Transformer fault root cause analysis method based on large model iterative reasoning

Abstract

The invention relates to a transformer fault root cause analysis method based on large model iterative reasoning. The method comprises the steps of performing retrieval based on a fault phenomenon in combination with a fault instance knowledge base to obtain related fault instances; inputting the fault phenomenon and the related fault instance into a large model to obtain a fault root cause reasoning result; performing hybrid retrieval based on semantic vectors and keywords on the basis of the fault phenomenon and the fault root cause reasoning result in combination with the fault instance knowledge base to obtain a candidate retrieval result set; and reordering the candidate retrieval result set by adopting a large model, and determining a target fault reason based on a fault phenomenon in combination with a thinking chain technology. By means of retrieval enhancement, the defect of a large model in the aspect of reasoning fact accuracy can be overcome, the efficiency and accuracy of power grid transformer fault root cause analysis are improved, meanwhile, analysis of the large model on fault causes can be continuously refined and improved on the basis of the iterative reasoning process, and the accuracy and reliability of a reasoning result are remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a transformer fault root cause analysis method based on large model iterative reasoning. Background Art

[0002] In power systems, root cause analysis of transformer failures is a key technology. This process mainly involves determining the root cause behind the failure based on the abnormal phenomena manifested by the failure. As research progresses, this field has gradually evolved from initially relying on machine learning algorithms to draw causal relationship maps to implementing fault root cause analysis based on deep learning neural networks, which inputs fault phenomena and domain rules, and the model outputs the cause of the failure. However, such methods usually require a large amount of labeled data for supervised training, which consumes huge time and labor costs.

[0003] Recently, large language models have provided a more efficient solution for transformer fault root cause analysis with their powerful logical reasoning ability and domain adaptability. By inputting fault instances into the large language model, the model is assisted by few-sample context learning to infer possible fault causes based on the fault phenomenon. Although the large model can directly infer the root cause of the transformer fault based on the abnormal phenomenon, it still faces many challenges in practical applications. First, the large model has the problem of "hallucination", and the generated fault cause may not match the actual situation or be logically inconsistent. In addition, the knowledge related to transformers in the power field contains a large number of professional terms and complex cross-domain features, which poses a challenge to the task understanding and reasoning ability of the large language model, resulting in its poor performance in the complex task of transformer fault root cause analysis.

[0004] Therefore, in the relevant technology, there is an urgent need for a method that can solve the hallucination problem in the application of large language models in the vertical field of electricity and improve the reasoning ability of large language models in the complex task of transformer fault root cause analysis. Summary of the invention

[0005] Based on this, it is necessary to provide a transformer fault root cause analysis method based on large model iterative reasoning, which can solve the hallucination problem in the application of large language models in the vertical field of power and improve the reasoning ability of large language models in complex tasks of transformer fault root cause analysis.

[0006] In a first aspect, the present application provides a transformer fault root cause analysis method based on large model iterative reasoning. The method comprises: Based on the fault phenomenon and the fault instance knowledge base, search to obtain relevant fault instances; Input the fault phenomenon and related fault instances into the large model to obtain the fault root cause reasoning result; Perform a hybrid retrieval based on semantic vectors and keywords by combining the fault phenomenon and the inference result of the fault root cause with the fault instance knowledge base to obtain a set of candidate retrieval results; Use a large model to re-rank the set of candidate retrieval results, and determine the target fault cause based on the fault phenomenon combined with the chain of thought technique.

[0007] Optionally, in an embodiment of the present application, the hybrid retrieval based on semantic vectors and keywords includes: Use a pre-trained embedding model to perform vector encoding on the fault phenomenon and the inference result of the fault root cause, and calculate the cosine similarity with the instance vectors in the fault instance knowledge base to obtain matching instances.

[0008] Optionally, before using the pre-trained embedding model to perform vector encoding on the fault phenomenon and the inference result of the fault root cause in an embodiment of the present application, it includes: Use the power fault instance library to construct a fine-tuning data set to fine-tune the pre-trained embedding model.

[0009] Optionally, in an embodiment of the present application, the hybrid retrieval based on semantic vectors and keywords further includes: Based on the BM25 algorithm, calculate the correlation score of the frequency of occurrence of keywords in the fault instances of the fault instance knowledge base relative to the frequency of occurrence of keywords in the fault instance knowledge base to obtain matching instances.

[0010] Optionally, in an embodiment of the present application, the use of a large model to re-rank the set of candidate retrieval results includes: Set a sliding window to re-rank the relevance of the candidate retrieval results within the window in turn.

[0011] In a second aspect, the present application also provides a transformer fault root cause analysis device based on iterative inference of a large model. The device includes: A primary retrieval module for retrieving based on the fault phenomenon combined with the fault instance knowledge base to obtain relevant fault instances; A primary inference module that inputs the fault phenomenon and the relevant fault instances into a large model to obtain an inference result of the fault root cause; A hybrid retrieval module for performing a hybrid retrieval based on semantic vectors and keywords by combining the fault phenomenon and the inference result of the fault root cause with the fault instance knowledge base to obtain a set of candidate retrieval results; A fault cause determination module for using a large model to re-rank the set of candidate retrieval results and determining the target fault cause based on the fault phenomenon combined with the chain of thought technique.

[0012] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the steps of the methods described in the above embodiments.

[0013] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in each of the above embodiments are implemented.

[0014] The above-mentioned transformer fault root cause analysis method based on large model iterative reasoning, first, based on the fault phenomenon combined with the fault instance knowledge base, a search is performed to obtain relevant fault instances; then, the fault phenomenon and the relevant fault instance are input into the large model to obtain the fault root cause reasoning result; then, based on the fault phenomenon and the fault root cause reasoning result combined with the fault instance knowledge base, a hybrid search based on semantic vectors and keywords is performed to obtain a candidate search result set; finally, the large model is used to reorder the candidate search result set, and the target fault cause is determined based on the fault phenomenon combined with the thinking chain technology. In other words, the use of retrieval enhancement can solve the shortcomings of the large model in the accuracy of reasoning facts, improve the efficiency and accuracy of the root cause analysis of power grid transformer faults, and at the same time, based on the iterative reasoning process, the large model can continuously refine and improve the analysis of the fault cause, which significantly improves the accuracy and reliability of the reasoning results. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 An application environment diagram of a transformer fault root cause analysis method based on large model iterative reasoning in one embodiment; Figure 2 A schematic flow chart of a transformer fault root cause analysis method based on large model iterative reasoning in one embodiment; Figure 3 is a schematic diagram of a retrieval method based on a semantic vector in one embodiment; Figure 4 is a schematic diagram of a keyword-based search method in an embodiment; Figure 5 A flowchart of specific steps of a transformer fault root cause analysis method based on large model iterative reasoning in one embodiment; Figure 6 It is a structural block diagram of a transformer fault root cause analysis device based on large model iterative reasoning in one embodiment; Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0017] The transformer fault root cause analysis method based on large model iterative reasoning provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be integrated on the server, or it can be placed on the cloud or other network servers. Among them, the terminal can be but not limited to various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented as an independent server or a server cluster consisting of multiple servers.

[0018] In one embodiment, Figure 2 As shown in the figure, a transformer fault root cause analysis method based on large model iterative reasoning is provided. Figure 1 The server in the example is used to illustrate the following steps: S201: Searching a fault instance knowledge base based on the fault phenomenon to obtain relevant fault instances.

[0019] In the embodiment of the present application, first, a fault phenomenon is constructed. and cause of failure Fault Instance Knowledge Base This method will iterate the retrieval-reasoning process T times. In the first round of iteration, based on the fault phenomenon Search from the fault instance knowledge base to obtain relevant fault instances ,Related fault instances include fault phenomena and fault causes.

[0020] S203: Input the fault phenomenon and related fault instances into the large model to obtain the fault root cause reasoning result.

[0021] In the embodiment of the present application, after that, the fault phenomenon and the related fault instances retrieved Input into the big model, and the big model generates new fault root cause reasoning results In specific applications, Qwen-7B can be selected as the base large model.

[0022] S205: Based on the fault phenomenon and the fault root cause reasoning result combined with the fault instance knowledge base, a hybrid search based on semantic vectors and keywords is performed to obtain a candidate search result set.

[0023] In the embodiment of the present application, after that, the reasoning enhanced retrieval is performed, taking the tth iteration as an example, based on the fault phenomenon And the inference results of the large model in the previous iteration , a hybrid search based on semantic vectors and keywords is performed from the fault instance knowledge base to obtain a set of candidate search results. Among them, the search based on semantic vectors adopts a common intensive search method. At the same time, when facing short text or semantically ambiguous fault phenomenon queries, there will be a problem of missing important keywords, resulting in the information retrieved using semantic vectors being irrelevant to the fault query information. Therefore, a keyword-based search method is used to solve this problem.

[0024] Specifically, in one embodiment of the present application, the hybrid search based on semantic vectors and keywords includes: The fault phenomenon and the fault root cause reasoning result are vector-encoded by using a pre-trained embedding model, and the cosine similarity with the instance vector in the fault instance knowledge base is calculated to obtain a matching instance.

[0025] In one embodiment of the present application, Figure 3 As shown in the figure, the pre-trained embedding model bge-base-zh is used to retrieve the fault phenomenon of the input And the inference results of the large model in the previous iteration Vector encoding is performed, and then the cosine similarity between the vector and the instance vector in the fault instance knowledge base is calculated. The k fault instances with the highest similarity are determined as the most matching ones. In specific applications, considering factors such as inference cost, k=5 is usually taken as the optimal value. The specific calculation formula is as follows:

[0026] in, represents the pre-trained embedding model encoder, represents the parameters of the encoder, represents the cosine similarity calculation of the vector, and Representative fault instance knowledge base The mth instance in Indicates the instance and fault symptom The matching score.

[0027] In one embodiment of the present application, before the pre-trained embedding model is used to vector encode the fault phenomenon and the fault root cause reasoning result, the method includes: A fine-tuning dataset is constructed using a power fault instance library to fine-tune the pre-trained embedding model.

[0028] In one embodiment of the present application, in order to enable the pre-trained embedding model to accurately reflect domain-specific corpus information and enhance its ability to understand the relevance of fault queries and retrieval content, the power fault instance library is sampled and the embedding model is fine-tuned for the task-specific domain. Specifically, JSON data in the format of {"query":str,"pos":List[str],"neg":List[str]} is constructed as a fine-tuning dataset, where query is the fault phenomenon, pos is the corresponding fault cause, and neg is the mismatched fault cause. The constructed fine-tuning dataset is trained and fine-tuned on the pre-trained embedding model.

[0029] In one embodiment of the present application, the hybrid search based on semantic vectors and keywords further includes: Based on the BM25 algorithm, the correlation score of the frequency of occurrence of a keyword in a single fault instance in the fault instance knowledge base relative to the frequency of occurrence of the keyword in all fault instances in the fault instance knowledge base is calculated to obtain a matching instance.

[0030] In one embodiment of the present application, Figure 4 As shown in the figure, the fault phenomena are And the inference results of the large model in the previous iteration And failure examples Perform Chinese word segmentation, and then calculate and troubleshoot based on the results after word segmentation And the inference results of the large model in the previous iteration Related fault examples The scores of the most relevant k matching fault instances are obtained. In specific applications, considering factors such as inference cost, k=5 is usually taken as the optimal value. The specific calculation formula is as follows:

[0031] in, The current fault phenomenon And the inference results of the large model in the previous iteration The result after word segmentation is score ( ) is the current query and failure instances The relevance score of for The words, for exist The frequency of occurrence in The corresponding length is is the average length of the fault instance data set, and are adjustable parameters, usually set to 1.2 and 0.75 respectively. for The corresponding inverse document frequency is calculated as:

[0032] in, is the number of fault instance samples, To include The number of fault instance samples.

[0033] S207: Reorder the candidate search result set using a large model, and determine the target fault cause based on the fault phenomenon combined with the thought chain technology.

[0034] In the embodiment of the present application, the large language model LLMs can understand the complex relationship between the query and the document, including semantic relevance, user intent and contextual information of the document, and use LLMS to directly sort a set of candidate retrieval results and select the top k instances to constitute the retrieval results of the current iteration round. ,In specific applications, considering factors such as inference cost, k = 5 is usually taken as the ,optimal value setting.,As shown in Table 1, by providing LLMs with clear instructions and ,background information of the fault phenomena, we guide LLMs to directly output the ,ranking results of the paragraphs instead of generating intermediate relevance scores.

[0035]

[0036] in, is the current fault phenomenon, passage_1 and passage_2 are the retrieved related results.

[0037] In addition, considering that there is a long text that exceeds the LLMs output context length, in one embodiment of the present application, the re-ranking of the candidate search result set using a large model includes: A sliding window is set to reorder the candidate search results in the window in order of relevance.

[0038] In one embodiment of the present application, a sliding window is used to sequentially reorder the relevance of the search results in the window by setting the window size and the moving step.

[0039] Afterwards, as shown in Table 2, the root cause reasoning of retrieval enhancement is based on the fault phenomenon and the current round of search results , through the prompt template (as shown in Table 1) to guide the large model to generate the corresponding fault root cause reasoning results Specifically, the algorithm retrieves the k matched instances as sample data for the large model's few-sample learning, and gradually guides the large model to analyze the fault phenomenon through the thinking chain technology. .

[0040]

[0041] in, , All of them are in the fault instance knowledge base. Through the above methods, the current fault analysis results can be obtained.

[0042] The above process will be repeated continuously. During the iteration process, if contains the field "The cause of the failure is:", which means that the large model has successfully inferred the failure phenomenon The corresponding fault cause, at this time As a cause of failure , the iteration ends; in addition, when the preset number of iterations T is reached, the model will also stop iterating.

[0043] In the above-mentioned transformer fault root cause analysis method based on large model iterative reasoning, first, based on the fault phenomenon combined with the fault instance knowledge base, a search is performed to obtain relevant fault instances; then, the fault phenomenon and the relevant fault instance are input into the large model to obtain the fault root cause reasoning result; then, based on the fault phenomenon and the fault root cause reasoning result combined with the fault instance knowledge base, a hybrid search based on semantic vectors and keywords is performed to obtain a candidate search result set; finally, the large model is used to reorder the candidate search result set, and the target fault cause is determined based on the fault phenomenon combined with the thinking chain technology. In other words, the use of retrieval enhancement can solve the shortcomings of the large model in the accuracy of reasoning facts, improve the efficiency and accuracy of the root cause analysis of power grid transformer faults, and at the same time, based on the iterative reasoning process, the large model can continuously refine and improve the analysis of the fault cause, which significantly improves the accuracy and reliability of the reasoning results.

[0044] The following is a specific example to illustrate the specific implementation steps of a transformer fault root cause analysis method based on large model iterative reasoning in this application. Figure 5 As shown, first, enter the current fault phenomenon , perform the first search in combination with the fault instance knowledge base, input the search results and the current fault phenomenon into the large model reasoning, and obtain the reasoning result , the first round of iteration ends. Then, the second round of iteration begins, based on the fault phenomenon and the root cause reasoning results Combine the fault instance knowledge base to perform a hybrid search based on semantic vectors and keywords to obtain a candidate search result set. Use the big model to reorder the candidate search result set to obtain the search result. Perform a second big model reasoning based on the search result and the current fault phenomenon to obtain the reasoning result. , the second round of iteration ends, and so on, until during the iteration process, if contains the field "The cause of the failure is:", which means that the large model has successfully inferred the failure phenomenon The corresponding fault cause, at this time As a cause of failure , the iteration ends; in addition, when the preset number of iterations T is reached, the model will also stop iterating.

[0045] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0046] Based on the same inventive concept, the embodiment of the present application also provides a transformer fault root cause analysis device based on large model iterative reasoning for implementing the above-mentioned transformer fault root cause analysis method based on large model iterative reasoning. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the transformer fault root cause analysis device based on large model iterative reasoning provided below can refer to the above limitations on the transformer fault root cause analysis method based on large model iterative reasoning, and will not be repeated here.

[0047] In one embodiment, Figure 6 As shown, a transformer fault root cause analysis device 600 based on large model iterative reasoning is provided, comprising: an initial retrieval module 601, an initial reasoning module 603, a hybrid retrieval module 605 and a fault cause determination module 607, wherein: The initial search module 601 is used to search based on the fault phenomenon in combination with the fault instance knowledge base to obtain relevant fault instances.

[0048] The initial reasoning module 603 inputs the fault phenomenon and related fault instances into the large model to obtain the fault root cause reasoning result.

[0049] The hybrid retrieval module 605 is used to perform a hybrid retrieval based on semantic vectors and keywords based on the fault phenomenon and fault root cause reasoning results combined with the fault instance knowledge base to obtain a candidate retrieval result set.

[0050] The fault cause determination module 607 is used to reorder the candidate search result set using a large model, and determine the target fault cause based on the fault phenomenon combined with the thought chain technology.

[0051] In one embodiment of the present application, the hybrid retrieval module is further used to: The fault phenomenon and the fault root cause reasoning result are vector-encoded by using a pre-trained embedding model, and the cosine similarity with the instance vector in the fault instance knowledge base is calculated to obtain a matching instance.

[0052] In one embodiment of the present application, the hybrid retrieval module is further used to: A fine-tuning dataset is constructed using a power fault instance library to fine-tune the pre-trained embedding model.

[0053] In one embodiment of the present application, the hybrid retrieval module is further used to: Based on the BM25 algorithm, the correlation score of the frequency of occurrence of a keyword in a single fault instance in the fault instance knowledge base relative to the frequency of occurrence of the keyword in all fault instances in the fault instance knowledge base is calculated to obtain a matching instance.

[0054] In one embodiment of the present application, the fault cause determination module is further used to: A sliding window is set to reorder the candidate search results in the window in order of relevance.

[0055] Each module in the above-mentioned transformer fault root cause analysis device based on large model iterative reasoning can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0056] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a transformer fault root cause analysis method based on large model iterative reasoning is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0057] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0058] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0059] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0060] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0061] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0062] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0063] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0064] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A transformer fault root cause analysis method based on large model iterative reasoning, characterized in that: The method comprises: Based on the fault phenomenon and the fault instance knowledge base, search to obtain relevant fault instances; Input the fault phenomenon and related fault instances into the large model to obtain the fault root cause reasoning result; Based on the fault phenomenon and the fault root cause reasoning result combined with the fault instance knowledge base, a hybrid search based on semantic vectors and keywords is performed to obtain a candidate search result set; A large model is used to reorder the candidate search result set, and the target fault cause is determined based on the fault phenomenon combined with the thinking chain technology.

2. A transformer fault root cause analysis method based on large model iterative reasoning according to claim 1, characterized in that: The hybrid search based on semantic vectors and keywords includes: The fault phenomenon and the fault root cause reasoning result are vector-encoded by using a pre-trained embedding model, and the cosine similarity with the instance vector in the fault instance knowledge base is calculated to obtain a matching instance.

3. A transformer fault root cause analysis method based on large model iterative reasoning according to claim 2, characterized in that: Before the pre-trained embedding model is used to vector encode the fault phenomenon and the fault root cause reasoning result, the following steps are included: A fine-tuning dataset is constructed using a power fault instance library to fine-tune the pre-trained embedding model.

4. A transformer fault root cause analysis method based on large model iterative reasoning according to claim 1, characterized in that: The hybrid search based on semantic vectors and keywords also includes: Based on the BM25 algorithm, the correlation score of the frequency of occurrence of a keyword in a single fault instance in the fault instance knowledge base relative to the frequency of occurrence of the keyword in all fault instances in the fault instance knowledge base is calculated to obtain a matching instance.

5. The transformer fault root cause analysis method based on large model iterative reasoning according to claim 1 is characterized in that: The adopting the large model to reorder the candidate search result set comprises: A sliding window is set to reorder the candidate search results in the window in order of relevance.

6. A transformer fault root cause analysis device based on large model iterative reasoning, characterized in that: The device comprises: The initial search module is used to search based on the fault phenomenon and the fault instance knowledge base to obtain relevant fault instances; The initial reasoning module inputs the fault phenomenon and related fault instances into the large model to obtain the fault root cause reasoning result; A hybrid retrieval module, used to perform a hybrid retrieval based on semantic vectors and keywords based on the fault phenomenon and the fault root cause reasoning result combined with the fault instance knowledge base to obtain a candidate retrieval result set; The fault cause determination module is used to reorder the candidate search result set using a large model, and determine the target fault cause based on the fault phenomenon combined with the thinking chain technology.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images