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

Abstract

The invention belongs to the technical field of data exception identification, and particularly relates to a main transformer operation multi-mode data exception identification method, which comprises the following steps: step 1, data acquisition: collecting various data in operation of a main transformer, including current, voltage, temperature, vibration and noise; step 2, data preprocessing: cleaning and sorting the collected data, including noise removal, missing value filling and data smoothing operation; step 3, feature extraction: extracting key features, including statistical features and frequency domain features, from the preprocessed data, and converting the original data into a form for analysis; step 4, modeling and training: utilizing a machine learning or deep learning model to train the features, wherein the model comprises a decision tree and a support vector machine; according to the invention, through a multi-level and multi-dimensional anomaly detection and processing mechanism, the safety and reliability of the operation of the transformer are improved, and the efficiency and cost of operation and maintenance management are optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of data anomaly identification, and in particular relates to an anomaly identification method for multi-modal data of main transformer operation. Background Art

[0002] The multimodal data of main transformer operation is a comprehensive concept that covers a wide range of data types captured by various sensors and advanced monitoring equipment during transformer operation. These data not only involve various aspects of the electrical field, but also include key indicators such as voltage fluctuations, current intensity, power output, and frequency stability, which together depict the electrical performance and operating status of the transformer. At the same time, temperature data also plays an important role, such as the temperature change of transformer oil and the temperature distribution of windings. These data can accurately reflect the thermal state of the transformer, which is crucial for preventing failures caused by overheating.

[0003] In addition, acoustic data and vibration data are also important in the condition monitoring of the main transformer. By analyzing the characteristics of the sound waves generated when the transformer is running, the changes in the sound pattern under abnormal conditions can be captured, providing a strong basis for fault diagnosis. Vibration data can reveal the working status of the mechanical components inside the transformer, and any abnormal vibration pattern may be an early signal of mechanical failure.

[0004] In addition to the above data, environmental data is also an indispensable part. Environmental factors such as humidity and temperature have a direct impact on the operation of the transformer. Too high or too low humidity and extreme temperature may cause the transformer to deteriorate in performance or even fail.

[0005] The existing multimodal data of main transformer operation generally adopts long-term recording and manual input for anomaly identification. The timeliness of anomaly detection is poor, which is not convenient for timely discovery of potential problems in operation. The expansion of faults has a greater impact on the system, the delay of human intervention is large, the stability and reliability of the system are poor, and the complexity of manual analysis further increases the difficulty of anomaly detection, which is not conducive to the anomaly identification of multimodal data of main transformer operation. For this reason, we propose an anomaly identification method for multimodal data of main transformer operation to solve the above problems. Summary of the invention

[0006] The purpose of the present invention is to provide a method for identifying anomalies in multimodal data of main transformer operation, which can not only improve the safety and reliability of transformer operation, but also optimize the efficiency and cost of operation and maintenance management through a multi-level and multi-dimensional anomaly detection and processing mechanism.

[0007] The technical solution adopted by the present invention is as follows:

[0008] A method for identifying abnormalities in multi-modal data of main transformer operation, the method comprising the following steps:

[0009] Step 1. Data collection: Collect various data during the operation of the main transformer, including current, voltage, temperature, vibration, and noise;

[0010] Step 2. Data preprocessing: Clean and organize the collected data, including removing noise, filling missing values, and data smoothing operations;

[0011] Step 3. Feature extraction: extract key features from the preprocessed data, including statistical features and frequency domain features, and convert the raw data into a form for analysis;

[0012] Step 4. Modeling and training: Use machine learning or deep learning models to train features. Models include decision trees, support vector machines, and neural networks.

[0013] Step 5. Anomaly detection: Use the trained model to predict real-time data, identify abnormal data that does not conform to the normal operating mode, and use threshold setting and anomaly scoring methods to determine the degree of anomaly;

[0014] Step 6. Anomaly diagnosis: Conduct in-depth analysis of detected anomalies, identify the causes of anomalies, and analyze them in combination with expert knowledge and system operation status;

[0015] Step 7. Alarm and response: When an abnormality is detected, the system generates an alarm signal to notify the operator to check or take measures. The response measures include shutting down for maintenance and adjusting operating parameters.

[0016] Step 8. System feedback and optimization: Adjust and optimize the model and anomaly detection algorithm based on actual operation conditions and feedback information;

[0017] Step 9. Result evaluation and model validation: Perform a systematic evaluation of the anomaly detection results, using cross-validation and independent test sets to evaluate the model's performance indicators, including accuracy, precision, recall, and F1-score.

[0018] In a preferred embodiment, the data acquisition also includes determining the frequency and accuracy requirements of data acquisition, selecting corresponding types of sensors according to the variables to be monitored, installing the sensors at key locations of the main transformer, including windings, current measurement points, and temperature measurement points, configuring the data acquisition system, including selecting appropriate data acquisition instruments and setting acquisition parameters, the sensors transmit real-time measured data to the data acquisition system, monitoring the data acquisition process in real time, and storing the collected data in a database or file system.

[0019] In a preferred embodiment, the data preprocessing also includes checking whether there are duplicate records in the data set, filling missing values ​​using interpolation methods, including mean interpolation, median interpolation, and forward / backward filling, predicting missing values ​​using a regression model, detecting outliers through standard deviation and quartiles, processing outliers, and correcting them to values ​​within a reasonable range, converting data to the same scale using Z-score standardization, converting data to a unified format or unit, performing sliding window averaging on data, assigning higher weights to recent data, using filters to remove high-frequency noise, retaining the main trends of the data, dividing the data set into a training set, a validation set, and a test set, and merging data from different sources or at different times.

[0020] In a preferred embodiment, the feature extraction also includes converting time domain data into frequency domain data, determining the maximum and minimum values ​​in the data, describing the range of the signal, finding the highest point of the signal, calculating the difference between the maximum and minimum values, describing the amplitude of the signal, performing wavelet transform on the time series data, generating a time-frequency graph, displaying the distribution of the signal in time and frequency, selecting features from the extracted features through correlation analysis and feature importance evaluation, reducing high-dimensional feature data to low-dimensional space, combining various extracted features into a comprehensive feature vector, and saving the final feature data to a database or file.

[0021] In a preferred embodiment, the modeling and training also include making decisions through a tree structure, performing classification or regression by finding the optimal separating hyperplane, performing complex nonlinear mapping by simulating a biological nervous system, using training set data to train the model, iteratively optimizing model parameters, adjusting model hyperparameters on a validation set, using grid search and random search to find the best hyperparameter combination, using accuracy, precision, recall, and F1 score indicators to evaluate the classification performance of the model, using mean square error, root mean square error, and determination coefficient indicators to evaluate the regression performance of the model, using a cross-validation method to further verify the generalization ability of the model, evaluating the final performance of the model on a test set, and adjusting the model structure or algorithm based on the evaluation results.

[0022] In a preferred embodiment, the anomaly detection also includes setting up a data inflow mechanism to transmit real-time data from a source to a detection system, processing noise and errors in the data in real time, performing feature calculations on real-time data, using a trained model to predict or score real-time features, comparing the anomaly score output by the model with a preset threshold, marking the data point as an anomaly when the score or prediction result output by the model exceeds the threshold, triggering a corresponding alarm according to the level of the anomaly, setting up a multi-level alarm system, sending different levels of notifications according to the severity of the anomaly, and configuring the system to automatically execute preset response operations when an anomaly is detected, including shutting down equipment, switching to a backup system, and adjusting operating parameters.

[0023] In a preferred embodiment, the abnormality diagnosis also includes confirming whether the detected abnormality is real and valid, eliminating false alarms, recording basic information of the abnormality, including the time when the abnormality occurred, the type of abnormality, relevant data points and model output, collecting all relevant data before and after the abnormality occurs, including system logs, sensor data, and operation records, integrating the collected relevant data, comparing the characteristics of the abnormal data with the characteristics of normal data, analyzing the trends of the abnormal data and related characteristics, comparing the abnormal situation with historical data and known failure modes, verifying the initially identified cause, and confirming the cause through simulation, testing or adjustment of system parameters.

[0024] In a preferred embodiment, the alarm and response also includes the system continuously monitoring key indicators and data streams through a built-in anomaly detection mechanism or model, detecting data anomalies or exceeding preset thresholds, and triggering an anomaly detection signal. The anomaly detection signal is further analyzed to determine whether it is necessary to generate an alarm signal, including comparison with historical data and identification of the anomaly type. The system generates an alarm signal, which includes the anomaly type, time, and location. The alarm signal is immediately sent to the operator through a preset notification channel, which includes sound and light alarms, text messages, emails, and mobile application push. After receiving the alarm signal, the operator checks to confirm the authenticity and severity of the anomaly, and selects response measures based on the type and urgency of the anomaly. The response measures include shutdown for maintenance, adjustment of operating parameters, and startup of backup equipment.

[0025] In a preferred embodiment, the system feedback and optimization also includes real-time monitoring of system performance, including the accuracy, response speed and processing effect of anomaly detection, recording detailed information of abnormal events, including the frequency, type, processing measures and results of occurrence, collecting feedback from operators and system users, including satisfaction with the accuracy of system alarms, timeliness of notifications and processing suggestions, diagnosing problems in the system based on the collected feedback information, adjusting and optimizing the anomaly detection algorithm based on the feedback analysis results, retraining the model, using the latest data and feedback information to optimize the model, analyzing abnormal data and feedback data, identifying abnormal patterns and system performance bottlenecks, and evaluating the performance of the current model, including detection accuracy, recall rate, and F1 score indicators.

[0026] In a preferred embodiment, the result evaluation and model verification also include retrospective analysis based on real-time operation data and historical data to verify the robustness and adaptability of the model, and based on the evaluation results, targeted adjustment of feature selection and model parameters to optimize the performance of the algorithm.

[0027] The technical effects achieved by the present invention are:

[0028] Combining multiple data sources such as current, voltage, temperature, vibration, and noise, it can capture the operating status of the main transformer more comprehensively and improve the accuracy of anomaly detection. By extracting different features such as time domain and frequency domain, it can better identify potential abnormal patterns and improve the sensitivity and accuracy of anomaly detection. Through real-time data collection and anomaly detection, it can timely discover potential problems in operation, avoid the expansion of faults, and reduce the impact of equipment failures on the system. The anomaly detection system can automatically trigger an alarm and take corresponding response measures, such as shutting down for maintenance or adjusting parameters, thereby reducing the delay of human intervention and improving the stability and reliability of the system.

[0029] Through effective anomaly detection, potential problems can be identified in advance, thereby reducing the frequency and severity of equipment failures and reducing maintenance and downtime costs. The system's feedback and optimization functions can help formulate more scientific maintenance plans, adjust maintenance strategies according to actual operating conditions, and improve resource utilization efficiency. Through pre-processing steps such as denoising, filling missing values, and data standardization, the quality and consistency of data can be improved, providing more reliable input data for model training. Machine learning or deep learning models are used to automatically extract features and model, reducing the complexity and time investment of manual analysis;

[0030] The system can conduct in-depth analysis of detected anomalies, identify the causes of anomalies, and conduct comprehensive evaluations based on expert knowledge to provide strong support for decision-making. Through analysis of system performance and user feedback, it can continuously optimize anomaly detection algorithms and models to improve the overall performance of the system. The automated alarm and response mechanism can reduce the burden on operators and improve their monitoring and maintenance efficiency of the system. Through a variety of data display methods, operators can more intuitively understand the system status and abnormal conditions, making it easier to respond quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of a method for identifying abnormalities in multi-modal data of main transformer operation according to the present invention. DETAILED DESCRIPTION

[0032] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0033] See also Figure 1 As shown, the present invention provides a method for identifying abnormalities in multi-modal data of main transformer operation, and the abnormality identification method comprises the following steps:

[0034] Step 1. Data collection: Collect various data during the operation of the main transformer, including current, voltage, temperature, vibration, and noise;

[0035] Step 2. Data preprocessing: Clean and organize the collected data, including removing noise, filling missing values, and data smoothing operations;

[0036] Step 3. Feature extraction: extract key features from the preprocessed data, including statistical features and frequency domain features, and convert the raw data into a form for analysis;

[0037] Step 4. Modeling and training: Use machine learning or deep learning models to train features. Models include decision trees, support vector machines, and neural networks.

[0038] Step 5. Anomaly detection: Use the trained model to predict real-time data, identify abnormal data that does not conform to the normal operating mode, and use threshold setting and anomaly scoring methods to determine the degree of anomaly;

[0039] Step 6. Anomaly diagnosis: Conduct in-depth analysis of detected anomalies, identify the causes of anomalies, and analyze them in combination with expert knowledge and system operation status;

[0040] Step 7. Alarm and response: When an abnormality is detected, the system generates an alarm signal to notify the operator to check or take measures. The response measures include shutting down for maintenance and adjusting operating parameters.

[0041] Step 8. System feedback and optimization: Adjust and optimize the model and anomaly detection algorithm based on actual operation conditions and feedback information;

[0042] Step 9. Result evaluation and model validation: Perform a systematic evaluation of the anomaly detection results, using cross-validation and independent test sets to evaluate the model's performance indicators, including accuracy, precision, recall, and F1-score.

[0043] Data acquisition also includes determining the frequency and accuracy requirements of data acquisition, selecting the corresponding type of sensor according to the variables to be monitored, installing the sensors at key locations of the main transformer, including windings, current measurement points, and temperature measurement points, configuring the data acquisition system, including selecting appropriate data acquisition instruments and setting acquisition parameters, the sensors transmitting real-time measured data to the data acquisition system, monitoring the data acquisition process in real time, and storing the collected data in a database or file system.

[0044] Data preprocessing also includes checking whether there are duplicate records in the dataset, using interpolation methods to fill missing values, including mean interpolation, median interpolation, and forward / backward interpolation, using regression models to predict missing values, detecting outliers through standard deviation and quartiles, processing outliers, and correcting them to values ​​in a reasonable range, using Z-score standardization to convert data to the same scale, converting data to a unified format or unit, performing sliding window averaging on data, giving higher weight to recent data, using filters to remove high-frequency noise, retaining the main trends of the data, dividing the dataset into training, validation, and test sets, and merging data from different sources or at different times.

[0045] Feature extraction also includes converting time domain data into frequency domain data, determining the maximum and minimum values ​​in the data, describing the range of the signal, finding the highest point of the signal, calculating the difference between the maximum and minimum values, describing the amplitude of the signal, performing wavelet transform on time series data, generating time-frequency graphs, displaying the distribution of the signal in time and frequency, selecting features from the extracted features through correlation analysis and feature importance assessment, reducing high-dimensional feature data to low-dimensional space, combining the extracted features into a comprehensive feature vector, and saving the final feature data to a database or file.

[0046] Modeling and training also include making decisions through tree structures, classification or regression by finding the best separating hyperplane, complex nonlinear mapping by simulating biological nervous systems, using training set data to train models, iteratively optimizing model parameters, adjusting model hyperparameters on the validation set, using grid search and random search to find the best hyperparameter combination, using accuracy, precision, recall, and F1 score indicators to evaluate the classification performance of the model, using mean square error, root mean square error, and determination coefficient indicators to evaluate the regression performance of the model, using cross-validation methods to further verify the generalization ability of the model, evaluating the final performance of the model on the test set, and adjusting the model structure or algorithm based on the evaluation results.

[0047] Anomaly detection also includes setting up a data inflow mechanism to transfer real-time data from the source to the detection system, processing noise and errors in the data in real time, performing feature calculations on real-time data, using trained models to predict or score real-time features, comparing the anomaly score output by the model with a preset threshold, marking the data point as an anomaly when the score or prediction result output by the model exceeds the threshold, triggering corresponding alarms based on the level of the anomaly, setting up a multi-level alarm system, sending different levels of notifications based on the severity of the anomaly, and configuring the system to automatically perform preset response operations when an anomaly is detected, including shutting down equipment, switching to a backup system, and adjusting operating parameters.

[0048] Abnormal diagnosis also includes confirming whether the detected abnormalities are real and valid, eliminating false alarms, recording basic information about the abnormalities, including the time when the abnormality occurred, the type of abnormality, relevant data points and model outputs, collecting relevant data before and after the abnormality occurs, including system logs, sensor data, and operation records, integrating the collected relevant data, comparing the characteristics of the abnormal data with those of normal data, analyzing the trends of abnormal data and related characteristics, comparing the abnormal conditions with historical data and known failure modes, verifying the initially identified causes, and confirming the causes through simulation, testing or adjusting system parameters.

[0049] Alarm and response also includes the system's built-in anomaly detection mechanism or model, which continuously monitors key indicators and data streams, detects data anomalies or exceeds preset thresholds, and triggers anomaly detection signals. The anomaly detection signals are further analyzed to determine whether an alarm signal needs to be generated, including comparison with historical data and identification of anomaly types. The system generates an alarm signal, which includes the type, time, and location of the anomaly. The alarm signal is immediately sent to the operator through preset notification channels, which include sound and light alarms, text messages, emails, and mobile application push. After receiving the alarm signal, the operator checks to confirm the authenticity and severity of the anomaly, and selects response measures based on the type and urgency of the anomaly. Response measures include shutdown for maintenance, adjustment of operating parameters, and startup of backup equipment.

[0050] System feedback and optimization also include real-time monitoring of system performance, including the accuracy of anomaly detection, response speed and processing effect, recording detailed information of abnormal events, including the frequency, type, processing measures and results of occurrence, collecting feedback from operators and system users, including satisfaction with the accuracy of system alarms, timeliness of notifications and processing suggestions, diagnosing problems in the system based on the collected feedback information, adjusting and optimizing anomaly detection algorithms based on feedback analysis results, retraining models, optimizing models using the latest data and feedback information, analyzing abnormal data and feedback data, identifying abnormal patterns and system performance bottlenecks, and evaluating the performance of the current model, including detection accuracy, recall rate, and F1 score indicators;

[0051] Result evaluation and model verification also include retrospective analysis based on real-time operation data and historical data to verify the robustness and adaptability of the model. Based on the evaluation results, feature selection and model parameters are adjusted in a targeted manner to optimize the performance of the algorithm.

[0052] In the present invention, by combining multiple data sources such as current, voltage, temperature, vibration, noise, etc., the operating status of the main transformer can be captured more comprehensively, and the accuracy of abnormality detection can be improved. By extracting different features such as time domain and frequency domain, potential abnormal modes can be better identified, the expansion of faults can be avoided, and the impact of equipment failures on the system can be reduced. The abnormality detection system can automatically trigger an alarm and take corresponding response measures, such as shutting down for maintenance or adjusting parameters, thereby reducing the delay of human intervention and improving the stability and reliability of the system.

[0053] Through effective anomaly detection, potential problems can be identified in advance, thereby reducing the frequency and severity of equipment failures and reducing maintenance and downtime costs. The system's feedback and optimization functions can help formulate more scientific maintenance plans, adjust maintenance strategies according to actual operating conditions, and improve resource utilization efficiency. Through pre-processing steps such as denoising, filling missing values, and data standardization, the quality and consistency of data can be improved, providing more reliable input data for model training. Machine learning or deep learning models are used to automatically extract features and model, reducing the complexity and time investment of manual analysis;

[0054] The system can conduct in-depth analysis of detected anomalies, identify the causes of anomalies, and conduct comprehensive evaluations based on expert knowledge to provide strong support for decision-making. Through analysis of system performance and user feedback, it can continuously optimize anomaly detection algorithms and models to improve its monitoring and maintenance efficiency of the system. Through a variety of data display methods, operators can more intuitively understand the system status and abnormal situations, making it easier to respond quickly.

[0055] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. A method for identifying abnormalities in multi-modal data of main transformer operation, characterized in that: The abnormality identification method comprises the following steps: Step 1. Data collection: Collect various data during the operation of the main transformer, including current, voltage, temperature, vibration, and noise; Step 2. Data preprocessing: Clean and organize the collected data, including removing noise, filling missing values, and data smoothing operations; Step 3. Feature extraction: extract key features from the preprocessed data, including statistical features and frequency domain features, and convert the raw data into a form for analysis; Step 4. Modeling and training: Use machine learning or deep learning models to train features. Models include decision trees, support vector machines, and neural networks. Step 5. Anomaly detection: Use the trained model to predict real-time data, identify abnormal data that does not conform to the normal operating mode, and use threshold setting and anomaly scoring methods to determine the degree of anomaly; Step 6. Anomaly diagnosis: Conduct in-depth analysis of detected anomalies, identify the causes of anomalies, and analyze them in combination with expert knowledge and system operation status; Step 7. Alarm and response: When an abnormality is detected, the system generates an alarm signal to notify the operator to check or take measures. The response measures include shutting down for maintenance and adjusting operating parameters. Step 8. System feedback and optimization: Adjust and optimize the model and anomaly detection algorithm based on actual operation conditions and feedback information. Step 9. Result evaluation and model validation: Perform a systematic evaluation of the anomaly detection results, using cross-validation and independent test sets to evaluate the model's performance indicators, including accuracy, precision, recall, and F1-score.

2. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1 is characterized in that: The data acquisition also includes determining the frequency and accuracy requirements of data acquisition, selecting corresponding types of sensors according to the variables to be monitored, installing the sensors at key locations of the main transformer, including windings, current measurement points, and temperature measurement points, configuring the data acquisition system, including selecting appropriate data acquisition instruments and setting acquisition parameters, the sensors transmitting real-time measured data to the data acquisition system, monitoring the data acquisition process in real time, and storing the collected data in a database or file system.

3. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1 is characterized in that: The data preprocessing also includes checking whether there are duplicate records in the data set, using interpolation methods to fill missing values, including mean interpolation, median interpolation, and forward / backward filling, using regression models to predict missing values, detecting outliers through standard deviations and quartiles, processing outliers, and correcting them to values ​​within a reasonable range, using Z-score standardization to convert data to the same scale, converting data to a unified format or unit, performing sliding window averaging on data, giving higher weights to recent data, using filters to remove high-frequency noise, retaining the main trends of the data, dividing the data set into training sets, validation sets, and test sets, and merging data from different sources or at different times.

4. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1 is characterized in that: The feature extraction also includes converting time domain data into frequency domain data, determining the maximum and minimum values ​​in the data, describing the range of the signal, finding the highest point of the signal, calculating the difference between the maximum and minimum values, describing the amplitude of the signal, performing wavelet transform on the time series data, generating a time-frequency graph, displaying the distribution of the signal in time and frequency, selecting features from the extracted features through correlation analysis and feature importance evaluation, reducing high-dimensional feature data to low-dimensional space, combining the extracted features into a comprehensive feature vector, and saving the final feature data to a database or file.

5. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1 is characterized in that: The modeling and training also include making decisions through a tree structure, performing classification or regression by finding the best separating hyperplane, performing complex nonlinear mapping by simulating a biological nervous system, using training set data to train the model, iteratively optimizing model parameters, adjusting model hyperparameters on a validation set, using grid search and random search to find the best hyperparameter combination, using accuracy, precision, recall, and F1 score indicators to evaluate the classification performance of the model, using mean square error, root mean square error, and determination coefficient indicators to evaluate the regression performance of the model, using a cross-validation method to further verify the generalization ability of the model, evaluating the final performance of the model on a test set, and adjusting the model structure or algorithm based on the evaluation results.

6. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1 is characterized in that: The anomaly detection also includes setting up a data inflow mechanism to transmit real-time data from a source to a detection system, processing noise and errors in the data in real time, performing feature calculations on real-time data, using a trained model to predict or score real-time features, comparing the anomaly score output by the model with a preset threshold, marking the data point as an anomaly when the score or prediction result output by the model exceeds the threshold, triggering a corresponding alarm according to the level of the anomaly, setting up a multi-level alarm system, sending different levels of notifications according to the severity of the anomaly, and configuring the system to automatically execute preset response operations when an anomaly is detected, including shutting down equipment, switching to a backup system, and adjusting operating parameters.

7. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1 is characterized in that: The abnormality diagnosis also includes confirming whether the detected abnormality is real and valid, eliminating false alarms, recording basic information of the abnormality, including the time when the abnormality occurred, the type of abnormality, relevant data points and model output, collecting all relevant data before and after the abnormality occurs, including system logs, sensor data, and operation records, integrating the collected relevant data, comparing the characteristics of the abnormal data with the characteristics of normal data, analyzing the trends of the abnormal data and related characteristics, comparing the abnormal situation with historical data and known failure modes, verifying the initially identified cause, and confirming the cause through simulation, testing or adjustment of system parameters.

8. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1 is characterized in that: The alarm and response also includes the system continuously monitoring key indicators and data streams through a built-in anomaly detection mechanism or model, detecting data anomalies or exceeding preset thresholds, and triggering anomaly detection signals. The anomaly detection signals are further analyzed to determine whether it is necessary to generate an alarm signal, including comparison with historical data and identification of anomaly types. The system generates an alarm signal, which includes the type, time, and location of the anomaly. The alarm signal is immediately sent to the operator through a preset notification channel, which includes sound and light alarms, text messages, emails, and mobile application push. After receiving the alarm signal, the operator checks to confirm the authenticity and severity of the anomaly, and selects response measures based on the type and urgency of the anomaly. The response measures include shutting down for maintenance, adjusting operating parameters, and starting backup equipment.

9. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1, characterized in that: The system feedback and optimization also includes real-time monitoring of system performance, including the accuracy, response speed and processing effect of anomaly detection, recording detailed information of abnormal events, including the frequency, type, processing measures and results of occurrence, collecting feedback from operators and system users, including satisfaction with the accuracy of system alarms, timeliness of notifications and processing suggestions, diagnosing problems in the system based on the collected feedback information, adjusting and optimizing the anomaly detection algorithm based on the feedback analysis results, retraining the model, using the latest data and feedback information to optimize the model, analyzing abnormal data and feedback data, identifying abnormal patterns and system performance bottlenecks, and evaluating the performance of the current model, including detection accuracy, recall rate, and F1 score indicators, using confusion matrix to analyze false positives and negatives, and identifying the performance of the model under different working conditions.

10. The method for identifying abnormalities in multi-modal data of main transformer operation according to claim 1, characterized in that: The result evaluation and model verification also include retrospective analysis based on real-time operation data and historical data to verify the robustness and adaptability of the model, and based on the evaluation results, targeted adjustments to feature selection and model parameters to optimize the performance of the algorithm.

Images