapping relationship, ensuring that the corresponding fine-tuned low-rank matrix can be quickly and accurately applied when needed, and improving the efficiency and accuracy of the operation and maintenance process.
[0054] In a possible implementation, integrate the LORA fine-tuning module into the generative model, and step S300 further includes: Step S370, determine the original parameter matrix of the generative model, where the original parameter matrix is determined based on the training of the question-and-answer graph. Specifically, the original parameter matrix is the core component of the generative model, containing all the parameters learned by the model during the training process. These parameters determine the behavior and performance of the model and are the basis for the model to generate accurate question-and-answer content. Load the question-and-answer graph data from the storage medium, which includes question-and-answer pairs, the relationships between question-and-answer pairs, and related attribute information. Use the loaded question-and-answer graph data as training data and train the generative model through supervised learning. During the training process, the model continuously adjusts its parameter matrix to minimize the difference between the predicted output and the true output. When the model training is completed, save its parameter matrix as the original parameter matrix. This matrix contains all the knowledge learned by the model during the training process and is the basis for fine-tuning.
[0055] Step S380, combine the original parameter matrix with the fine-tuning low-rank matrix as the way to integrate the LORA fine-tuning module into the generative model. Specifically, in the LORA fine-tuning module, each fine-tuning low-rank matrix corresponds to a specific fine-tuning type for fine-tuning the generative model in a specific scenario. These matrices have a low rank, so the amount of parameter updates can be reduced while maintaining the model performance. The way of combining the original parameter matrix with the fine-tuning low-rank matrix is the key to implementing the integration of the LORA fine-tuning module into the generative model. This combination is achieved through matrix addition or multiplication, that is, adding the fine-tuning low-rank matrix to the original parameter matrix or multiplying them to complete, and use the combined parameter matrix to update the generative model. This implementation method combines the original parameter matrix with the fine-tuning low-rank matrix, integrates the LORA fine-tuning module into the generative model, and improves the performance and adaptability of the power equipment operation and maintenance application.
[0056] Step S400, receive the operation and maintenance scenario data, combine with the generative model for power operation and maintenance analysis and scenario-based fine-tuning, and generate power operation and maintenance data. Specifically, receive the real-time data from the power equipment operation and maintenance site, including equipment status data, fault records, maintenance records, etc., input these data into the generative model, and perform scenario-based fine-tuning on the output of the generative model according to the operation and maintenance scenario and the fine-tuning instruction set, and use the generative model for power operation and maintenance analysis to generate power operation and maintenance data that meets specific requirements.
[0057] In a possible implementation manner, when combining with the generative model for power operation and maintenance analysis and scenario-based fine-tuning, step S400 further includes: Step S410: Identify the operation and maintenance scenario data, and trigger the target fine-tuned low-rank matrix by matching the trigger rules. Specifically, preprocess the received operation and maintenance scenario data, including data cleaning, format conversion, etc., to ensure the quality and consistency of the data. Utilize the preprocessed data and combine it with predefined scenario recognition algorithms or models to identify the type or characteristics of the current operation and maintenance scenario. Match the identified scenario type or characteristics with the predefined trigger rules to determine which fine-tuned low-rank matrix should be triggered. According to the matching result, select the corresponding target fine-tuned low-rank matrix from the LORA fine-tuning module.
[0058] Step S420: Using the target fine-tuned low-rank matrix as the output constraint, the generative model performs operation and maintenance analysis in the power scenario on the operation and maintenance scenario data to determine the power operation and maintenance data. Specifically, load the trained generative model and the LORA fine-tuning module, and use the preprocessed operation and maintenance scenario data and the selected target fine-tuned low-rank matrix as inputs and pass them to the generative model. Under the constraint of the target fine-tuned low-rank matrix, the generative model comprehensively analyzes the input data to generate power operation and maintenance data, including equipment status assessment, fault prediction, maintenance suggestions, etc. This implementation method can identify the operation and maintenance scenario data and trigger the target fine-tuned low-rank matrix, enabling the generative model to perform intelligent analysis for specific scenarios, provide accurate equipment status assessment and maintenance suggestions, and achieve scenario-based intelligent operation and maintenance.
[0059] In a possible implementation manner, after generating the power operation and maintenance data, step S400 further includes: Step S430: Determine a preset expansion period and integrate the periodic power operation and maintenance data. Specifically, according to the characteristics of power equipment and operation and maintenance requirements, set a reasonable expansion period for regularly integrating and analyzing periodic power operation and maintenance data, such as weekly, monthly, or quarterly, etc. During the expansion period, regularly collect power operation and maintenance data, including equipment status, fault records, maintenance records, etc. Integrate the collected periodic power operation and maintenance data, remove duplicate or invalid data, and ensure the accuracy and integrity of the data. The periodic power operation and maintenance data has time series characteristics and reflects the operating conditions and change trends of power equipment over a period of time.
[0060] Step S440: Based on the power operation and maintenance data, expand the Q&A graph to determine the expanded graph. Specifically, preprocess the integrated power operation and maintenance data, including data cleaning, format conversion, etc., so as to match and associate with the Q&A graph. Match the preprocessed data with the Q&A graph, find new Q&A pairs and relationships, and add them to the graph. At the same time, update and optimize the existing Q&A pairs and relationships in the graph. After matching and expansion, a new Q&A graph, that is, the expanded graph, is obtained. This graph contains more power knowledge and operation and maintenance experience, and can provide richer training data and inference basis for the generative model.
[0061] Step S450: Update the generative model based on the expanded graph. Specifically, load the original generative model, including its parameters and structure, etc. Extract Q&A pairs and relationships from the expanded graph as training data, and use the training data to train the generative model, including parameter update, structure adjustment, etc. Evaluate the trained model, including tests on aspects such as accuracy and robustness. According to the evaluation results, optimize and adjust the model to improve its performance. Replace the original generative model with the trained and optimized model to complete the update of the model. This implementation method can provide more training data and inference basis for the generative model by continuously integrating new power operation and maintenance data and expanding the Q&A graph, thereby improving the accuracy and adaptability of the model.
[0062] Step S500: Perform terminal visualization on the power operation and maintenance data to guide the operation and maintenance management of power equipment. Specifically, display the generated power operation and maintenance data to users or managers through visualization means (such as charts, reports, etc.). Use the visualized data to guide the operation and maintenance management work of power equipment, such as formulating maintenance plans and optimizing equipment maintenance strategies.
[0063] An efficient adaptation method for a generative model for power equipment operation and maintenance provided by an embodiment of the present invention first calls the power equipment operation and maintenance records, integrates power standard knowledge, mines a Q&A graph and a fine-tuning instruction set based on power knowledge, then supervises the training of the generative model based on the Q&A graph, then clusters the fine-tuning instruction set, sets the fine-tuning type, and introduces an initial low-rank matrix to supervise the training of the LORA fine-tuning module. Incorporate the LORA fine-tuning module into the generative model. The LORA fine-tuning module contains multiple parallel fine-tuning low-rank matrices. Then receive operation and maintenance scenario data, combine with the generative model for power operation and maintenance analysis and scenario-based fine-tuning to generate power operation and maintenance data. Finally, perform terminal visualization on the power operation and maintenance data to guide the operation and maintenance management of power equipment, achieving the technical effect of improving the adaptability and decision-making reliability of the model.
[0064] The following is an apparatus embodiment of the present invention, which can be used to implement the method embodiment of the present invention. For details not disclosed in the apparatus embodiment, please refer to the method embodiment of the present invention.
[0065] Please refer to Figure 4 , in an embodiment of the present invention, a generative model construction system for power equipment operation and maintenance inspection is provided, including: A data mining module, configured to call power equipment operation and maintenance records and integrate power standard knowledge, and mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge; A first training module, configured to train an initial generative model in a supervised training manner based on the question-and-answer graph; A second training module, configured to perform clustering processing on the obtained fine-tuning instruction set, and set multiple fine-tuning types; based on the set multiple fine-tuning types, introduce an initialized low-rank matrix, and train a LORA fine-tuning module in a supervised training manner; wherein, the LORA fine-tuning module includes a plurality of parallel fine-tuning low-rank matrices; A fusion module, configured to integrate the LORA fine-tuning module into the initial generative model to obtain a final composite generative model; wherein, the composite generative model is configured to receive power operation and maintenance scenario data, the LORA fine-tuning module performs scenario-based fine-tuning on the initial generative model according to the power operation and maintenance scenario data to obtain a fine-tuned generative model, and the fine-tuned generative model performs power operation and maintenance analysis according to the power operation and maintenance scenario data to obtain a power operation and maintenance result.
[0066] In an embodiment of the present invention, a computer device is provided. The computer device includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may also be 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. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used to execute the operations of the method for constructing a generative model for power equipment operation and maintenance inspection.
[0067] In one embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and, of course, the extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space, and the operating system of the terminal is stored in this storage space. Moreover, one or more instructions suitable for being loaded and executed by the processor are stored in this storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM (Random Access Memory) or a non-volatile memory, such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the method for constructing a generative model for power equipment operation and maintenance inspection in the above embodiments.
[0068] In one embodiment of the present invention, a computer program product is provided. The computer program product includes a computer program, and the computer program can be used to cause a computer or a processor to control the execution of the steps of a method according to any implementation manner of the above method for constructing a generative model for power equipment operation and maintenance inspection.
[0069] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, optical memories, etc.) containing computer-usable program codes.
[0070] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 one process or multiple processes and / or blocks Figure 1a device for the functions specified in one or more boxes.
[0071] These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, and the instruction device implements the functions specified in one Figure 1 process or more processes and / or boxes Figure 1 a box or more boxes.
[0072] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 process or more processes and / or boxes Figure 1 a box or more boxes.
[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the specific implementation manners of the present invention can still be modified or equivalently replaced, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.
Claims
1. A method for constructing a generative model for operation and maintenance of power equipment, characterized in that It includes the following steps: Call the operation and maintenance records of power equipment and integrate power standard knowledge to mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge; Based on the question-and-answer graph, train an initial generative model in a supervised training manner; Cluster the obtained fine-tuning instruction set and set multiple fine-tuning types; based on the set multiple fine-tuning types, introduce an initial low-rank matrix and train a LORA fine-tuning module in a supervised training manner; wherein, the LORA fine-tuning module includes multiple parallel fine-tuning low-rank matrices; Integrate the LORA fine-tuning module into the initial generative model to obtain a final composite generative model; wherein, the composite generative model is used to receive power operation and maintenance scenario data, the LORA fine-tuning module performs scenario-based fine-tuning on the initial generative model according to the power operation and maintenance scenario data to obtain a fine-tuned generative model, and the fine-tuned generative model performs power operation and maintenance analysis according to the power operation and maintenance scenario data to obtain a power operation and maintenance result.
2. The method for constructing a generative model for power equipment operation and maintenance according to claim 1, wherein In the step of calling the operation and maintenance records of power equipment and integrating power standard knowledge to mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge, when mining the question-and-answer graph based on power knowledge, the following processing is performed: Traverse the operation and maintenance records of power equipment, perform task homologous clustering, mine the entities and relationships of power operation and maintenance, and construct an operation and maintenance knowledge graph; Based on the question-and-answer logic, transform the operation and maintenance knowledge graph to obtain an initial question-and-answer graph; wherein, the question-and-answer logic is determined based on the characteristics of operation and maintenance tasks; Traverse the power standard knowledge and compensate the initial question-and-answer graph to obtain a question-and-answer graph based on power knowledge.
3. The method for constructing a generative model for power equipment operation and maintenance according to claim 1, wherein In the step of calling the operation and maintenance records of power equipment and integrating power standard knowledge to mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge, when mining the fine-tuning instruction set, the following processing is performed: Traverse the operation and maintenance records of power equipment and determine the operation and maintenance task characteristics and the tendency characteristics of operation and maintenance results of each record data; Combine the operation and maintenance task characteristics and the tendency characteristics to generate the fine-tuning instruction set.
4. The method for constructing a generative model for power equipment operation and maintenance according to claim 1, wherein The step of introducing an initial low-rank matrix based on the set multiple fine-tuning types and training a LORA fine-tuning module in a supervised training manner specifically includes: Traverse all fine-tuning types, complete the training of the fine-tuning low-rank matrices corresponding to each fine-tuning type; combine all the fine-tuning low-rank matrices in parallel to obtain a LORA fine-tuning module; Among them, when traversing all fine-tuning types, for the n fine-tuning type, based on the instruction cluster corresponding to the fine-tuning type, obtain fine-tuning samples; based on the fine-tuning samples, perform supervised training on the initialized low-rank matrix to determine the n fine-tuned low-rank matrix; The LORA fine-tuning module is provided with multiple trigger rules, and each trigger rule is respectively mapped to each fine-tuning low-rank matrix.
5. The method for constructing a generative model for power equipment operation and maintenance according to claim 1, wherein The steps of integrating the LORA fine-tuning module into the initial generative model to obtain the final composite generative model include: Combining the original parameter matrix of the initial generative model with the fine-tuning low-rank matrix of the LORA fine-tuning module to obtain the final composite generative model.
6. A generative model construction system for the operation and maintenance of power equipment, characterized in that, Including: A data mining module, configured to call power equipment operation and maintenance records and integrate power standard knowledge, and mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge; A first training module, configured to train an initial generative model in a supervised training manner based on the question-and-answer graph; A second training module, configured to perform clustering processing on the obtained fine-tuning instruction set and set multiple fine-tuning types; based on the set multiple fine-tuning types, introduce an initial low-rank matrix, and train a LORA fine-tuning module in a supervised training manner; wherein, the LORA fine-tuning module includes a plurality of parallel fine-tuning low-rank matrices; A fusion module, configured to integrate the LORA fine-tuning module into the initial generative model to obtain the final composite generative model; wherein, the composite generative model is configured to receive power operation and maintenance scenario data, the LORA fine-tuning module performs scenario-based fine-tuning on the initial generative model according to the power operation and maintenance scenario data to obtain a fine-tuned generative model, and the fine-tuned generative model performs power operation and maintenance analysis according to the power operation and maintenance scenario data to obtain a power operation and maintenance result.
7. The generative model construction system for power equipment operation and maintenance according to claim 6, wherein In the step of calling power equipment operation and maintenance records and integrating power standard knowledge to mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge, when mining the question-and-answer graph based on power knowledge, the following processing is performed: Traverse the power equipment operation and maintenance records, perform task homology clustering, mine the entities and relationships of power operation and maintenance, and construct an operation and maintenance knowledge graph; Based on the question-and-answer logic, transform the operation and maintenance knowledge graph to obtain an initial question-and-answer graph; wherein, the question-and-answer logic is determined based on the characteristics of the operation and maintenance tasks; Traverse the power standard knowledge, and compensate the initial question-and-answer graph to obtain a question-and-answer graph based on power knowledge.
8. The generative model construction system for power equipment operation and maintenance according to claim 6, wherein In the step of calling power equipment operation and maintenance records and integrating power standard knowledge to mine a question-and-answer graph and a fine-tuning instruction set based on power knowledge, when mining the fine-tuning instruction set, the following processing is performed: Traverse the power equipment operation and maintenance records, and determine the operation and maintenance task characteristics and the tendency characteristics of the operation and maintenance results of each record data; Combine the operation and maintenance task characteristics with the tendency characteristics to generate the fine-tuning instruction set.
9. The generative model construction system for power equipment operation and maintenance according to claim 6, wherein The step of introducing an initial low-rank matrix based on the set multiple fine-tuning types and training a LORA fine-tuning module in a supervised training manner specifically includes: Traverse all fine-tuning types, complete the training of the fine-tuning low-rank matrices corresponding to each fine-tuning type; combine all the fine-tuning low-rank matrices in parallel to obtain the LORA fine-tuning module; Among them, when traversing all fine-tuning types, for the n fine-tuning type, based on the instruction cluster corresponding to the fine-tuning type, obtain fine-tuning samples; based on the fine-tuning samples, perform supervised training on the initialized low-rank matrix to determine the n fine-tuned low-rank matrix; The LORA fine-tuning module is provided with multiple triggering rules, and each triggering rule is respectively mapped to each fine-tuning low-rank matrix.
10. A generative model construction system for power equipment operation and maintenance according to claim 6, wherein, The step of integrating the LORA fine-tuning module into the initial generative model to obtain the final composite generative model includes: Combining the original parameter matrix of the initial generative model with the fine-tuning low-rank matrix of the LORA fine-tuning module to obtain the final composite generative model.
11. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the generative model construction method for power equipment operation and maintenance according to any one of claims 1 to 5.
12. A non-transitory computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the generative model construction method for power equipment operation and maintenance according to any one of claims 1 to 5.
13. A computer program product, characterized in that, The computer program product includes a computer program, the computer program includes computer program code means stored on a computer-readable medium or carrier wave, and the computer program code means is configured to enable a computer or a processor to control the execution of the steps of the generative model construction method for power equipment operation and maintenance according to any one of claims 1 to 5.