Electrical equipment fault diagnosis method based on question and answer mode and computer system

Abstract

The invention belongs to the technical field of electrical equipment fault diagnosis, and particularly relates to an electrical equipment fault diagnosis method based on a question and answer mode and a computer system. The method comprises the following steps: S1, acquiring electrical equipment data; s2, inputting the electrical equipment data into the trained fault identification and diagnosis small model to obtain fault description information of the electrical equipment; and S3, inputting the fault description information into the trained electrical equipment large model to obtain a similar fault record similar to the fault description information and / or a fault solution of the fault description information. In the whole fault diagnosis process, only the electrical equipment large model is a large model, the data volume required for training the fault diagnosis identification small model is far smaller than the data volume required for training the large model, the hardware requirement required for training is lower, and the training cost is greatly reduced. The technical problem that in the prior art, the training cost is high due to the fact that a large data volume is needed in the model training process is solved.

Description

Technical Field

[0001] This invention belongs to the field of electrical equipment fault diagnosis technology, specifically relating to a fault diagnosis method and computer system for electrical equipment based on a question-and-answer approach. Background Technology

[0002] Electrical equipment is the foundation of new power systems, and its operational reliability directly affects the safe and stable operation of the power grid. Electrical equipment operation and maintenance (O&M) encompasses equipment inspection, repair, troubleshooting, and performance testing. Condition monitoring and fault diagnosis ensure the normal operation of critical equipment and the safe and stable operation of the system. Traditional O&M primarily relies on manual inspections, which struggle to detect hidden defects, leading to missed opportunities for critical intervention. Furthermore, O&M personnel face challenges in learning and have limited experience, hindering their ability to develop effective troubleshooting solutions and significantly reducing O&M efficiency. Therefore, it is necessary to employ intelligent methods such as large-scale modeling technology to achieve rapid fault diagnosis and solution output for electrical equipment, thereby improving O&M efficiency.

[0003] Chinese invention patent application CN117763107A, published on March 26, 2024, discloses a method for power defect image detection based on a text-image question-answering multimodal model. This method constructs a text-image question-answering multimodal model based on a visual model and a language dialogue model, and trains the model using the Low-Rank Adaptive Relationship (LoRA) method and Q-Former. Then, by performing text-image question-answering on the trained model, information such as the power scene and its defects can be obtained. This technical solution uses a fusion of large visual and semantic models to construct the text-image question-answering multimodal model, requiring a large amount of data during training and placing high demands on the hardware, thus increasing hardware costs. Furthermore, this technical solution can only achieve defect identification and cannot provide maintenance personnel with solutions or similar historical cases.

[0004] Chinese invention patent application CN117612189A, published on February 7, 2024, discloses a training method and system for a large-scale cognitive model of power equipment operation and maintenance. The method includes: acquiring relevant text and image data related to power equipment operation and maintenance; inputting the relevant text and image data into a Transformer architecture for pre-training to obtain a large-scale cognitive model of power equipment operation and maintenance. This technical solution also employs a fusion of visual and semantic large-scale models to construct the large-scale cognitive model of power equipment operation and maintenance; therefore, it also faces the technical challenge of requiring a large amount of data during training.

[0005] In summary, existing technologies suffer from the technical problems of requiring a large amount of data and demanding hardware for model training, resulting in high training costs. Summary of the Invention

[0006] The purpose of this invention is to provide a fault diagnosis method and computer system for electrical equipment based on a question-and-answer approach, in order to solve the technical problem that the model training process in the prior art requires a large amount of data, resulting in high training costs.

[0007] To address the aforementioned technical problems, the present invention provides a technical solution for a question-and-answer-based method for diagnosing electrical equipment faults, comprising the following steps:

[0008] S1. Obtain electrical equipment data;

[0009] S2. Input the electrical equipment data into the trained fault identification and diagnosis mini-model to obtain fault description information of the electrical equipment.

[0010] The fault identification and diagnosis small model is a pre-established small model that takes electrical equipment data as input and fault description information that represents electrical equipment fault information as output.

[0011] S3. Input the fault description information into the trained electrical equipment large model to obtain similar fault records and / or fault description information with similar fault description information;

[0012] The large-scale electrical equipment model is a pre-established model that takes fault description information representing electrical equipment fault information as input and similar fault records and / or fault solutions as output.

[0013] The beneficial effects of the above technical solution are as follows: The technical solution of the electrical equipment fault diagnosis method based on question-and-answer method of the present invention belongs to an improved invention. When performing fault diagnosis, the present invention first uses a small fault diagnosis identification model to obtain fault description information. Then, the large electrical equipment model outputs similar fault records and / or fault solutions based on the fault description information output by the small fault diagnosis identification model. Throughout the process, only the large electrical equipment model needs to be used, while the amount of data required to train the small fault diagnosis identification model is much smaller than the amount of data required to train the large model, and the hardware requirements for training are also lower, greatly reducing training costs. The present invention solves the technical problem of high training costs caused by the large amount of data required for model training in the prior art.

[0014] Furthermore, the fault identification and diagnosis small model includes at least one of the following: an image-based fault identification and diagnosis model with image input, a speech-based fault identification and diagnosis model with speech input, and a natural language-based fault identification and diagnosis model with text input.

[0015] When the fault identification and diagnosis model includes two or more of the following: image-based fault identification and diagnosis model, voice-based fault identification and diagnosis model, and natural language-based fault identification and diagnosis model, the outputs of each type of fault identification and diagnosis model are concatenated to obtain the fault description information.

[0016] Furthermore, the fault description information input into the large-scale electrical equipment model is vectorized fault description information obtained after vectorization processing; the large-scale electrical equipment model obtains the similar fault records and / or the fault solutions according to the following methods:

[0017] The electrical equipment large model screens the vectorized fault description information in the vector library of the electrical equipment large model, and combines the prompt words to output similar fault records and / or fault solutions.

[0018] Furthermore, the large model of the electrical equipment is obtained in the following way:

[0019] After incremental training and / or instruction fine-tuning of the general large model base based on power sector data, a vertical large model base for electrical equipment is obtained; then, a vector library of power sector data is constructed, and the electrical equipment vertical large model base is trained based on the vector library to obtain the electrical equipment large model.

[0020] Furthermore, the vector library is obtained by sequentially parsing, segmenting, and normalizing the historical fault report documents.

[0021] Furthermore, the electrical equipment data acquired by S1 is the preprocessed data of the collected raw data. The preprocessing includes at least one of format conversion, data processing to supplement missing data, and data cleaning to remove duplicate and useless data.

[0022] Furthermore, the fault description information can be text, images, or voice.

[0023] The present invention also provides a technical solution for a computer system: a computer system including a processor, the processor being used to execute a computer program to implement the steps of the question-and-answer based electrical equipment fault diagnosis method as described above. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the software architecture of an embodiment of the electrical equipment fault diagnosis method based on the question-and-answer method of the present invention.

[0025] Figure 2 This is a flowchart illustrating the question-and-answer method for diagnosing electrical equipment faults, as described in an embodiment of the present invention.

[0026] Figure 3This is a flowchart illustrating the output of similar fault records and fault solutions in an embodiment of the electrical equipment fault diagnosis method based on a question-and-answer approach of the present invention.

[0027] Figure 4 This is a flowchart illustrating the training process of a large-scale electrical equipment model, representing an embodiment of the question-and-answer-based electrical equipment fault diagnosis method of the present invention. Detailed Implementation

[0028] In existing technologies, to implement a text-based question-and-answer fault diagnosis model, the model often requires two large models: a large visual model and a large semantic model. Training these large models requires a large amount of data and significant computing power, resulting in high training costs. This invention, in its fault diagnosis process, first utilizes a small fault diagnosis and identification model to obtain fault description information. Then, the large electrical equipment model outputs similar fault records and / or fault solutions based on the fault description information from the small model. Throughout this process, only the large electrical equipment model is needed, and the amount of data required to train the small model is far less than that required to train the large model. The hardware requirements for training are also lower, significantly reducing training costs. This invention solves the technical problem of high training costs caused by the large amount of data required for model training in existing technologies.

[0029] Example of a question-and-answer based electrical equipment fault diagnosis method:

[0030] like Figure 2 As shown, the question-and-answer based electrical equipment fault diagnosis method of this embodiment includes the following steps:

[0031] S1. Obtain electrical equipment data.

[0032] In this embodiment, the collected raw electrical equipment data is also preprocessed. The preprocessing operations include format conversion, data processing to supplement missing data, and data cleaning to remove duplicate and useless data.

[0033] In other embodiments, the preprocessing operation may include at least one of format conversion, data processing to supplement missing data, and data cleaning to remove duplicate and useless data, with the appropriate preprocessing operation selected based on the nature of the acquired electrical equipment data.

[0034] S2. Input the electrical equipment data into the trained fault identification and diagnosis mini-model to obtain fault description information of the electrical equipment.

[0035] The fault identification and diagnosis small model is a pre-selected small model that takes electrical equipment data as input and fault description information that represents electrical equipment fault information as output.

[0036] Specifically, in this embodiment, the fault identification and diagnosis mini-model is a deep learning network model that includes an image-based fault identification and diagnosis model (object detection, image segmentation, image classification) with image input, a speech-based fault identification and diagnosis model with speech input, a natural language-based fault identification and diagnosis model with text input, and time series prediction.

[0037] Correspondingly, the electrical equipment data in S1 corresponding to the fault identification and diagnosis mini-model can be multimodal data including video, images, sound, text, etc.

[0038] The establishment and training of various fault diagnosis and identification models can refer to existing technologies, and will not be described in detail here.

[0039] S3. Input the fault description information output by the fault diagnosis and identification small model into the trained electrical equipment large model to obtain similar fault records and fault solutions with similar fault description information.

[0040] The large-scale electrical equipment model is a pre-established model that takes fault description information representing electrical equipment fault information as input and outputs similar fault records and fault solutions. Specifically, such as... Figure 3 As shown, the fault description information is vectorized to form a standardized document, and the standardized document is input into the electrical equipment large model. The electrical equipment large model filters the standardized document formed by the vectorized fault description information in the vector library, and outputs historical fault records and fault solutions similar to the fault description information in combination with prompt words.

[0041] It should be noted that when the input data is multimodal data, various fault identification and diagnosis models will distinguish the input image, voice or text information respectively. The system will then connect and integrate the results according to the logical relationship between the results and output them to the large model to obtain similar fault records and fault solutions that are similar to the fault description information.

[0042] When the fault identification and diagnosis model includes two or more of the following: image-based fault identification and diagnosis model, voice-based fault identification and diagnosis model, and natural language-based fault identification and diagnosis model, the outputs of each type of fault identification and diagnosis model are concatenated to obtain the fault description information.

[0043] Specifically, in this embodiment, the fault description is in text form. Of course, in other embodiments, the output fault description can be in other forms such as images or voice.

[0044] In other implementations, the large model of electrical equipment can also be configured to output information that is only similar fault records or only fault solutions.

[0045] Specifically, the vector library in the large model of electrical equipment is constructed in the following way: the vector library is obtained by sequentially parsing, segmenting and normalizing historical fault report documents.

[0046] like Figure 4 As shown, the large-scale electrical equipment model in this embodiment is obtained in the following way:

[0047] After incremental training and instruction fine-tuning of a general large-scale model based on power sector data, a vertical large-scale model for electrical equipment was obtained. Then, a vector library of power sector data was constructed, and the electrical equipment vertical large-scale model was trained based on this vector library to obtain the electrical equipment large-scale model (i.e.,...). Figure 4 (A large-scale dialogue model for the electrical equipment industry).

[0048] Specifically, the instruction fine-tuning technique used in this embodiment is QLoRA (Query-specific Low-Rank Adaptation) technology.

[0049] In other implementations, a large model base for electrical equipment can be obtained by incrementally training only on a general large model base.

[0050] In other implementations, if there is limited data available in the power sector, incremental training can be skipped, and only fine-tuning of instructions can be performed to obtain the large model base for the electrical equipment category.

[0051] The question-and-answer based electrical equipment fault diagnosis method in this embodiment can be as follows: Figure 1 The software architecture is shown below. Specifically, it can include a data layer, a platform layer, and an application layer. The data layer is the corpus for the electrical equipment fault diagnosis question-and-answer system, including power expert knowledge datasets, equipment operation datasets, and equipment maintenance datasets. This corpus is used to incrementally train and fine-tune the general large model base to obtain the electrical equipment vertical large model base. Simultaneously, a power domain data vector library is constructed, and the electrical equipment vertical large model base is trained based on this vector library to obtain the electrical equipment large model. The platform layer includes a basic model and a dedicated electrical equipment operation and maintenance model: the basic model includes a language large model (i.e., the electrical equipment large model) and a fault identification and diagnosis small model, used to construct the dedicated electrical equipment operation and maintenance model; the dedicated electrical equipment operation and maintenance model integrates the language large model and the fault identification and diagnosis small model, employing electrical equipment domain knowledge internalization enhancement and large-small model collaborative decision-making technology to achieve electrical equipment fault identification and diagnosis, proactive push of similar fault cases, and intelligent question-and-answer for electrical equipment fault solutions. The application layer is the electrical large model operation and maintenance assistant, serving as the carrier of the dedicated electrical equipment operation and maintenance model, enabling proactive push of similar fault cases and intelligent question-and-answer for electrical equipment fault solutions.

[0052] Computer system example:

[0053] A computer system includes a processor for executing a computer program to implement the steps of the question-and-answer based electrical equipment fault diagnosis method as described above. The specific question-and-answer based electrical equipment fault diagnosis method has been described in sufficient detail in the above-described embodiments and will not be repeated here.

[0054] This invention has the following characteristics:

[0055] This invention requires less data for model training, effectively reducing the training and inference costs of large models. Simultaneously, this patent enables the analysis and diagnosis of multimodal data from electrical equipment, including video, images, sound, and text. Based on fault descriptions of electrical equipment, it provides maintenance personnel with similar historical cases and fault solutions, achieving a closed-loop process for electrical equipment defect identification, diagnosis, and maintenance. This effectively improves the efficiency of electrical equipment maintenance, contributes to the construction of new power systems, and has broad application prospects in the power sector.

[0056] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still make modifications to the technical solutions described in the foregoing embodiments without creative effort, or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing a fault of an electrical equipment based on a question-and-answer method, characterized by, The method comprises the following steps: S1, acquiring electrical equipment data; S2, inputting the electrical equipment data into a trained fault identification and diagnosis small model to obtain fault description information of the electrical equipment; The fault identification and diagnosis small model is a pre-established small model taking electrical equipment data as input and taking fault description information representing electrical equipment fault information as output; S3, inputting the fault description information into a trained electrical equipment large model to obtain similar fault records similar to the fault description information and / or fault solutions of the fault description information; The electrical equipment large model is a pre-established large model taking fault description information representing electrical equipment fault information as input and taking the similar fault records and / or the fault solutions as output.

2. The question-and-answer-based electrical equipment failure diagnosis method according to claim 1, characterized by, The fault identification and diagnosis small model comprises at least one of an image-based fault identification and diagnosis model taking images as input, a voice-based fault identification and diagnosis model taking voice as input, and a natural language-based fault identification and diagnosis model taking text as input; When the fault identification and diagnosis small model comprises two or more of the image-based fault identification and diagnosis model, the voice-based fault identification and diagnosis model, and the natural language-based fault identification and diagnosis model, the outputs of the various fault identification and diagnosis models are concatenated to obtain the fault description information.

3. The question-and-answer-based electrical equipment fault diagnosis method according to claim 1, characterized by, The fault description information input into the electrical equipment large model is vectorized fault description information obtained after vectorization processing; the electrical equipment large model obtains the similar fault records and / or the fault solutions in the following manner: The electrical equipment large model screens the vectorized fault description information in the vector library of the electrical equipment large model, and outputs the similar fault records and / or the fault solutions in combination with prompt words engineering.

4. The question-and-answer-based electrical equipment failure diagnosis method according to claim 1 or 3, characterized by, The electrical equipment large model is obtained in the following manner: An electrical equipment vertical large model base is obtained by incrementally training and / or instruction fine-tuning a general large model base based on power field data; then a vector library of power field data is constructed, and the electrical equipment vertical large model base is trained based on the vector library to obtain the electrical equipment large model.

5. The question-and-answer-based electrical equipment failure diagnosis method according to claim 3, characterized by, The vector library is obtained by sequentially performing document parsing, document segmentation, and normalization processing on historical fault report documents.

6. The question-and-answer-based electrical equipment failure diagnosis method according to claim 1, characterized by, The electrical equipment data acquired in S1 is preprocessed data obtained by preprocessing collected raw data, and the preprocessing comprises at least one of format conversion, data processing for supplementing missing data, and data cleaning for removing duplicate and useless data.

7. The question-and-answer-based electrical equipment failure diagnosis method according to claim 1 or 2, characterized by, The modality of the fault description information is text, picture, or voice.

8. A computer system comprising a processor, characterized in that The processor is configured to execute a computer program to implement the steps of the electrical equipment fault diagnosis method based on the question and answer mode according to any one of claims 1-7.

Images