An electrical equipment fault diagnosis method and device based on a vector knowledge base and an agent

Abstract

The application discloses an electrical equipment fault diagnosis method and device based on a vector knowledge base and an agent, and relates to the technical field of power system operation and maintenance. The electrical equipment fault diagnosis method based on the vector knowledge base and the agent comprises the following steps: receiving demand information input by an operation and maintenance personnel; performing prompt word optimization on the demand information to obtain a semantic segment with a unified format as an optimized prompt word; inputting the optimized prompt word into the agent; querying the vector knowledge base by the agent to determine whether similar data matching the optimized prompt word exist; if the similar data exist, inputting the matched similar data into a large language model as input, generating a diagnosis conclusion according to reasoning of the large language model, and outputting the diagnosis conclusion; and if the similar data do not exist, calling an external method to constrain the output of the large language model, generating a constrained diagnosis conclusion, and outputting the diagnosis conclusion. The diagnosis result conforms to the power operation and maintenance specification, and is suitable for fault diagnosis of electrical equipment.

Description

Technical Field

[0001] This invention relates to a method and apparatus for diagnosing electrical equipment faults based on a vector knowledge base and intelligent agents, belonging to the field of power system operation and maintenance technology. Background Technology

[0002] Substation main equipment (such as transformers, circuit breakers, disconnectors, and instrument transformers) is the core carrier of power transmission and distribution in the power system, and its operating status directly affects the safety and stability of the power grid. Substation main equipment faults are characterized by their high degree of concealment and complex data types, and traditional fault diagnosis methods have significant limitations.

[0003] Substation operation and maintenance data includes multimodal heterogeneous data such as sensor time-series measurement data and relay protection action text. Traditional machine learning methods mostly rely on a single data type for diagnosis, which cannot achieve deep integration of multimodal data and form data silos.

[0004] The fault diagnosis experience of power grid experts is mostly in the form of text reports and on-site oral accounts, which is difficult to digitize and structure. This makes it difficult for new operation and maintenance personnel to quickly reuse historical experience, and the diagnostic efficiency is greatly affected by the experience level of the personnel.

[0005] While existing large-scale model algorithms possess powerful natural language understanding and logical reasoning capabilities, they suffer from the "illusion" problem and are unable to efficiently retrieve massive historical fault cases from substations, making it difficult to deeply integrate large-scale models with power industry scenarios. Summary of the Invention

[0006] The purpose of this invention is to provide a method and device for diagnosing electrical equipment faults based on a vector knowledge base and an intelligent agent. By combining a vector database and a skill base, it enables the feature-based storage and rapid matching of multimodal data, providing support for historical typical cases, and solving the problem of difficult fault diagnosis and analysis of electrical equipment in substations.

[0007] To achieve the above objectives, the present invention is implemented using the following technical solution.

[0008] On the one hand, the present invention provides a method for diagnosing electrical equipment faults based on vector knowledge bases and intelligent agents, including:

[0009] Receive the requirement information input by maintenance personnel;

[0010] The required information is optimized by selecting prompt words to obtain semantic fragments with a uniform format, which are then used as optimized prompt words.

[0011] The optimized prompts are input into the agent;

[0012] The agent queries the vector knowledge base to determine whether similar data matching the optimized prompt words exists.

[0013] If similar data exists, the matched similar data is used as input to the large language model, and a diagnostic conclusion is generated and output based on the reasoning of the large language model.

[0014] If no similar data exists, an external method is invoked to constrain the output of the large language model, generating and outputting constrained diagnostic conclusions.

[0015] Optionally, the process of optimizing the prompt words for the demand information is as follows:

[0016] Obtain the requirements information entered by the operations and maintenance personnel from the dialog window;

[0017] The requirements information input by maintenance personnel is optimized to obtain semantic fragments with consistent format;

[0018] The prompt word optimization method uses rules for electrical equipment fault diagnosis to limit the optimization format of prompt words, and the rules are written using Markdown syntax;

[0019] Input requirements information Compared with the optimized prompt words Semantic similarity comparison is performed to prevent semantic changes during the optimization process. The formula is expressed as:

[0020] ;

[0021] In the formula, Represents cosine similarity. This represents the similarity value before and after optimization;

[0022] When the similarity value is higher than the threshold, no manual review is required; otherwise, manual review is required.

[0023] Optionally, the vector knowledge base construction process includes:

[0024] Collecting private data about electrical equipment within the power system;

[0025] The private data is formatted and then stored in the Milvus database.

[0026] The optimized prompts are matched with database content using search enhancement techniques to obtain matching results with a confidence level higher than a preset threshold.

[0027] Optionally, the formatting process for the private data includes:

[0028] For text data, the following operations are performed sequentially: original document extraction, text content extraction, block processing, vectorization, and storage. The block processing uses a sliding window to segment the text content.

[0029] For image data, the following operations are performed sequentially: original image preprocessing, image inpainting, vectorization, dimensionality reduction, and storage. The image inpainting includes at least one of noise reduction processing and super-resolution processing, and the dimensionality reduction is achieved through principal component analysis.

[0030] Optionally, the step of invoking an external method to constrain the output of the large language model includes:

[0031] Build a skill library, which includes script calling rules and corresponding script methods;

[0032] Based on the optimized prompt words, the matching script method is invoked to generate the corresponding script execution result;

[0033] The script execution result is used as input to a large language model. The large language model performs reasoning and output processing to generate a constrained diagnostic conclusion, expressed by the following formula:

[0034] ;

[0035] in, This indicates the optimized prompt words. This refers to the script method called from the skill library. This indicates a call to the large language model. This represents the diagnostic conclusion output by the large language model.

[0036] Optionally, the electrical equipment fault diagnosis method also includes fine-tuning the large language model:

[0037] The LoRA fine-tuning method is adopted to introduce a low-rank matrix on the basis of the original model parameters of the large language model. Only the parameters of the low-rank matrix are updated to train a large language model suitable for the field of electrical equipment fault diagnosis.

[0038] Optionally, after outputting the diagnostic conclusion, the diagnostic conclusion and the feedback data from the operation and maintenance personnel are stored in the vector knowledge base as newly added fault cases.

[0039] According to a preset cycle, the newly added fault cases are used to iteratively optimize the large language model to obtain an optimized large language model for subsequent diagnostic tasks.

[0040] Secondly, the present invention provides an electrical equipment fault diagnosis device based on a vector knowledge base and an intelligent agent, comprising:

[0041] The prompt word optimization module is used to: receive the requirement information input by maintenance personnel;

[0042] The required information is optimized by selecting prompt words to obtain semantic fragments with a uniform format, which are then used as optimized prompt words.

[0043] The fault diagnosis module is used to: input the optimized prompt words into the intelligent agent;

[0044] The agent queries the vector knowledge base to determine whether similar data matching the optimized prompt words exists.

[0045] If similar data exists, the matched similar data is used as input to the large language model, and a diagnostic conclusion is generated and output based on the reasoning of the large language model.

[0046] If no similar data exists, an external method is invoked to constrain the output of the large language model, generating and outputting constrained diagnostic conclusions.

[0047] Thirdly, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent as described in any of the first aspects.

[0048] Fourthly, the present invention provides a computer device / equipment / system, characterized in that it comprises:

[0049] Memory, used to store computer programs / instructions;

[0050] A processor for executing the computer program / instructions to implement the steps of the electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent as described in any of the first aspects.

[0051] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0052] 1. This invention can fine-tune a large model by integrating power industry standards and fault cases, and the diagnostic results conform to power operation and maintenance specifications, and are applicable to fault diagnosis of electrical equipment;

[0053] 2. This invention can handle the input content from different operations and maintenance personnel. Due to differences in experience, different operations and maintenance personnel may have different ways of expressing themselves. This invention can standardize the input format of operations and maintenance personnel, making it easier for the Agent to output more accurate content.

[0054] 3. This invention achieves unified storage and association of electrical equipment image, text and other data through a multimodal vector database, breaking down "data silos" and effectively improving the utilization rate of cross-modal fault data;

[0055] 4. This invention digitizes the fault diagnosis experience of power grid experts by vectorizing text and storing the results in a database, enabling newly hired maintenance personnel to quickly reuse historical experience.

[0056] 5. This invention constructs a fusion architecture of vector retrieval and large model inference, enabling the large model to diagnose based on real historical fault cases and power industry standards, effectively solving the illusion problem of large models in power scenarios. Attached Figure Description

[0057] Figure 1 This is a flowchart of the electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent of the present invention;

[0058] Figure 2 This is a flowchart of the text data formatting process in this invention;

[0059] Figure 3 This is a flowchart of the image data formatting process in this invention. Detailed Implementation

[0060] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0061] Example 1:

[0062] This embodiment introduces a method for diagnosing electrical equipment faults based on a vector knowledge base and an intelligent agent, such as... Figure 1 As shown, it includes:

[0063] Receive the requirement information input by maintenance personnel;

[0064] The required information is optimized by selecting prompt words to obtain semantic fragments with a uniform format, which are then used as optimized prompt words.

[0065] The optimized prompts are input into the agent;

[0066] The agent queries the vector knowledge base to determine whether similar data matching the optimized prompt words exists.

[0067] If similar data exists, the matched similar data is used as input to the large language model, and a diagnostic conclusion is generated and output based on the reasoning of the large language model.

[0068] If no similar data exists, an external method is invoked to constrain the output of the large language model, generating and outputting constrained diagnostic conclusions.

[0069] I. Optimization of prompt words

[0070] The system acquires the requirements input by operations and maintenance personnel from the dialog window. By adhering to Markdown format constraints, and employing a large-scale model optimization technique, the system refines the personnel's requirements into semantically consistent fragments, which serve as optimized prompts. These optimized prompts better align with the input habits of the large language model, effectively improving the accuracy of subsequent inference.

[0071] To prevent semantic bias from being introduced during the optimization process, this embodiment will input the requirement information. Compared with the optimized prompt words The formula for semantic similarity comparison is as follows:

[0072] ;

[0073] In the formula, Represents cosine similarity. This represents the similarity value before and after optimization;

[0074] When the similarity is higher than the preset threshold, no manual review is required; when the similarity is not higher than the preset threshold, manual review is triggered, and the optimization results are confirmed or corrected manually.

[0075] II. Data Acquisition and Preprocessing

[0076] When constructing the vector knowledge base, this invention needs to collect private data about electrical equipment within the power system, including but not limited to expert knowledge, State Grid standards, and historical fault cases. This mainly includes text data and image data, and different preprocessing methods are required.

[0077] like Figure 2 As shown, for text data such as documents, the text data is first extracted to obtain the text content from the original document; then, a sliding window is used to divide the text data into blocks, the size of which is set according to the document content and server configuration to ensure that each block contains complete semantic units; the data obtained from the blocks is vectorized to output vectors about the inherent knowledge; finally, this part of the data is stored in the Milvus database.

[0078] like Figure 3 As shown, for image data such as pictures, the image data is first processed by image inpainting, including image noise reduction, image super-resolution reconstruction, image cropping and other related techniques to improve image quality. The processed image is then processed by a pre-trained convolutional neural network to extract image feature vectors, and the generated vectors are compressed and reduced in dimensionality by algorithms such as principal component analysis to reduce storage space and computational overhead. Finally, this part of the data is stored in the Milvus database.

[0079] The steps can be uniformly described as follows, depending on the type of data:

[0080] ;

[0081] In the formula, This represents input data such as text and images, which are then processed... The model is processed and ultimately output as vector data of the same dimension. .

[0082] III. Constructing a Vector Knowledge Base

[0083] This embodiment uses Milvus as the vector knowledge base. Milvus is a database system specifically designed for vector search, which can well support the storage needs of massive amounts of data within the power system. It features low-latency queries and has an active open-source community and a mature Java ecosystem, making it better suited to the existing business architecture of the power system.

[0084] After selecting the database, the process involves storing previous expert knowledge, State Grid regulations, and historical fault cases. This content provides the inherent knowledge support for subsequent steps and also serves as the dataset for fine-tuning the large language model.

[0085] IV. Fine-tuning of the Large Language Model and Design of Inference Links

[0086] Large models typically have tens or even hundreds of gigabytes of parameters. Retraining them is not only difficult but also requires significant computational resources and time. Therefore, this embodiment uses the LoRA fine-tuning method to perform small-scale parameter training and updates for power system data. The steps are as follows:

[0087] ;

[0088] In the formula, Indicates the original model parameters. Represents the input vector. This indicates a small number of newly added parameters. This represents the output vector after fine-tuning the parameters.

[0089] ;

[0090] The original model parameters are added to the incremental parameters obtained through LoRA fine-tuning using a matrix addition operation to obtain the new fused weight parameters. This is used to replace the original model parameters, thereby improving the output. It incorporates both the learning of original knowledge and the ability to adapt to specific domain tasks, ultimately enabling the model to possess the ability to adapt to specific domains of the power system and professional reasoning capabilities.

[0091] Leveraging the reasoning and analytical capabilities of large-scale models, and based on an existing knowledge base, data analysis is performed on electrical equipment to output results such as fault prediction and fault diagnosis. The result flow is as follows:

[0092] The input data is analyzed and matched against a knowledge base. If a match is successful, the result is used as input to a larger model, which then derives better analytical conclusions as the output of the application.

[0093] If a match cannot be found, the data is directly used as input to the large model, and the result is output by the large model. An external method is called to constrain the output of the large language model, generate a constrained diagnostic conclusion, and output it. More human resources are needed here to verify the reasoning results.

[0094] Invoking external methods to constrain the output of the large language model includes:

[0095] Build a skill library, which includes script calling rules and corresponding script methods;

[0096] Based on the optimized prompt words, the matching script method is invoked to generate the corresponding script execution result;

[0097] The script execution result is used as input to a large language model. The large language model performs reasoning and output processing to generate a constrained diagnostic conclusion, expressed by the following formula:

[0098] ;

[0099] in, This indicates the optimized prompt words. This refers to the script method called from the skill library. This indicates a call to the large language model. This represents the diagnostic conclusion output by the large language model.

[0100] V. Feedback and Vector Library Iterative Optimization

[0101] After the maintenance personnel implement the solution and resolve the issues as suggested, they need to enter the feedback data into the system. The entered data can be stored in a vector database as a new set of fault cases. Alternatively, the new data can be stored in a large model fine-tuning dataset, and the large model can be lightweightly updated and fine-tuned monthly or periodically.

[0102] This invention can utilize the reasoning capabilities of vector databases and large AI models to construct a novel electrical equipment fault analysis function. This method can integrate the power system with an expert knowledge base, which largely solves the illusion problem of large models in the power system field and improves fault diagnosis efficiency.

[0103] Example 2:

[0104] Based on the same inventive concept as Embodiment 1, this embodiment introduces an electrical equipment fault diagnosis device based on a vector knowledge base and an intelligent agent, comprising:

[0105] The prompt word optimization module is used to: receive the requirement information input by maintenance personnel;

[0106] The required information is optimized by selecting prompt words to obtain semantic fragments with a uniform format, which are then used as optimized prompt words.

[0107] The fault diagnosis module is used to: input the optimized prompt words into the intelligent agent;

[0108] The agent queries the vector knowledge base to determine whether similar data matching the optimized prompt words exists.

[0109] If similar data exists, the matched similar data is used as input to the large language model, and a diagnostic conclusion is generated and output based on the reasoning of the large language model.

[0110] If no similar data exists, an external method is invoked to constrain the output of the large language model, generating and outputting constrained diagnostic conclusions.

[0111] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0112] Example 3:

[0113] Based on the same inventive concept as other embodiments, this embodiment introduces a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent as described in any of the embodiments.

[0114] Example 4:

[0115] Based on the same inventive concept as other embodiments, this embodiment introduces a computer device / apparatus / system, characterized in that it includes:

[0116] Memory, used to store computer programs / instructions;

[0117] A processor is configured to execute the computer program / instructions to implement the steps of the electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent as described in any of Embodiment 1.

[0118] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0120] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0122] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for diagnosing electrical equipment faults based on vector knowledge bases and intelligent agents, characterized in that, include: Receive the requirement information input by maintenance personnel; The required information is optimized by selecting prompt words to obtain semantic fragments with a uniform format, which are then used as optimized prompt words. The optimized prompts are input into the agent; The agent queries the vector knowledge base to determine whether similar data matching the optimized prompt words exists. If similar data exists, the matched similar data is used as input to the large language model, and a diagnostic conclusion is generated and output based on the reasoning of the large language model. If no similar data exists, an external method is invoked to constrain the output of the large language model, generating and outputting constrained diagnostic conclusions.

2. The electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent according to claim 1, characterized in that, The process of optimizing prompts for demand information is as follows: Obtain the requirements information entered by the operations and maintenance personnel from the dialog window; The requirements information input by maintenance personnel is optimized to obtain semantic fragments with consistent format; The prompt word optimization method uses rules for electrical equipment fault diagnosis to limit the optimization format of prompt words, and the rules are written using Markdown syntax; Input requirements information Compared with the optimized prompt words Semantic similarity comparison is performed to prevent semantic changes during the optimization process. The formula is expressed as: ； In the formula, Represents cosine similarity. This represents the similarity value before and after optimization; When the similarity value is higher than the threshold, no manual review is required; otherwise, manual review is required.

3. The electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent according to claim 1, characterized in that, The process of constructing the vector knowledge base includes: Collecting private data about electrical equipment within the power system; The private data is formatted and then stored in the Milvus database. The optimized prompts are matched with database content using search enhancement techniques to obtain matching results with a confidence level higher than a preset threshold.

4. The electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent according to claim 3, characterized in that, The formatting process for private data includes: For text data, the following operations are performed sequentially: original document extraction, text content extraction, block processing, vectorization, and storage. The block processing uses a sliding window to segment the text content. For image data, the following operations are performed sequentially: original image preprocessing, image inpainting, vectorization, dimensionality reduction, and storage. The image inpainting includes at least one of noise reduction processing and super-resolution processing, and the dimensionality reduction is achieved through principal component analysis.

5. The electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent according to claim 1, characterized in that, The method of invoking external methods to constrain the output of the large language model includes: Build a skill library, which includes script calling rules and corresponding script methods; Based on the optimized prompt words, the matching script method is invoked to generate the corresponding script execution result; The script execution result is used as input to a large language model. The large language model performs reasoning and output processing to generate a constrained diagnostic conclusion, expressed by the following formula: ； in, This indicates the optimized prompt words. This refers to the script method called from the skill library. This indicates a call to the large language model. This represents the diagnostic conclusion output by the large language model.

6. The electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent according to claim 1, characterized in that, This also includes fine-tuning the training of the large language model: The LoRA fine-tuning method is adopted to introduce a low-rank matrix on the basis of the original model parameters of the large language model. Only the parameters of the low-rank matrix are updated to train a large language model suitable for the field of electrical equipment fault diagnosis.

7. The electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent according to claim 1, characterized in that, After outputting the diagnostic conclusion, the diagnostic conclusion and the feedback data from the operation and maintenance personnel are stored in the vector knowledge base as newly added fault cases. According to a preset cycle, the newly added fault cases are used to iteratively optimize the large language model to obtain an optimized large language model for subsequent diagnostic tasks.

8. An electrical equipment fault diagnosis device based on vector knowledge base and intelligent agent, characterized in that, include: The prompt word optimization module is used to: receive the requirement information input by maintenance personnel; The required information is optimized by selecting prompt words to obtain semantic fragments with a uniform format, which are then used as optimized prompt words. The fault diagnosis module is used to: input the optimized prompt words into the intelligent agent; The agent queries the vector knowledge base to determine whether similar data matching the optimized prompt words exists. If similar data exists, the matched similar data is used as input to the large language model, and a diagnostic conclusion is generated and output based on the reasoning of the large language model. If no similar data exists, an external method is invoked to constrain the output of the large language model, generating and outputting constrained diagnostic conclusions.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent as described in any one of claims 1 to 7.

10. A computer device / equipment / system, characterized in that, include: Memory, used to store computer programs / instructions; A processor for executing the computer program / instructions to implement the steps of the electrical equipment fault diagnosis method based on vector knowledge base and intelligent agent as described in any one of claims 1 to 7.

Images