Improved RAG-based intelligent operation and maintenance method, apparatus and device for power transformation equipment, and medium

Through the improved RAG method and vertical field large language model, combined with spectral clustering and breadth-first search algorithm, the problem that precise solutions are difficult to provide in substation operation and maintenance is solved, the operation and maintenance efficiency is improved, and the hallucination phenomenon is reduced. Accurate and reliable business solutions.

CN119027093AActive Publication Date: 2024-11-26YELLOW RIVER CONSERVANCY TECHN INST

Patent Information

Application Number
CN202411110144.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-26
Estimated Expiration
2044-08-14

AI Technical Summary

Technical Problem

The existing technology relies on expert experience in substation operation and maintenance. The general large language model is difficult to provide accurate and reliable solutions, and there are hallucinations, which affect operation and maintenance efficiency and security.

Method used

Using the improved RAG (Retrieval-Augmented Generation) method, by constructing a large language model in vertical fields, combining spectral clustering and breadth-first search algorithms, vector similarity graphs are constructed, and accurate retrieval and reasoning of substation operation and maintenance knowledge is provided. Comprehensive and accurate solutions.

Benefits of technology

It significantly improves operation and maintenance work efficiency, reduces illusion phenomena, provides more accurate and reliable business solutions, and enhances intelligent support for operation and maintenance decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027093A_ABST
    Figure CN119027093A_ABST
Patent Text Reader

Abstract

The invention discloses a power transformation equipment intelligent operation and maintenance method and device based on an improved RAG, equipment and a medium, and the method comprises the steps: constructing different vector similarity graphs, and obtaining a vertical question and answer model; optimizing a breadth-first search algorithm; based on the power transformation operation and maintenance business problem, optimizing a breadth-first search algorithm, a vector similarity graph corresponding to the knowledge database and a vertical question and answer model, generating an answer result for the business problem, and obtaining an evaluation result used for judging whether the answer result meets a standard or not; generating a third evaluation result used for judging whether the business problem contains a fault and fault information; generating a third answer result including the cause of the fault, and generating a fourth answer result for indicating the standard related to the fault at the same time; and generating an optimal answer scheme for the business question based on all the answer results and the evaluation results, the business question and the vertical question and answer model. According to the invention, business problems in operation and maintenance can be answered more comprehensively, accurately and reliably, and the working efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of operation and maintenance of substation equipment, and specifically relates to an intelligent operation and maintenance method, device, equipment and medium for substation equipment based on improved RAG. Background Art

[0002] In the substation operation and maintenance work, there are many types of substation equipment, many inspection and maintenance items, changeable fault information, complicated operating procedures, dangerous working scenes, and artificial subjective consciousness. The professional ability of the operation and maintenance personnel is extremely high, which makes the current operation and maintenance work rely heavily on expert experience, and it takes a long time for the operation and maintenance personnel to find information. In order to reduce the work pressure of the operation and maintenance personnel and improve the operation and maintenance efficiency, most of the existing research is to design a question-and-answer system. The operation and maintenance personnel provide questions or keywords through the question-and-answer system. The question-and-answer system uses retrieval technology to find matching answers in professional databases, and feeds back the answers found word for word to the operation and maintenance personnel to assist the operation and maintenance personnel in making further decisions. The matching answers given in this form are relatively simple, and cannot be associated with more comprehensive technical issues to assist in decision-making. There is a lack of intelligent auxiliary suggestions or more reasonable solutions, and it is too dependent on the input of the operation and maintenance personnel, which is more likely to be biased, so that the question-and-answer system provides invalid answers.

[0003] The emergence of large language models has greatly promoted the development and application of intelligent question-answering systems. Compared with intelligent question-answering based on database retrieval and matching, large language model technology makes question-answering generation more in line with human language patterns. It can use the model's powerful reasoning and generation capabilities to give more human-like answers or suggestions based on questions, generate richer solutions, and further assist operation and maintenance personnel in making decisions more intelligently. However, due to the wide range, large amount, and complex logical relationships of knowledge in the field of substation operation and maintenance, the operation and maintenance question-answering system of the general large language model relies heavily on expert experience in substation operation and maintenance work, and it takes a long time for operation and maintenance personnel to find information, which cannot meet specific application scenarios and needs, resulting in deviations between the generated solutions and actual needs. For example, when dealing with problems such as fault analysis and processing, it is difficult to provide accurate and reliable solutions. At the same time, due to the inevitable existence of hallucinations in large language models, in substation operation and maintenance business scenarios, hallucinations may have a huge and catastrophic impact on personnel safety, equipment safety, and power supply reliability. Summary of the invention

[0004] In view of the above technical problems, the present invention proposes a method, device, equipment and medium for intelligent operation and maintenance of substation equipment based on improved RAG based on a large language model, which realizes the high intelligence of the substation operation and maintenance system, can accurately answer complex problems in the substation operation and maintenance process, and can also provide a variety of comprehensive and reasonable solutions through the powerful reasoning ability of the large language model, assisting the operation and maintenance personnel in decision-making, and significantly improving the efficiency of operation and maintenance work. In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0005] An intelligent operation and maintenance method for substation equipment based on improved RAG includes the following steps:

[0006] S1, construct vector similarity graphs based on the preset database for storing different substation operation and maintenance knowledge, and obtain a vertical question-answering model based on the technical question-answering database and the large language model;

[0007] The databases of different substation operation and maintenance knowledge include knowledge databases, fault databases and standard databases, and the technical question and answer database belongs to the knowledge database;

[0008] S2, improving the breadth-first search algorithm based on the spectral clustering method, and obtaining a spectral clustering BFS search algorithm for performing similar text retrieval on the vector similarity graph in step S1 based on the retrieval vector;

[0009] S3, collect business questions, and use the vertical question-answering model obtained in step S1 to determine whether the business question is a substation operation and maintenance business. If so, execute step S4, otherwise re-collect business questions;

[0010] S4, using the business question obtained in step S3 as a retrieval vector and using the algorithm in step S2 to traverse the vector similarity graph corresponding to the knowledge database, generating an answer result for the business question based on the traversal result, the business question and the vertical question-answering model, using the answer result as a retrieval vector to search the vector similarity graph corresponding to the standard database, and obtaining an evaluation result for determining whether the answer result meets the standard in the standard database based on the retrieval result, the answer result and the vertical question-answering model;

[0011] S5, based on the business question, the answer result obtained in step S4 and the vertical question-answering model, a third evaluation result is generated to determine whether the business question contains a fault and fault information, and whether the device has a fault is determined according to the third evaluation result. If not, return to step S3, otherwise execute step S6;

[0012] S6, using the third evaluation result as a retrieval vector and using the algorithm of step S2 to search the vector similarity graph corresponding to the fault database, generating a third answer result including the cause of the fault based on the search result, the third evaluation result and the vertical question-answering model, and generating a fourth answer result indicating the standard involved in the fault based on the search result, the business question, the third evaluation result and the vertical question-answering model;

[0013] S7, based on all the answer results and evaluation results obtained in step S4, step S5 and step S6, the business problem and the vertical question-answering model, generates an optimal solution for the business problem based on the knowledge database to provide relevant knowledge, determine whether the equipment is faulty and the cause of the fault, and indicate the standards involved in the fault.

[0014] In step S1, the method for constructing the vector similarity graph includes the following steps:

[0015] ①, pre-process the data in each database in step S1 to obtain the corresponding text database;

[0016] ②, segment and vectorize the text database obtained in step ① to obtain corresponding vector feature data;

[0017] ③. Use the similarity calculation method to calculate the similarity between any two vector feature data of each text database in step ②. Take the vector feature data as nodes and construct undirected edges between the nodes based on the preset first similarity threshold to obtain the corresponding vector similarity graph. At the same time, construct edge weights based on the similarity between the two nodes corresponding to the undirected edges.

[0018] The step S2 comprises the following steps:

[0019] i, calculate the normalized Laplacian matrix based on the degree matrix and similarity matrix of the vector similarity graph;

[0020] ii. Perform eigendecomposition on the normalized Laplace matrix to obtain the first eigenvalue and the first eigenvector, and use the K-means algorithm to cluster the first eigenvector and divide it into several clusters;

[0021] iii. Calculate the similarity between the search vector and the cluster center of each cluster and each node in the cluster using a similarity calculation method, and use a breadth-first search algorithm based on a second similarity threshold to search the vector similarity graph to generate similar texts.

[0022] The knowledge database includes a technical question and answer database and a theoretical knowledge database.

[0023] An intelligent operation and maintenance system for substation equipment based on improved RAG, comprising:

[0024] User interaction module: used to collect business issues;

[0025] Professional knowledge base construction module: used to collect and classify professional knowledge data in the field of substation operation and maintenance, and construct different vector similarity graphs based on the classified professional knowledge data;

[0026] Vertical large model building module: used to build a vertical question-answering model based on the classified technical question-answering database obtained by the large language model and professional knowledge base building module;

[0027] Search algorithm building module: used to generate a spectral clustering BFS search algorithm based on the spectral clustering method to improve the breadth-first search algorithm and obtain a spectral clustering BFS search algorithm for similar text retrieval based on the retrieval vector for the vector similarity graph obtained by the professional knowledge base building module;

[0028] Knowledge reasoning module: used to use the business questions collected by the user interaction module as retrieval vectors to traverse the vector similarity graph corresponding to the knowledge database using the algorithm of the search algorithm building module, generate answer results for the business questions based on the traversal results, business questions and vertical question-answering models, use the answer results as retrieval vectors to search the vector similarity graph corresponding to the standard database, and obtain an evaluation result for determining whether the answer result meets the standards in the standard database based on the retrieval results, the answer results and the vertical question-answering model;

[0029] Fault reasoning module: used to generate a third evaluation result for determining whether the business question contains a fault and corresponding fault information based on the business question, all answer results obtained by the knowledge reasoning module and the vertical question-answering model, and determine whether the device has a fault according to the third evaluation result. If not, return to step S3, otherwise execute step S6;

[0030] Standard output module: used to use the third evaluation result output by the fault reasoning module as a retrieval vector to search the vector similarity graph corresponding to the fault database using the algorithm of the search algorithm building module, and generate a third answer result including the cause of the fault based on the search result, the third evaluation result and the vertical question-answering model, and generate a fourth answer result for indicating the standard involved in the fault based on the search result, the business questions collected by the user interaction module, the third evaluation result and the vertical question-answering model;

[0031] Optimal answer generation module: used to generate the optimal answer solution for the business problem based on the knowledge database to provide corresponding knowledge, determine whether the equipment is faulty and the cause of the fault, and indicate the standards involved in the fault based on all answer results and evaluation results obtained by the knowledge reasoning module, fault reasoning module and standard output module, business problems collected by the user interaction module and the vertical question and answer model.

[0032] An electronic device comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the intelligent operation and maintenance method of substation equipment based on improved RAG as described above is implemented.

[0033] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent operation and maintenance method for substation equipment based on improved RAG as described above.

[0034] Beneficial effects of the present invention:

[0035] 1. Use a large amount of professional domain knowledge to fine-tune the general large model and perform retrieval augmented generation (RAG) to build a large language model in a vertical field. This can alleviate the illusion phenomenon in the large model reasoning process, make the business solution more comprehensive and accurate, and improve the work efficiency in the business field;

[0036] 2. Use self-assessment prompt optimization algorithm to provide more comprehensive, accurate and reliable answers to business problems in substation operation and maintenance work, give more intelligent solutions, and improve work efficiency in the business field. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 It is a schematic diagram of the module structure of the present invention.

[0039] Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Although the operation and maintenance systems using large language models in the existing technology have improved intelligence, they lack professional expert experience and are difficult to provide accurate and comprehensive answers. In addition, in highly professional questions, such as "What should be done when a 220kV or higher voltage transformer is hit by a short circuit in the nearby area and does not trip?", the general knowledge of the large language model is superficial and cannot give accurate values. It may even cause hallucinations, give false answers, mislead operation and maintenance personnel, and cause inevitable losses. Therefore, the current technical solutions cannot provide comprehensive and effective intelligent support in the highly professional substation operation and maintenance work.

[0042] This application improves and upgrades the substation equipment operation and maintenance system. On the basis of the traditional or large language model operation and maintenance system, it adds a professional knowledge base to build a vertical question-answering model, realizes system question-answering through images, sounds and texts, and assists in decision-making. At the same time, through the powerful professional reasoning ability of the vertical question-answering model, it provides more intelligent solutions for various professional problems that arise during the operation and maintenance process, such as "What should I do if a light gas alarm signal appears on phase C of the transformer, the heavy gas protection trips, and the electrical quantity protection does not work?", which significantly improves work efficiency.

[0043] This embodiment provides a method for intelligent operation and maintenance of substation equipment based on improved RAG, including the following steps:

[0044] S1, collect and classify professional knowledge data in the field of substation operation and maintenance, obtain a vertical question-answering model based on the classified professional knowledge data and the question-answering model, and construct different types of vector similarity graphs based on the classified professional knowledge data;

[0045] This application first categorizes the professional knowledge data in the field of substation operation and maintenance, including knowledge database, fault database and standard database. In specific operations, each type of database can be further divided according to actual conditions. In this embodiment, the knowledge database can also be specifically divided into a technical question and answer database, a theoretical knowledge database and a practical skills database. The fault database refers to a fault case database, and the standard database refers to a standard specification and strong clause database. Among them, a vertical question and answer model is obtained based on the data and question and answer model in the technical question and answer database.

[0046] The technical question-answer database includes several question-answer pairs, each of which, as a piece of data, includes questions and corresponding answers in the substation operation and maintenance process. For example, the question is: What are the basic contents of substation switching operations? The answers are: (1) Line shutdown and power supply operations; (2) Transformer shutdown and power supply operations; (3) Busbar reversal and busbar power supply and shutdown operations; (4) Installation and removal of grounding wires (closing and opening grounding switches); (5) Grid paralleling and decoupling operations; (6) Transformer voltage regulation operations; (7) Station power supply switching operations. (8) Relay protection and automatic device activation and deactivation operations, and operations to change the set values ​​of relay protection and automatic devices; (9) Other special operations.

[0047] The theoretical knowledge database includes AC and DC power supply systems, anti-error operation and management, online monitoring system for dissolved gas in oil, construction and management of smart substations and other substation operation and maintenance theoretical knowledge, such as: general technical principles of anti-error logic, specific requirements for circuit breakers, disconnectors, circuit breaker trolleys, grounding switches, and network (cabinet) doors, among which, (1) when the circuit breaker trolley is switched from "working", "testing" to "maintenance", the circuit breaker in this interval should be in the open position. (2) When the circuit breaker trolley is turned to the "working" position, the grounding switches on both sides should be in the open position and the grounding wires should be removed, including the grounding switches and grounding wires connected through the main transformer, grounding transformer, station transformer, capacitor, busbar, cable, etc., and the rear cabinet door should be closed.

[0048] The practical skills database includes substation switching operation, oil-immersed transformer sampling and diagnostic analysis, GIS gas chamber diagnostic analysis, GIS ultrasonic partial discharge detection technology, GIS ultra-high frequency partial discharge detection technology, equipment precise temperature measurement and other substation operation and maintenance practical skills. For example, during the operation ticket filling stage, the following items should be filled in the operation ticket: (1) Equipment that should be pulled in and out [circuit breakers (switches), disconnectors (knife switches), grounding switches (devices), etc.], electricity testing, installation and removal of grounding wires, closing (installing) or disconnecting (removing) air circuit breakers (switches) and fuses of control circuits or voltage transformer circuits, switching protection circuits and automation devices, and checking whether there is no voltage, etc. (2) Check the position of the equipment after pulling in and out the equipment [circuit breakers (switches), disconnectors (knife switches), grounding switches (devices), etc.]. (3) When performing power-off and power-on operations, before pulling in and out the disconnectors (knife switches) or pulling out and pushing in the trolley-type circuit breakers (switches), check whether the circuit breakers (switches) are in the open position. (4) Before and after reversing loads or performing disconnection and paralleling operations, check the operation of related power supplies and load distribution. (5) Before closing the switch and supplying power after equipment maintenance, check that the grounding switch (device) within the power supply range has been opened and the grounding wire has been removed.

[0049] The fault case database includes typical cases of substation operation and maintenance such as electrical misoperation accidents, substation shutdown accidents, large transformer accidents, reactive power compensation accidents, transformer accidents, GIS and switchgear accidents, and relay protection accidents. For example, the accident of "switching operation in the wrong interval, mistakenly closing the grounding switch under power" involves the following clauses: (1) Before operation, the system mode, equipment name, number and location should be checked. (Q / GDW1799.1-2013 "State Grid Corporation Electric Power Safety Work Regulations Substation Part" 5.3.6.2) (2) The operator should clarify the purpose and sequence of operation, analyze the possible dangerous points during the operation and take corresponding measures. Read, repeat, monitor and operate the operation ticket item by item, confirm that the equipment status is consistent with the content of the operation ticket and tick it. (State Grid Safety Supervision

[2018] No. 1119 "Safety Management Regulations for Preventing Electrical Misoperation" 3.1.3) (3) The switching operation should be officially issued by the on-duty control personnel and the operation and maintenance person in charge, and use the operation ticket that has been reviewed and approved in advance, and operate item by item according to the order filled in the operation ticket. (Article 66 of the State Grid Corporation of China Substation Operation and Maintenance Management Regulations). The response measures to the accident are: (1) Strictly implement the switching operation management regulations, follow the formal control instructions, use the operation ticket that has been approved in advance, and the guardian and operator shall operate in sequence. Non-operating personnel and on-site inspection personnel are not allowed to perform operations. (2) Fully implement the "five prevention" management regulations, strictly implement the unlocking key management system, and strengthen the management of the unlocking key use approval process. (3) Comprehensively check the unlocking password settings of the computer key debugging function, and strictly prohibit the use of unlocking passwords. Strengthen software program management, require suppliers to cancel the debugging mode, block the unlocking function, and trace and evaluate the responsibility for illegal use.

[0050] The standard specifications and mandatory database include the State Grid Corporation of China's substation operation and maintenance management regulations and twenty-eight types of equipment operation and maintenance details (oil-immersed transformer operation and maintenance details, circuit breaker operation and maintenance details, combined electrical appliance operation and maintenance details, etc.), the State Grid Corporation of China's eighteen major grid anti-accident measures, the State Grid Corporation of China's power safety work regulations substation part and other national or industry standards and specifications, such as: Circuit breaker emergency application shutdown regulations "1.5.1 If any of the following situations are found during operation, the operation and maintenance personnel shall immediately report to the control personnel to apply for equipment shutdown, and stay away from the equipment before shutdown. a) The bushing is seriously damaged a) There is severe overheating or sparking in the conductive circuit components. c) There is severe gas leakage in the SF6 circuit breaker, and an operation lockout signal is issued. d) There is smoke in the arc extinguishing chamber of the low-oil circuit breaker or abnormal noise inside. e) There is severe oil leakage in the low-oil circuit breaker, and the oil level is not visible. f) There is a bursting sound inside the high-oil circuit breaker. g) There are abnormal phenomena such as cracks or discharge sounds in the arc extinguishing chamber of the vacuum circuit breaker. h) The explosion-proof membrane of the floor-standing tank circuit breaker is deformed or damaged. i) The hydraulic and pneumatic operating mechanisms lose pressure, and the energy storage spring of the energy storage mechanism is damaged. j) Other situations that should be urgently stopped according to the actual situation on site. "

[0051] The question-answering model is based on large language models such as LLaMA3, ChatGLM and other open source models. At present, the question-answering models in substation operation and maintenance are mainly based on traditional database retrieval, and a few use general large language models to enhance the intelligence of the system. However, the general large language models lack a lot of professional knowledge in the field of electricity during the training process, which makes the generated solutions deviate from the actual needs, resulting in serious hallucinations, and it is difficult to provide accurate and reliable solutions.

[0052] When fine-tuning the vertical question-answering model in this application, the data in the technical question-answering database is randomly divided into a training set and a test set. The training set is used to fine-tune the training model, and the test set is used to evaluate the performance of the fine-tuned model. In the later specific application process, new technical question-answering data can be continuously collected based on user feedback and business problems to further fine-tune and improve the vertical question-answering model.

[0053] The steps of constructing different types of vector similarity graphs based on the classified professional knowledge data include:

[0054] ①, build a corresponding text database based on the classified professional knowledge data;

[0055] After collecting professional knowledge data, all data can be pre-processed according to business needs, such as data cleaning, word segmentation and tokenization, to generate a text database. The above are all prior arts, and this embodiment will not be described in detail.

[0056] ②, use the Word2Vec model to vectorize the data preprocessed in step ① to obtain several vector feature data;

[0057] The preprocessed data is vectorized using Word2Vec. Specifically, vector feature data may be generated based on a paragraph, a chapter, or a sentence in the text database. In this embodiment, a vector feature data is generated based on a paragraph.

[0058] ③, calculating the similarity between any two vector feature data of each text database in step ②, taking the vector feature data as nodes, constructing undirected edges between the nodes based on a preset first similarity threshold to obtain a corresponding vector similarity graph, and constructing edge weights based on the similarity between the two nodes corresponding to the undirected edges;

[0059] When constructing a vector similarity graph, you can construct a respective vector similarity graph based on each text database, or you can combine two or more text databases according to actual needs, and construct a vector similarity graph based on the combined text database. This embodiment constructs a first vector similarity graph based on a technical question and answer text database for priority retrieval of business issues; constructs a second vector similarity graph based on a text database after combining theoretical knowledge and practical skills, which is used as a supplementary retrieval of the first vector similarity graph; constructs a third vector similarity graph based on a fault case text database to assist in solving fault cause analysis in operation and maintenance; constructs a fourth vector similarity graph based on a standard specification and strong text database for answer result evaluation. This application is based on the reasoning framework of the Tree of Thoughts (ToT), and uses LLM to simulate the human decision-making process, so that the answers obtained are more in line with the way humans think. Compared with directly searching once to obtain search results, multiple searches through the thinking tree can obtain more comprehensive answers.

[0060] The vector similarity graph (VSG) uses G n Indicates that G n = {V n ,E n},V n represents the set of nodes in the nth vector similarity graph, E n Represents the set of edges in the nth vector similarity graph.

[0061] In order to improve the retrieval efficiency, after calculating the similarity between vector feature data using the similarity calculation method, the edges with lower similarity are filtered based on the first similarity threshold γ. Based on this method, all vector similarity graphs are updated and optimized. The set of edges after the nth vector similarity graph is updated using E′ n The corresponding expression is:

[0062] E′ n ={(v n,m ,v n,m′ )∈E n |sim(v n,m ,v n,m′ )≥θ}; (1)

[0063] Where sim(·) represents the similarity function. In this embodiment, cosine similarity is used for measurement, θ=0.7, and m≠m′, v n,m represents the mth node in the nth vector similarity graph, v n,m′ represents the m′th node in the nth vector similarity graph, v n,m ∈V n , v n,m′ ∈V n .

[0064] S2, improving the breadth-first search (BFS) algorithm based on the spectral clustering method to obtain a spectral clustering BFS search algorithm for performing similar text retrieval on the vector similarity graph obtained in step S1 based on the retrieval vector;

[0065] The breadth-first search algorithm (BFS) is used to traverse the vector similarity graph, and vector segments with high similarity are retrieved as candidate results, i.e., similar texts, based on the second similarity threshold θ. When a vector segment with a similarity lower than the second similarity threshold θ is encountered during the traversal process, further searching for neighbor nodes is stopped.

[0066] When BFS traverses the first vector similarity graph, it needs to traverse multiple nodes and edges, and the search efficiency is low. This application innovatively proposes a spectral clustering BFS search algorithm, which uses spectral clustering in BFS to classify the first vector similarity graph, thereby guiding the search process, making the search more efficient, and adopts a pruning strategy to effectively limit the search range and reduce the search space. The specific process of the algorithm is as follows:

[0067] i, calculate the normalized Laplace matrix L according to the degree matrix and similarity matrix of the vector similarity graph;

[0068] The calculation formula of the normalized Laplace matrix L is:

[0069]

[0070] Where I represents the identity matrix, the elements on the diagonal are all 1, and the remaining elements are all 0, D represents the degree matrix, and W represents the similarity matrix.

[0071] Each element on the diagonal of the degree matrix D ij The calculation formula is:

[0072]

[0073] Where N represents the total number of nodes in the first vector similarity graph, W ij represents the value of the i-th row and j-th column in the similarity matrix W. The calculation method of the similarity matrix and the similarity matrix is ​​a prior art and will not be described in detail in this embodiment.

[0074] ii. Perform eigendecomposition on the normalized Laplace matrix L to obtain the first eigenvalue and the first eigenvector, and use the K-means algorithm to cluster the first eigenvector and divide it into several clusters;

[0075] The first eigenvector is clustered using the K-means algorithm into k clusters (k=5 in this embodiment), the cluster information of each cluster is saved, and the average value of all edge weights in the cluster is used as the cluster center.

[0076] The expression of the eigendecomposition of the normalized Laplace matrix L is:

[0077] L=UΛU T ; (4)

[0078] Where U is an orthogonal matrix whose columns are the first eigenvectors of L, and Λ is a diagonal matrix whose diagonal elements are the first eigenvalues ​​of L.

[0079] iii, calculating the similarity between the retrieval vector and the cluster center and nodes of each cluster, and based on the second similarity threshold θ, using the breadth-first search (BFS) algorithm to retrieve the vector similarity graph to generate similar texts;

[0080] When traversing, first calculate the similarity between the search vectors such as business questions, answer results, etc. and each cluster center. If it is lower than the second similarity threshold, no further search will be performed on the cluster. Otherwise, the cluster will be searched starting from the cluster center. When searching, when the similarity between a node and the business question is lower than the second similarity threshold, the subsequent nodes of the node are pruned, and its second-order neighboring nodes are pruned, further reducing the search complexity. Here, the second similarity threshold θ = 0.7, and the first k′ vectors with the highest similarity in each cluster are selected as similar texts, and k′ is a positive integer.

[0081] S3, collect business questions from operation and maintenance personnel, and use the vertical question-answering model obtained in step S1 to determine whether the business question is a substation operation and maintenance business. If so, execute step S4, otherwise re-collect business questions;

[0082] The type of the business question can be an image, sound or text. If the business question is an image or sound, it is necessary to use a multimodal model to process and convert it into text before inputting the business question into the vertical question-answering model. If the business question is text, the text can be directly input into the vertical question-answering model. When processing images, the multimodal model is an image-to-text model, and when processing sounds, the multimodal model is a speech-to-text model. The image-to-text model includes CLIP-ViL, ViT-BERT, SimVLM, BLIP, Florence, L-Verse, Oscar, VinVL, MURAL, CoCa, etc., and the speech-to-text model includes convolutional neural networks (VGG, GoogLeNet, ResNet), recurrent neural networks (LSTM, GRU), adversarial generative networks (GAN), Transformer networks, attention mechanisms, etc. The above is the prior art, and this embodiment will not describe the process of requesting a text sequence in detail.

[0083] Operation and maintenance personnel can ask business questions to the vertical question-answering model through the user interface and obtain business solutions. The solutions generated by the vertical question-answering model can be text, or converted into images or sounds and returned to the user.

[0084] S4, using the business question obtained in step S3 as a retrieval vector and using the algorithm in step S2 to traverse the vector similarity graph corresponding to the knowledge database, generating an answer result for the business question based on the traversal result, the business question and the vertical question-answering model, using the answer result as a retrieval vector to search the vector similarity graph corresponding to the standard database, and obtaining an evaluation result for determining whether the answer result meets the standard in the standard database based on the retrieval result, the answer result and the vertical question-answering model;

[0085] In this embodiment, the vector similarity graph corresponding to the vector similarity graph corresponding to the knowledge database is a first vector similarity graph and a second vector similarity graph. Specifically, when executing this step, step S4 includes:

[0086] a. Using the method of step S2, the first vector similarity graph is traversed and screened based on the business question obtained in step S3 to obtain the first similar text, the first similar text and the business question are filled into the first preset prompt, and the filled first preset prompt is input into the vertical question-answering model to generate the first answer result, using the method of step S2, the fourth vector similarity graph is traversed and screened based on the first answer result to obtain the first evaluation text, the first answer result and the first evaluation text are filled into the second preset prompt, and the filled second preset prompt is input into the vertical question-answering model to obtain the first evaluation result used to determine whether the first answer result meets the standard in the standard database. If not, execute step b, otherwise, execute step S5;

[0087] The first preset prompt and the second preset prompt are both prompt words. The first preset prompt is used to prompt the vertical question and answer model to obtain the corresponding answer based on the fill-in body, and the second preset prompt is used to prompt the vertical question and answer model to evaluate whether the answer result meets the correctness of the standard.

[0088] In step a, the business question and the first answer result obtained in step S3 are used as retrieval vectors to search and traverse the first vector similarity graph and the fourth vector similarity graph, respectively, to generate two traversal results, namely the first similar text and the first evaluation text. Specifically, firstly, the method of step S2 is used to traverse and filter the first vector similarity graph based on the business question, namely question Q, to obtain the first similar text, namely text fragment 1-1, and then fill the text fragment 1-1 and question Q into the first preset prompt to obtain prompt1-1. Prompt1-1 is: [Answer the question based on {question Q}, {text fragment 1-1}, and do not add other sentences. ], and then prompt1-1 is input into the vertical question-answering model to obtain the first answer result, namely Figure 2 Answer 1-1 in .

[0089] Then, the method of step S2 is used to traverse and filter the fourth vector similarity graph based on answer 1-1 to obtain the first evaluation text, that is, text fragment 1-2. Answer 1-1 and text fragment 1-2 are filled into the second preset prompt to obtain prompt2-1, and prompt2-1 is: [For {answer 1-1}, please analyze it from the aspects of rationality and standardization according to {text fragment 1-2}. If rationality and standardization are met, please return 1; if rationality and standardization are not met, please return 0. ]. Finally, prompt2-1 is input into the vertical question-answering model for evaluation to obtain the first evaluation result, that is, answer 1-2.

[0090] Text fragment 1-1 is the answer to the business question in the technical question and answer database. The answer is input into the vertical question and answer model together with the business question, which can assist in guiding the generation of the answer result, thereby improving the quality and accuracy of the answer result. Based on the answer 1-1, the fourth vector similarity graph can be traversed to judge and evaluate whether the answer result meets the national or industry standards according to the relevant data in the standard specification and strong text database, and assist the operation and maintenance personnel to determine the feasibility of the answer result. When the answer 1-2 is zero, it means that the answer 1-1 cannot be used as a reference solution to the business problem, and further retrieval of the theoretical knowledge text database is required.

[0091] b. Using the method of step S2, traverse and filter the second vector similarity graph based on the business question obtained in step S3 to obtain the second similar text, fill the second similar text and the business question into the first preset prompt, input the filled first preset prompt into the vertical question-answering model to generate a second answer result, using the method of step S2, traverse and filter the fourth vector similarity graph based on the second answer result to obtain the second evaluation text, fill the second answer result and the second evaluation text into the second preset prompt, input the filled second preset prompt into the vertical question-answering model to obtain a second evaluation result for determining whether the second answer result meets the standard in the standard database, if not, return to step S2, otherwise execute step S5;

[0092] In step b, the business question and the second answer result are used as retrieval vectors to search and traverse the second vector similarity graph and the fourth vector similarity graph, respectively, to generate two retrieval results, namely the second similar text and the second evaluation text. Specifically, the method of step S2 is used to traverse and filter the second vector similarity graph based on the business question to obtain the second similar text, that is, the text segment 2-1, and the text segment 2-1 and the question Q are filled into the first preset prompt to obtain prompt1-2, and prompt1-2 is: [Answer the question according to {question Q}, {text segment 2-1}, and do not add other sentences. ]. Then prompt1-2 is input into the vertical question-answering model to obtain the second answer result, that is, answer 2-1.

[0093] Using the method of step S2, the fourth vector similarity graph is traversed and screened based on answer 2-1 to obtain the second evaluation text, that is, text fragment 2-2. Answer 2-1 and text fragment 2-2 are filled into the second preset prompt to obtain prompt2-2, and prompt2-2 is: [For {answer 2-1}, please analyze it from the aspects of rationality and standardization according to {text fragment 2-2}. If rationality and standardization are met, please return 1; if rationality and standardization are not met, please return 0. ]. Finally, prompt2-2 is input into the vertical question-answering model for evaluation to obtain the second evaluation result, that is, answer 2-2.

[0094] Text fragment 2-1 retrieves relevant knowledge about business problems from a text database that is a combination of theoretical knowledge and practical skills. This knowledge retrieval is input into the vertical question-and-answer model together with the business problem, which can assist in guiding the generation of answer results, thereby improving the quality and accuracy of the answer results. Based on answer 2-1, traversing the fourth vector similarity graph can determine whether the second answer result complies with national or industry standards based on standard specifications and relevant data in the strong text database, helping operation and maintenance personnel to determine the feasibility of the answer results. When answer 2-2 is zero, it means that neither the technical question-and-answer text database nor the text database that is a combination of theoretical knowledge and practical skills can find answers and knowledge related to the business problem, and the operation and maintenance personnel need to consider solving the problem from other directions.

[0095] S5, fill the business question, the first answer result or the second answer result into the third preset prompt, input the filled third preset prompt into the vertical question-answering model to obtain a third evaluation result for determining whether the business question contains a fault and corresponding fault information, determine whether the device has a fault according to the third evaluation result, if not, return to step S3, otherwise output the third evaluation result and execute step S6;

[0096] Fill the business question, the first answer result or the second answer result into the third preset prompt to obtain prompt3-1, which is: [Please analyze whether {question Q}, {answer 1-1} or {answer 2-1} contain fault or defect phenomena, accident information. If they do, please return 1 and output the fault or defect phenomena, accident information, and do not add other statements. If they do not, please return 0. ]. Input prompt3-1 into the vertical question-answering model to obtain the third evaluation result, which is answer 3-1.

[0097] The vertical question-answering model generates answer 3-1 based on the business question and answer 1-1 / answer 2-1, which can assist the operation and maintenance personnel in determining the fault or defect status of the currently described equipment. When answer 3-1 is zero, it means that the business question does not contain fault information.

[0098] S6, using the method of step S2 to traverse the third vector similarity graph based on the fault information in the third evaluation result obtained in step S5 to obtain a third similar text, filling the business question, the third evaluation result, and the third similar text into a fourth preset prompt, inputting the filled fourth preset prompt into the vertical question-answering model to obtain a third answer result, filling the business question, the third evaluation result, and the third similar text into a fifth preset prompt, and inputting the filled fifth preset prompt into the vertical question-answering model to obtain a fourth answer result;

[0099] In step S6, the fault information is used as a search vector to search and traverse the third vector similarity graph, and a third similar text is generated as the search result. Specifically, the method of step S2 is used to traverse the third vector similarity graph based on the fault information obtained in step S5 to obtain the third similar text, that is, text segment 4-1, and question Q, answer 3-1, and text segment 4-1 are filled into the fourth preset prompt to obtain prompt4-1, prompt4-1 is: [For {question Q}, {answer 3-1}, please analyze the possible causes of the fault or defect phenomenon in question Q based on {text segment 4-1}, and provide treatment measures. Do not add other sentences. ]. Input prompt4-1 into the vertical question-answering model to obtain the third answer result, that is, answer 4-1.

[0100] Fill question Q, answer 3-1, and text segment 4-1 into the fifth preset prompt to obtain prompt 5-1, which is: [For {question Q}, {answer 3-1}, please analyze the clauses involved in the occurrence of the fault or defect phenomenon in question Q based on {text segment 4-1}, and give countermeasures. Do not add other sentences. ]. Input prompt 5-1 into the vertical question-answering model to obtain the fourth answer result, that is, answer 5-1.

[0101] Based on answer 3-1, traversing the third vector similarity graph can determine whether there are similar fault cases before. Based on question Q, answer 3-1, and text fragment 4-1, the vertical question-answering model is input to obtain answer 4-1. The large model can output the cause of the fault and the corresponding countermeasures, providing a reliable solution for operation and maintenance personnel.

[0102] S7, filling the business question, the first answer result or the second answer result, the third answer result, the fourth answer result, the first evaluation result or the second evaluation result into the sixth preset prompt, and inputting the filled sixth preset prompt into the vertical question-answering model to obtain the optimal business solution corresponding to the business question;

[0103] Fill question Q, answer 1-1 or answer 2-1, answer 1-2 or answer 2-2, answer 4-1, answer 5-1 into the sixth preset prompt to obtain prompt 6-1, which is:

[0104] [Based on {Question Q}, {Answer 1-1} or {Answer 2-1}, {Answer 1-2} or {Answer 2-2}, {Answer 4-1}, {Answer 5-1}, answer business question Q. Please answer strictly according to the information provided and do not add other statements;

[0105] If {Answer 2-2} contains "0", please output "No relevant information found";

[0106] If {Answer 1-2} or {Answer 2-2} contains "1", and {Answer 3-1} contains "0", please output it in the format of question and answer according to the content of {Question Q}, {Answer 1-1} or {Answer 2-1};

[0107] If "1" is included in {Answer 1-2} or {Answer 2-2}, and "1" is included in {Answer 3-1}, please output the content of {Question Q}, {Answer 1-1} or {Answer 2-1}, {Answer 1-2} or {Answer 2-2}, {Answer 4-1}, and {Answer 5-1} in the format of question, answer, phenomenon, possible cause, treatment measures, involved clauses, and countermeasures. ].

[0108] In addition, the existing large language model can already analyze the possible causes of the failure or defect phenomenon in the problem, give treatment measures, as well as the clauses involved in the occurrence of the failure or defect phenomenon, and give countermeasures. However, the answers he gives are not comprehensive and accurate, and multiple rounds of questions and answers are required to obtain a solution. The vertical question-and-answer model of this application can provide a professional solution at one time through the self-evaluation prompt optimization algorithm. Based on the output of the optimal business solution, the operation and maintenance personnel can obtain the corresponding knowledge output based on the knowledge database, whether the equipment is faulty and the cause of the failure, the standards involved in the failure, and other related content, such as the following business problems and their optimal business solutions:

[0109] Business question: What are the reasons for the main transformer’s heavy gas, pressure release and differential protection tripping?

[0110] Optimal business solution (including phenomenon, possible causes, treatment measures, involved clauses and / or countermeasures):

[0111] Phenomenon: The main transformer is tripped due to heavy gas, pressure release and differential protection.

[0112] Possible cause: Foreign matter in the transformer high-voltage bushing causes a short circuit fault in the vicinity.

[0113] Treatment measures: a) Check the transformer bushing burst. b) After confirming that the circuit breakers on both sides of the transformer have tripped, check that the fan and submersible oil pump of the oil-cooled transformer have stopped running. c) Carefully check and verify the transformer differential protection action information, and at the same time check the protection action signals of other equipment, the primary and secondary circuits, the DC power supply system and the operation of the station power system. d) If the station power system loses power, normal power supply should be restored as soon as possible. e) According to the dispatching instructions or the "Special Regulations for Substation On-site Operation", adjust the transformer neutral point operation mode. f) Check whether the operating transformer is overloaded and put into the cooler according to the load situation. If the transformer is overloaded, report to the on-duty control personnel to transfer the load. g) Check the operation of the standby automatic device. If the standby automatic device operates correctly, exit the automatic device of the equipment according to the dispatching instructions. If the standby automatic device does not operate correctly, check that the circuit breaker acting on the standby automatic device is ready, exit the standby power supply automatic device according to the dispatching instructions, and immediately close the circuit breaker acted on by the standby automatic device after the action, and restore the load on the power-lost bus. h) Rapidly isolate the faulty transformer. i) Record and report the transformer protection action time and primary and secondary equipment inspection results. j) Arrange safety measures for the maintenance and testing of the faulty transformer in advance.

[0114] Clauses involved: 5.1.3.4 of the 18 Major Anti-Accident Measures of State Grid Corporation of China (2018 Revised Edition) and Preparation Instructions: "Regularly clean up floating objects, plastic greenhouses, color steel plate buildings, kites and tall trees in and around the substation. Special inspections should be carried out before and after strong winds to prevent foreign objects from floating and causing equipment short circuits."

[0115] Countermeasures: State Grid Corporation of China (SGCC) Notice on Issuing Sixteen Measures (Trial) to Prevent Substation Shutdown (2015) No. 376, 12.3 "Inflammable and explosive items in and around substations should be cleaned regularly to prevent accidents. Floating objects, plastic greenhouses, color steel plate buildings, kites, etc. in and around substations should be cleaned regularly. Special inspections should be carried out before strong winds to prevent foreign objects from floating and causing equipment short circuits. Burning of wasteland, burning of straw, blasting operations, dust emissions, etc. near substations should be stopped in a timely manner to prevent dust pollution from causing short circuits."

[0116] The embodiment of the present application also provides an intelligent operation and maintenance device for substation equipment based on improved RAG, including:

[0117] User interaction module: used to input business problems and process them to generate data that can be recognized by the big model;

[0118] Professional knowledge base construction module: used to collect and classify professional knowledge data in the field of substation operation and maintenance, and construct different vector similarity graphs based on the classified professional knowledge data;

[0119] Vertical large model building module: used to build a vertical question-answering model based on the classified technical question-answering database obtained by the large language model and professional knowledge base building module;

[0120] Search algorithm building module: used to generate a spectral clustering BFS search algorithm based on the spectral clustering method to improve the breadth-first search algorithm and obtain a spectral clustering BFS search algorithm for similar text retrieval based on the retrieval vector for the vector similarity graph obtained by the professional knowledge base building module;

[0121] Knowledge reasoning module: used to use the business questions collected by the user interaction module as retrieval vectors to traverse the vector similarity graph corresponding to the knowledge database using the algorithm of the search algorithm building module, generate answer results for the business questions based on the traversal results, business questions and vertical question-answering models, use the answer results as retrieval vectors to search the vector similarity graph corresponding to the standard database, and obtain an evaluation result for determining whether the answer result meets the standards in the standard database based on the retrieval results, the answer results and the vertical question-answering model;

[0122] Fault reasoning module: used to generate a third evaluation result for determining whether the business question contains a fault and corresponding fault information based on the business question, all answer results obtained by the knowledge reasoning module and the vertical question-answering model, and determine whether the device has a fault according to the third evaluation result. If not, return to step S3, otherwise execute step S6;

[0123] Standard output module: used to use the third evaluation result output by the fault reasoning module as a retrieval vector to search the vector similarity graph corresponding to the fault database using the algorithm of the search algorithm building module, and generate a third answer result including the cause of the fault based on the search result, the third evaluation result and the vertical question-answering model, and generate a fourth answer result for indicating the standard involved in the fault based on the search result, the business questions collected by the user interaction module, the third evaluation result and the vertical question-answering model;

[0124] Optimal answer generation module: used to generate the optimal answer solution for the business problem based on the knowledge database to provide corresponding knowledge, determine whether the equipment is faulty and the cause of the fault, and indicate the standards involved in the fault based on all answer results and evaluation results obtained by the knowledge reasoning module, fault reasoning module and standard output module, business problems collected by the user interaction module and the vertical question and answer model.

[0125] like Figure 1 As shown in the figure, the search algorithm building module, knowledge reasoning module, fault reasoning module, standard output module and optimal answer generation module together constitute the reasoning enhancement module, which improves the model output accuracy and work efficiency.

[0126] An embodiment of the present application also provides an electronic device, which includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the intelligent operation and maintenance method of substation equipment based on the improved RAG as described above is implemented.

[0127] The embodiment of the present application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned intelligent operation and maintenance method for substation equipment based on improved RAG. Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is executed, the embodiment of the intelligent operation and maintenance method for substation equipment based on improved RAG can be executed.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent operation and maintenance method for substation equipment based on improved RAG, It is characterized in that The steps include: S1, construct vector similarity graphs based on the preset database for storing different substation operation and maintenance knowledge, and obtain a vertical question-answering model based on the technical question-answering database and the large language model; The databases of different substation operation and maintenance knowledge include knowledge databases, fault databases and standard databases, and the technical question and answer database belongs to the knowledge database; S2, improving the breadth-first search algorithm based on the spectral clustering method, and obtaining a spectral clustering BFS search algorithm for performing similar text retrieval on the vector similarity graph in step S1 based on the retrieval vector; S3, collect business questions, and use the vertical question-answering model obtained in step S1 to determine whether the business question is a substation operation and maintenance business. If so, execute step S4, otherwise re-collect business questions; S4, using the business question obtained in step S3 as a retrieval vector and using the algorithm in step S2 to traverse the vector similarity graph corresponding to the knowledge database, generating an answer result for the business question based on the traversal result, the business question and the vertical question-answering model, using the answer result as a retrieval vector to search the vector similarity graph corresponding to the standard database, and obtaining an evaluation result for determining whether the answer result meets the standard in the standard database based on the retrieval result, the answer result and the vertical question-answering model; S5, based on the business question, the answer result obtained in step S4 and the vertical question-answering model, a third evaluation result is generated to determine whether the business question contains a fault and fault information, and whether the device has a fault is determined according to the third evaluation result. If not, return to step S3, otherwise execute step S6; S6, using the third evaluation result as a retrieval vector and using the algorithm of step S2 to search the vector similarity graph corresponding to the fault database, generating a third answer result including the cause of the fault based on the search result, the third evaluation result and the vertical question-answering model, and generating a fourth answer result indicating the standard involved in the fault based on the search result, the business question, the third evaluation result and the vertical question-answering model; S7, based on all the answer results and evaluation results obtained in step S4, step S5 and step S6, the business problem and the vertical question-answering model, generates an optimal solution for the business problem based on the knowledge database to provide relevant knowledge, determine whether the equipment is faulty and the cause of the fault, and indicate the standards involved in the fault.

2. According to claim 1, the intelligent operation and maintenance method of substation equipment based on improved RAG, It is characterized in that In step S1, the method for constructing the vector similarity graph includes the following steps: ①, pre-process the data in each database in step S1 to obtain the corresponding text database; ②, segment and vectorize the text database obtained in step ① to obtain corresponding vector feature data; ③. Use the similarity calculation method to calculate the similarity between any two vector feature data of each text database in step ②. Take the vector feature data as nodes and construct undirected edges between the nodes based on the preset first similarity threshold to obtain the corresponding vector similarity graph. At the same time, construct edge weights based on the similarity between the two nodes corresponding to the undirected edges.

3. According to claim 1, the intelligent operation and maintenance method of substation equipment based on improved RAG, It is characterized in that The step S2 comprises the following steps: i, calculate the normalized Laplacian matrix based on the degree matrix and similarity matrix of the vector similarity graph; ii. Perform eigendecomposition on the normalized Laplace matrix to obtain the first eigenvalue and the first eigenvector, and use the K-means algorithm to cluster the first eigenvector and divide it into several clusters; iii. Calculate the similarity between the search vector and the cluster center of each cluster and each node in the cluster using a similarity calculation method, and use a breadth-first search algorithm based on a second similarity threshold to search the vector similarity graph to generate similar texts.

4. According to claim 1, the intelligent operation and maintenance method of substation equipment based on improved RAG, It is characterized in that The knowledge database includes a technical question and answer database and a theoretical knowledge database.

5. An intelligent operation and maintenance system for substation equipment based on improved RAG, It is characterized in that include: User interaction module: used to collect business issues; Professional knowledge base construction module: used to collect and classify professional knowledge data in the field of substation operation and maintenance, and construct different vector similarity graphs based on the classified professional knowledge data; Vertical large model building module: used to build a vertical question-answering model based on the classified technical question-answering database obtained by the large language model and professional knowledge base building module; Search algorithm building module: used to generate a spectral clustering BFS search algorithm based on the spectral clustering method to improve the breadth-first search algorithm and obtain a spectral clustering BFS search algorithm for similar text retrieval based on the retrieval vector for the vector similarity graph obtained by the professional knowledge base building module; Knowledge reasoning module: used to use the business questions collected by the user interaction module as retrieval vectors to traverse the vector similarity graph corresponding to the knowledge database using the algorithm of the search algorithm building module, generate answer results for the business questions based on the traversal results, business questions and vertical question-answering models, use the answer results as retrieval vectors to search the vector similarity graph corresponding to the standard database, and obtain an evaluation result for determining whether the answer result meets the standards in the standard database based on the retrieval results, the answer results and the vertical question-answering model; Fault reasoning module: used to generate a third evaluation result for determining whether the business question contains a fault and corresponding fault information based on the business question, all answer results obtained by the knowledge reasoning module and the vertical question-answering model, and determine whether the device has a fault according to the third evaluation result. If not, return to step S3, otherwise execute step S6; Standard output module: used to use the third evaluation result output by the fault reasoning module as a retrieval vector to search the vector similarity graph corresponding to the fault database using the algorithm of the search algorithm building module, and generate a third answer result including the cause of the fault based on the search result, the third evaluation result and the vertical question-answering model, and generate a fourth answer result for indicating the standard involved in the fault based on the search result, the business questions collected by the user interaction module, the third evaluation result and the vertical question-answering model; Optimal answer generation module: used to generate the optimal answer solution for the business problem based on the knowledge database to provide corresponding knowledge, determine whether the equipment is faulty and the cause of the fault, and indicate the standards involved in the fault based on all answer results and evaluation results obtained by the knowledge reasoning module, fault reasoning module and standard output module, business problems collected by the user interaction module and the vertical question and answer model.

6. An electronic device comprising a processor and a memory, It is characterized in that The memory stores a computer program, and when the computer program is executed by the processor, the intelligent operation and maintenance method of substation equipment based on improved RAG as described in any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium storing a computer program, It is characterized in that When the computer program is executed by a processor, the intelligent operation and maintenance method for substation equipment based on improved RAG as described in any one of claims 1 to 4 is implemented.

Citation Information

Patent Citations

  • Information search method and system based on interactive document clustering

    CN103514183A

  • Question answering system and method based on vector and large language model

    CN117493513A

  • Substation operation and maintenance intelligent question answering method based on large model, related method and device

    CN117891831A

  • Question and answer method and device based on RAG and GNN and electronic equipment

    CN118210906A

  • Construction industry safety question and answer method and equipment based on large language model technology

    CN118332092A

Cited By

  • Equipment abnormal information collection method and system based on large model agent

    CN119624392A

  • User question and answer method and device, electronic equipment, storage medium and program product

    CN119938951A