Large model-based main transformer defect risk identification method

Through the combination of multimodal data processing and knowledge graph of the large model, the problem of medium and high training costs and insufficient feature coverage of main transformer defect risk identification is solved, and more accurate defect detection and risk assessment is achieved, reducing the burden on maintenance personnel.

CN118916804AActive Publication Date: 2024-11-08XIANGJIANG LAB

Patent Information

Application Number
CN202411407664.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-10
Publication Date
2024-11-08
Estimated Expiration
2044-10-10

AI Technical Summary

Technical Problem

In the identification of defect risk of main transformer, existing large models have problems such as high training costs, poor model updateability, and insufficient coverage of knowledge graphs and multimodal model features in the identification of defect risk, resulting in poor defect risk identification.

Method used

Using the main transformer defect risk identification method based on large models, collect multimodal data for preprocessing, build a knowledge graph and use machine learning methods to infer, introduce a multi-task learning framework to train multimodal large models, combined with entropy weight Fuzzy methods conduct risk assessment and provide decision support.

Benefits of technology

Improve the accuracy of defect detection, can understand the operating status and potential risks of the equipment more comprehensively, reduce the work burden of maintenance personnel, and significantly improve the complex relationship representation of the relationship between equipment through the combination of multimodal large model and knowledge graph. and processing capability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916804A_ABST
    Figure CN118916804A_ABST
Patent Text Reader

Abstract

The invention discloses a main transformer defect risk identification method based on a large model, and particularly relates to the technical field of large models, and the method comprises the steps: collecting the multi-modal data of a main transformer, and carrying out the preprocessing of the multi-modal data; constructing a knowledge graph of the main transformer, deducing the operation state and defect risk of the main transformer by applying a machine learning method, constructing a multi-modal large model according to the preprocessed multi-modal data and the knowledge graph, and introducing a multi-task learning framework to train defect identification and classification of the multi-modal large model; based on the output of the knowledge graph and the multi-modal large model, a fuzzy evaluation matrix is established according to the evaluation indexes of the risk state of the main transformer, the risk state of the main transformer is evaluated by combining an entropy weight fuzzy method, risk analysis is performed according to the evaluation result of the risk state of the main transformer, and decision support is provided for maintenance personnel according to the risk analysis. The defect risk distinguishing precision is effectively improved, and the problem that the model is not matched with the knowledge graph is avoided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of large models, and more specifically, to a main transformer defect risk identification method based on a large model. Background Art

[0002] Main transformer defect risk analysis is the process of identifying, evaluating and managing various defects that may occur in the transformer during operation. It aims to prevent and reduce the risks caused by transformer defects and ensure the safe and stable operation of the power system. Defect risk identification based on big models and knowledge graphs focuses on expanding and training big models and knowledge graphs themselves. The knowledge graph itself has good scalability, but the cost of model training is high, and the model may become outdated and invalid as the knowledge graph is updated. Therefore, the updatability of the model should be strengthened, and the effectiveness of the model in defect risk identification should be verified to avoid the problem of insufficient feature coverage of knowledge graphs and multimodal models.

[0003] In order to solve the above defects, a technical solution is now proposed. Summary of the invention

[0004] The purpose of the present invention is to provide a main transformer defect risk identification method based on a large model to solve the problems in the above-mentioned background technology.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solutions: a method for identifying main transformer defect risks based on a large model, the specific steps of which include collecting multimodal data of the main transformer and preprocessing the multimodal data, constructing a knowledge graph of the main transformer and using a machine learning method to infer the operating status and defect risk of the main transformer; A multimodal large model is constructed based on the preprocessed multimodal data and knowledge graph, and a multi-task learning framework is introduced to train the defect recognition and classification of the multimodal large model; Based on the output of knowledge graph and multimodal large model, a fuzzy evaluation matrix is ​​established according to the evaluation index of the risk status of the main transformer, and the risk status of the main transformer is evaluated by combining the entropy weight fuzzy method; Conduct risk analysis based on the assessment results of the risk status of the main transformer, and provide decision support to maintenance personnel based on the risk analysis.

[0006] Preferably, the method for preprocessing multimodal data is: Eliminate data records with outliers. For text data, remove punctuation noise. For missing values, the mean, median, or mode is used to fill them; Use image restoration technology to process the occluded or blurred parts in video or image data; The data containing time series information are aligned by timestamp to ensure the temporal consistency of data of different modalities; For image and video data, when multiple camera or sensor views are involved, spatial calibration and synchronization are performed to ensure spatial consistency of the data; Normalize the multimodal data. The normalization method is: , where x is the multimodal data before normalization, and y is the multimodal data after normalization; Perform feature extraction on multimodal data.

[0007] Preferably, feature processing is performed on different types of data in the multimodal data, and the processing method is: For text data, natural language processing technology is used to perform word segmentation, part-of-speech tagging, named entity recognition, extract keywords and phrases, and construct word vector representations; For image data, the convolutional neural network model is used to automatically extract the edge and texture features of the image; For audio data, the Mel-frequency cepstral coefficient method is used to extract the spectral characteristics and time domain characteristics of the audio; For the sensor data of the main transformer, the frequency domain features of the sensor data are extracted using fast Fourier transform, and the statistical features of the sensor data are extracted.

[0008] Preferably, the feature vectors of different multimodal data are concatenated to form a comprehensive feature vector, and the method for forming the comprehensive feature vector is: Convert text data output by natural language processing technology into fixed-length The feature vector of the image data extracted by the convolutional neural network model is converted into a fixed-length The feature vector of the audio data obtained by the Mel frequency cepstral coefficient method is converted into a fixed length The sensor data is normalized as a fixed-length feature vector The feature vectors of each feature vector are concatenated to form a comprehensive feature vector. , the total length of the comprehensive feature vector is .

[0009] Preferably, the method for constructing the knowledge graph is: Collect the operating data and technical specifications of the main transformer, use the graph database to model the data, define the entities and relationships between entities in the knowledge graph, convert the data into triples, and use natural language processing technology to automatically extract entities and relationships from documents, reports, and maintenance records; The knowledge graph is verified using actual transformer operation data, and is updated and expanded based on new accumulated data.

[0010] Preferably, based on the knowledge graph, the operating status and defect risk of the main transformer are inferred in combination with a machine learning method. The machine learning method used is a graph neural network, and the construction method of the graph neural network is: Represent the knowledge graph in the form of a graph, use the entities in the knowledge graph as nodes in the graph neural network, use the relationships between entities as edges in the graph neural network, and assign feature vectors to each node and edge; The feature matrix of all nodes’ feature vector combinations and the adjacency matrix representing the node connection relationship are used as the input of the graph neural network; The status of the main transformer is classified or risk assessed based on the output of the graph neural network.

[0011] Preferably, the process of inferring the operating status and defect risk of the main transformer is: The input includes a knowledge graph of the main transformer status, which includes node features and edge features and is represented in a graph form; After several layers of message passing and node updates, the graph neural network outputs the final state feature vector of each node; According to the output feature vector, the state of the main transformer is classified or evaluated to identify the defect state of the main transformer; Use the message passing mechanism of graph neural networks to propagate risk information from known risk nodes to other nodes; Through the output of the graph neural network, the nodes or edges with risks are identified and the results are output.

[0012] Preferably, based on the output of the knowledge graph and the multimodal large model, a fuzzy evaluation matrix is ​​established according to the evaluation index of the risk state of the main transformer, and the method for evaluating the risk state of the main transformer in combination with the entropy weight fuzzy method is as follows: The evaluation indicators of the main transformer include temperature, oil level, current, voltage, vibration level, insulation resistance, fault history, and environmental conditions; Define four levels of fuzzy linguistic variables as low risk, medium risk, high risk, and extremely high risk, and each fuzzy linguistic variable corresponds to a fuzzy membership function; For each evaluation indicator, the membership degree of the evaluation indicator under different fuzzy linguistic variables is calculated based on the actual measurement value or the output of the multimodal large model; Construct a fuzzy evaluation matrix R, and arrange the membership values ​​of each evaluation index under different risk states into a matrix form. Each row of the fuzzy evaluation matrix represents the membership distribution of an evaluation index, and each column represents the membership value under different risk states. For each evaluation indicator, the information entropy is calculated according to the membership distribution of the evaluation indicator. The calculation expression is: , where is a constant, is the membership degree of the jth evaluation indicator under the i-th risk state, n is the number of risk states, and the entropy weight of the evaluation indicator is calculated. The calculation expression is: , where m is the number of evaluation indicators; The fuzzy evaluation matrix R and the weight vector W are combined to obtain the fuzzy comprehensive evaluation matrix B. The expression of the fuzzy comprehensive evaluation matrix B is: , perform weighted summation on the fuzzy comprehensive evaluation matrix B to obtain the comprehensive membership vector V, and the calculation expression is: , where is the fuzzy evaluation matrix of the jth evaluation index; The risk status of the main transformer is determined according to the comprehensive membership vector V, and the risk status with the largest membership is taken as the final evaluation result and output.

[0013] Preferably, a risk analysis is performed based on the assessment result of the risk status of the main transformer, and decision support is provided to the maintenance personnel based on the risk analysis; According to the fuzzy comprehensive evaluation, the comprehensive membership vector V of the main transformer under different risk states is obtained. Combined with the total length of the comprehensive characteristic vector, the risk state with the highest membership is identified. The higher the comprehensive membership vector and the larger the total length of the comprehensive characteristic vector, the higher the credibility of the risk state with the highest membership. Use the knowledge graph to track and analyze the specific factors that lead to the current risk status. Combined with the output of the multimodal large model, verify the risk status according to the operating parameters of the main transformer. Use the relationships in the knowledge graph to analyze the chain reaction caused by the risk status, simulate the path of risk diffusion, predict the impact status of other components or systems, and identify high-risk areas or components. Determine the components or systems that require maintenance priority based on the risk analysis results. If the risk status is medium or high, prompt maintenance personnel to conduct detailed inspection and maintenance during the next planned shutdown. If the risk status is extremely high, maintenance personnel will be prompted to shut down the machine immediately and perform emergency maintenance to prevent the fault from expanding.

[0014] In the above technical solution, the technical effects and advantages provided by the present invention are: This application combines a multimodal big model with a knowledge graph. The knowledge graph can significantly represent the complex relationships between entities, integrate and uniformly represent data from different sources and in different formats, and improve the accuracy of defect detection by utilizing the correlation and complementarity between different modalities through a multimodal big model. The graph neural network can effectively process graph structured data. By combining a multimodal big model with a knowledge graph, it can more accurately capture the complex relationships between devices, between device parameters, and between them and environmental factors, thereby more comprehensively understanding the operating status and potential risks of the equipment, and providing decision support to maintenance personnel in an orderly and coherent manner, effectively reducing the burden on staff. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0016] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0017] 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.

[0018] Example 1: Please refer to Figure 1 As shown, the present invention is a main transformer defect risk identification method based on a large model, and the specific steps include collecting multimodal data of the main transformer and preprocessing the multimodal data, constructing a knowledge graph of the main transformer and using a machine learning method to infer the operating status and defect risk of the main transformer; A multimodal large model is constructed based on the preprocessed multimodal data and knowledge graph, and a multi-task learning framework is introduced to train the defect recognition and classification of the multimodal large model; Based on the output of knowledge graph and multimodal large model, a fuzzy evaluation matrix is ​​established according to the evaluation index of the risk status of the main transformer, and the risk status of the main transformer is evaluated by combining the entropy weight fuzzy method; Conduct risk analysis based on the assessment results of the risk status of the main transformer, and provide decision support to maintenance personnel based on the risk analysis.

[0019] Multimodal data is a collection of data that includes multiple different modes. The mode is the form of existence of the data or the source of information. Multimodal data includes the following types: Text data: including natural language text, numbers, symbols, etc.; Image data: data captured by visual sensors, containing visual information such as color, shape, texture, etc. Audio data: data captured by auditory sensors, including sound waveform, frequency, timbre and other characteristics; Video data: a data form that combines images and audio, a series of continuous image frames plus the corresponding audio stream; Other sensor data: including point cloud data from lidar, motion data from sensors, temperature data, pressure data, etc.; The characteristics of multimodal data are its richness and diversity, which can provide more comprehensive and accurate information. In practical applications, multimodal data fusion can integrate data from different sensors and different media. By comprehensively processing and analyzing multimodal data, it can achieve a deeper understanding and more accurate prediction of complex scenarios or situations.

[0020] The content of multimodal data includes equipment operation data, maintenance records, defect images, diagnostic reports, etc. The method for preprocessing multimodal data is as follows: Eliminate data records with outliers. For text data, remove punctuation noise. For missing values, the mean, median, or mode is used to fill them; Use image restoration technology to process the occluded or blurred parts in video or image data; The data containing time series information are aligned by timestamp to ensure the temporal consistency of data of different modalities; For image and video data, when multiple camera or sensor views are involved, spatial calibration and synchronization are performed to ensure spatial consistency of the data; Normalize the multimodal data. The normalization method is: , where x is the multimodal data before normalization, y is the multimodal data after normalization, and features are extracted from the multimodal data.

[0021] The feature processing of different types of data in multimodal data is performed by: For text data, natural language processing technology is used to perform word segmentation, part-of-speech tagging, named entity recognition, extract keywords and phrases, and construct word vector representations; For image data, the convolutional neural network model is used to automatically extract the edge and texture features of the image; For audio data, the Mel-frequency cepstral coefficient method is used to extract the spectral characteristics and time domain characteristics of the audio; For the sensor data of the main transformer, the frequency domain features of the sensor data are extracted using fast Fourier transform, and the statistical features of the sensor data are extracted.

[0022] Data containing time series information such as audio, video, sensor data, etc., image processing techniques used by convolutional neural network models such as edge detection and texture analysis, and statistical characteristics of sensor data including mean, variance, peak, etc.

[0023] The feature vectors of different multimodal data are concatenated to form a comprehensive feature vector. The method for forming the comprehensive feature vector is: Convert text data output by natural language processing technology into fixed-length The feature vector of the image data extracted by the convolutional neural network model is converted into a fixed-length The feature vector of the audio data obtained by the Mel frequency cepstral coefficient method is converted into a fixed length The sensor data is normalized as a fixed-length feature vector The feature vectors of each feature vector are concatenated to form a comprehensive feature vector. , the total length of the comprehensive feature vector is .

[0024] The comprehensive feature vector contains data features from different modes and can represent the overall operating status of the main transformer. The comprehensive feature vector is input into subsequent models, such as risk classifiers or prediction models, to obtain a comprehensive assessment of the current status of the transformer. The multimodal data fusion method can express different types of data features in a unified vector space through the splicing of feature vectors, providing a rich information basis for further analysis and decision-making.

[0025] Construct a knowledge graph of the main transformer and use machine learning methods to infer the operating status and defect risk of the main transformer. The construction method of the knowledge graph is as follows: Collect the operating data and technical specifications of the main transformer, use the graph database to model the data, define the entities and relationships between entities in the knowledge graph, convert the data into triples, and use natural language processing technology to automatically extract entities and relationships from documents, reports, and maintenance records; The knowledge graph is verified using actual transformer operation data, and is updated and expanded based on new accumulated data.

[0026] Common graph database tools include Neo4j, GraphDB, etc., and commonly used modeling languages ​​include RDF, OWL, etc.

[0027] The knowledge graph can integrate and uniformly represent data from different sources and in different formats, that is, multimodal data including text, images, videos, etc. In the risk assessment of the main transformer, various information such as historical fault records, real-time monitoring data, maintenance records, and expert experience of the equipment are incorporated into the knowledge graph to form a comprehensive knowledge base. This unified data representation method helps to reduce data silos and improve data utilization and consistency. The knowledge graph is based on graph theory and probabilistic graph models and can handle complex and diverse association analysis. In the main transformer risk assessment, by constructing the association between the main transformer and its related attributes, defect types, risk factors, etc., the interaction and influence between the various elements can be more intuitively displayed. This relationship expression ability helps to reveal potential risk patterns and trends, and provide a more accurate and comprehensive basis for risk assessment. The relevant attributes include voltage level, capacity, operating years, etc. The knowledge graph supports interactive exploratory analysis, allowing users to directly operate the nodes and edges in the knowledge graph through a graphical interface for flexible query and reasoning. In the main transformer risk assessment, users can conduct in-depth analysis of specific risk points as needed, or filter and sort according to risk type, severity and other conditions. Exploratory interactive analysis helps improve the efficiency and accuracy of risk assessment, and also makes risk assessment results easier to understand and interpret.

[0028] Based on the knowledge graph, the operating status and defect risk of the main transformer are inferred by combining machine learning methods. The machine learning method used is graph neural network. The construction method of graph neural network is as follows: Represent the knowledge graph in the form of a graph, use the entities in the knowledge graph as nodes in the graph neural network, use the relationships between entities as edges in the graph neural network, and assign feature vectors to each node and edge; The feature matrix of all nodes’ feature vector combinations and the adjacency matrix representing the node connection relationship are used as the input of the graph neural network; The status of the main transformer is classified or risk assessed based on the output of the graph neural network.

[0029] Each node and edge is given a feature vector. For example, the node coil contains features such as temperature and current, while the edge overheating leading to insulation breakdown is represented as a weight or risk level. The process of inferring the operating status and defect risk of the main transformer is: The input includes a knowledge graph of the main transformer status, which includes node features and edge features and is represented in a graph form; After several layers of message passing and node updates, the graph neural network outputs the final state feature vector of each node; According to the output feature vector, the state of the main transformer is classified or evaluated to identify the defect state of the main transformer; Use the message passing mechanism of graph neural networks to propagate risk information from known risk nodes to other nodes; Through the output of the graph neural network, the nodes or edges with risks are identified and the results are output.

[0030] Known risk nodes include detected defect states such as overheating, vibration, etc.

[0031] Knowledge graphs can significantly represent complex relationships between entities, while graph neural networks are good at processing such graph-structured data. By combining the two, we can more accurately capture the complex relationships between devices, device parameters, and their environmental factors, thereby more comprehensively understanding the operating status and potential risks of devices. Graph neural networks can learn deep feature representations of nodes through information transmission and aggregation mechanisms between nodes. The deep feature representation not only contains the information of the device itself, but also incorporates the information of other devices connected to it. Therefore, during the reasoning process, graph neural networks can more accurately infer the status and potential risks of the device; Since graph neural networks can learn universal representations of graph structured data, they generally have good generalization capabilities in different fields and different data sets. When using knowledge graphs and graph neural networks for equipment status monitoring and risk identification, the model can more easily adapt to new equipment and scenarios, avoiding a lot of retraining; Both knowledge graphs and graph neural networks have good scalability and can handle large-scale data sets. In practical applications, there are many devices and a large amount of data, and traditional reasoning methods may be difficult to cope with. The method that combines knowledge graphs and graph neural networks can efficiently process this data and achieve fast and accurate reasoning; Compared with other deep learning models, graph neural networks provide more explainability in the reasoning process. By visualizing the nodes and edges in the graph structure and the information transfer process between them, it is easier to understand how the model makes decisions. It is more suitable for application in the industrial field; In real applications, the status and potential risks of equipment are constantly changing. The combined knowledge graph and graph neural network approach supports dynamic updates of the knowledge graph and model parameters to reflect the changes. It helps to add new data or information to the knowledge graph and retrain or fine-tune the graph neural network model to update its reasoning ability.

[0032] Based on the output of knowledge graph and multimodal large model, a fuzzy evaluation matrix is ​​established according to the evaluation index of the risk status of the main transformer, and the risk status of the main transformer is evaluated by combining the entropy weight fuzzy method; The evaluation indicators of the main transformer include temperature, oil level, current, voltage, vibration level, insulation resistance, fault history, and environmental conditions; Define four levels of fuzzy linguistic variables as low risk, medium risk, high risk, and extremely high risk, and each fuzzy linguistic variable corresponds to a fuzzy membership function; For each evaluation indicator, the membership degree of the evaluation indicator under different fuzzy linguistic variables is calculated based on the actual measurement value or the output of the multimodal large model; Construct a fuzzy evaluation matrix R, and arrange the membership values ​​of each evaluation index under different risk states into a matrix form. Each row of the fuzzy evaluation matrix represents the membership distribution of an evaluation index, and each column represents the membership value under different risk states. For each evaluation indicator, the information entropy is calculated according to the membership distribution of the evaluation indicator. The calculation expression is: , where is a constant, is the membership degree of the jth evaluation indicator under the i-th risk state, n is the number of risk states, and the entropy weight of the evaluation indicator is calculated. The calculation expression is: , where m is the number of evaluation indicators; The fuzzy evaluation matrix R and the weight vector W are combined to obtain the fuzzy comprehensive evaluation matrix B. The expression of the fuzzy comprehensive evaluation matrix B is: , perform weighted summation on the fuzzy comprehensive evaluation matrix B to obtain the comprehensive membership vector V, and the calculation expression is: , where is the fuzzy evaluation matrix of the jth evaluation index; The risk status of the main transformer is determined according to the comprehensive membership vector V, and the risk status with the largest membership is taken as the final evaluation result and output.

[0033] Temperature includes coil temperature and oil temperature, fault history includes overheating and short circuit faults that have occurred, and environmental conditions include humidity and air temperature; In a specific example, for the temperature index, several temperature intervals are set and a corresponding degree of membership is set for each temperature interval.

[0034] Conduct risk analysis based on the assessment results of the risk status of the main transformer and provide decision support to maintenance personnel based on the risk analysis; According to the fuzzy comprehensive evaluation, the comprehensive membership vector V of the main transformer under different risk states is obtained. Combined with the total length of the comprehensive characteristic vector, the risk state with the highest membership is identified. The higher the comprehensive membership vector and the larger the total length of the comprehensive characteristic vector, the higher the credibility of the risk state with the highest membership. Use the knowledge graph to track and analyze the specific factors that lead to the current risk status. Combined with the output of the multimodal large model, verify the risk status according to the operating parameters of the main transformer. Use the relationships in the knowledge graph to analyze the chain reaction caused by the risk status, simulate the path of risk diffusion, predict the impact status of other components or systems, and identify high-risk areas or components. Determine the components or systems that require maintenance priority based on the risk analysis results. If the risk status is medium or high, prompt maintenance personnel to conduct detailed inspection and maintenance during the next planned shutdown. If the risk status is extremely high, maintenance personnel will be prompted to shut down the machine immediately and perform emergency maintenance to prevent the fault from expanding.

[0035] Combine the risk status assessment results with the online monitoring system, set warning thresholds, dynamically adjust risk assessment and maintenance strategies based on changes in real-time data, and generate a risk analysis report including the current risk status, major risk factors, potential failure paths, and recommended maintenance measures.

[0036] Dynamically adjust risk assessment and maintenance strategies. For example, as the temperature changes, the risk status assessment results are updated in real time and the maintenance plan is adjusted accordingly.

[0037] This application combines a multimodal big model with a knowledge graph. The knowledge graph can significantly represent the complex relationships between entities, integrate and uniformly represent data from different sources and in different formats, and improve the accuracy of defect detection by utilizing the correlation and complementarity between different modalities through a multimodal big model. The graph neural network can effectively process graph structured data. By combining a multimodal big model with a knowledge graph, it can more accurately capture the complex relationships between devices, between device parameters, and between them and environmental factors, thereby more comprehensively understanding the operating status and potential risks of the equipment, and providing decision support to maintenance personnel in an orderly and coherent manner, effectively reducing the burden on staff.

[0038] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0039] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of computer program goods. The computer program goods include one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0040] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0041] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0042] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0043] If the functions are implemented in the form of software functional units and sold or used as independent goods, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the prior art or the part of the technical solution can be embodied in the form of software goods, and the computer software goods are stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0044] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A main transformer defect risk identification method based on a large model, characterized in that: The specific steps include collecting multimodal data of the main transformer and preprocessing the multimodal data, building a knowledge graph of the main transformer and using machine learning methods to infer the operating status and defect risks of the main transformer; A multimodal large model is constructed based on the preprocessed multimodal data and knowledge graph, and a multi-task learning framework is introduced to train the defect recognition and classification of the multimodal large model; Based on the output of knowledge graph and multimodal large model, a fuzzy evaluation matrix is ​​established according to the evaluation index of the risk status of the main transformer, and the risk status of the main transformer is evaluated by combining the entropy weight fuzzy method; Conduct risk analysis based on the assessment results of the risk status of the main transformer, and provide decision support to maintenance personnel based on the risk analysis.

2. The main transformer defect risk identification method based on a large model according to claim 1 is characterized in that: The method for preprocessing multimodal data is: Eliminate data records with outliers. For text data, remove punctuation noise. For missing values, the mean, median, or mode is used to fill them; Use image restoration technology to process the occluded or blurred parts in video or image data; The data containing time series information are aligned by timestamp to ensure the temporal consistency of data of different modalities; For image and video data, when multiple camera or sensor views are involved, spatial calibration and synchronization are performed to ensure spatial consistency of the data; Normalize the multimodal data. The normalization method is: , where x is the multimodal data before normalization, and y is the multimodal data after normalization; Perform feature extraction on multimodal data.

3. The main transformer defect risk identification method based on a large model according to claim 2 is characterized in that: The feature processing of different types of data in multimodal data is performed by: For text data, natural language processing technology is used to perform word segmentation, part-of-speech tagging, named entity recognition, extract keywords and phrases, and construct word vector representations; For image data, the convolutional neural network model is used to automatically extract the edge and texture features of the image; For audio data, the Mel-frequency cepstral coefficient method is used to extract the spectral characteristics and time domain characteristics of the audio; For the sensor data of the main transformer, the frequency domain features of the sensor data are extracted using fast Fourier transform, and the statistical features of the sensor data are extracted.

4. The main transformer defect risk identification method based on a large model according to claim 3 is characterized in that: The feature vectors of different multimodal data are concatenated to form a comprehensive feature vector. The method for forming the comprehensive feature vector is: Convert text data output by natural language processing technology into fixed-length The feature vector of the image data extracted by the convolutional neural network model is converted into a fixed-length The feature vector of the audio data obtained by the Mel frequency cepstral coefficient method is converted into a fixed length The sensor data is normalized as a fixed-length feature vector The feature vectors of each feature vector are concatenated to form a comprehensive feature vector. , the total length of the comprehensive feature vector is .

5. The main transformer defect risk identification method based on a large model according to claim 1 is characterized in that: The construction method of the knowledge graph is: Collect the operating data and technical specifications of the main transformer, use the graph database to model the data, define the entities and relationships between entities in the knowledge graph, convert the data into triples, and use natural language processing technology to automatically extract entities and relationships from documents, reports, and maintenance records; The knowledge graph is verified using actual transformer operation data, and is updated and expanded based on new accumulated data.

6. The main transformer defect risk identification method based on a large model according to claim 5 is characterized in that: Based on the knowledge graph, the operating status and defect risk of the main transformer are inferred by combining machine learning methods. The machine learning method used is graph neural network. The construction method of graph neural network is as follows: Represent the knowledge graph in the form of a graph, use the entities in the knowledge graph as nodes in the graph neural network, use the relationships between entities as edges in the graph neural network, and assign feature vectors to each node and edge; The feature matrix of the feature vector combination of all nodes and the adjacency matrix representing the node connection relationship are used as the input of the graph neural network; The status of the main transformer is classified or risk assessed based on the output of the graph neural network.

7. The main transformer defect risk identification method based on a large model according to claim 6 is characterized in that: The process of inferring the operating status and defect risk of the main transformer is: The input includes the knowledge graph of the main transformer state, which includes node features and edge features and is represented in graph form; After several layers of message passing and node updates, the graph neural network outputs the final state feature vector of each node; According to the output feature vector, the state of the main transformer is classified or evaluated to identify the defect state of the main transformer; Use the message passing mechanism of graph neural networks to propagate risk information from known risk nodes to other nodes; Through the output of the graph neural network, the nodes or edges with risks are identified and the results are output.

8. The main transformer defect risk identification method based on a large model according to claim 1 is characterized in that: Based on the output of the knowledge graph and the multimodal large model, a fuzzy evaluation matrix is ​​established according to the evaluation indicators of the risk status of the main transformer. The method for evaluating the risk status of the main transformer combined with the entropy weight fuzzy method is as follows: The evaluation indicators of the main transformer include temperature, oil level, current, voltage, vibration level, insulation resistance, fault history, and environmental conditions; Define four levels of fuzzy linguistic variables as low risk, medium risk, high risk, and extremely high risk, and each fuzzy linguistic variable corresponds to a fuzzy membership function; For each evaluation indicator, the membership degree of the evaluation indicator under different fuzzy linguistic variables is calculated based on the actual measurement value or the output of the multimodal large model; Construct a fuzzy evaluation matrix R, and arrange the membership values ​​of each evaluation index under different risk states into a matrix form. Each row of the fuzzy evaluation matrix represents the membership distribution of an evaluation index, and each column represents the membership value under different risk states. For each evaluation indicator, the information entropy is calculated according to the membership distribution of the evaluation indicator. The calculation expression is: , where is a constant, is the membership degree of the jth evaluation indicator under the i-th risk state, n is the number of risk states, and the entropy weight of the evaluation indicator is calculated. The calculation expression is: , where m is the number of evaluation indicators; The fuzzy evaluation matrix R and the weight vector W are combined to obtain the fuzzy comprehensive evaluation matrix B. The expression of the fuzzy comprehensive evaluation matrix B is: , perform weighted summation on the fuzzy comprehensive evaluation matrix B to obtain the comprehensive membership vector V, and the calculation expression is: , where is the fuzzy evaluation matrix of the jth evaluation index; The risk status of the main transformer is determined according to the comprehensive membership vector V, and the risk status with the largest membership is taken as the final evaluation result and output.

9. The main transformer defect risk identification method based on a large model according to claim 8 is characterized in that: Conduct risk analysis based on the assessment results of the risk status of the main transformer and provide decision support to maintenance personnel based on the risk analysis; According to the fuzzy comprehensive evaluation, the comprehensive membership vector V of the main transformer under different risk states is obtained. Combined with the total length of the comprehensive characteristic vector, the risk state with the highest membership is identified. The higher the comprehensive membership vector and the larger the total length of the comprehensive characteristic vector, the higher the credibility of the risk state with the highest membership. Use the knowledge graph to track and analyze the specific factors that lead to the current risk status. Combined with the output of the multimodal large model, verify the risk status according to the operating parameters of the main transformer. Use the relationships in the knowledge graph to analyze the chain reaction caused by the risk status, simulate the path of risk diffusion, predict the impact status of other components or systems, and identify high-risk areas or components. Determine the components or systems that require maintenance priority based on the risk analysis results. If the risk status is medium or high, prompt maintenance personnel to conduct detailed inspection and maintenance during the next planned shutdown. If the risk status is extremely high, maintenance personnel will be prompted to shut down the machine immediately and perform emergency maintenance to prevent the fault from expanding.

Citation Information

Patent Citations

  • Multi-modal knowledge graph method based on power grid dispatching

    CN117171358A

  • Fire-fighting equipment detection and evaluation system and method based on video processing and deep learning

    CN118095867A

  • Transformer fault prediction method

    CN118690247A

Cited By

  • Risk monitoring method, device and equipment of source load storage network, medium and program product

    CN119647977A

  • Robot control method based on multi-modal large model

    CN119681909A

  • Abnormality identification method for multi-modal data of main transformer operation

    CN119939456A

  • Elevator fault classification method and system

    CN120004089A

  • Laser cladding knowledge base construction method based on large model

    CN120373435A