Voltage transformer online thermal stress monitoring system based on LSTM neural network

Abstract

The invention relates to the technical field of power system equipment monitoring, in particular to a voltage transformer online thermal stress monitoring system based on an LSTM neural network, which comprises a control unit, a state display unit, a thermal stress monitoring unit, a voice alarm unit and a data storage unit, wherein the control unit is respectively connected with the state display unit, the thermal stress monitoring unit, the voice alarm unit and the data storage unit and is used for receiving and processing data information transmitted by the units and sending command data information to the units to control the units to complete corresponding functions. The units of the constructed monitoring module perform real-time monitoring cooperatively, the overall operation is simple and convenient, user experience is emphasized, the working efficiency is improved, the LSTM network is good at processing sequence data and capturing a thermal stress related time dependency relationship, dependence on a traditional method is reduced, the prediction efficiency is improved, and the parallel computing capability of the LSTM network can quickly process data, calculate thermal stress and perform early warning. And equipment faults caused by overlarge thermal stress are prevented.

Description

Technical Field

[0001] The present invention relates to the technical field of power system equipment monitoring, and in particular to an online thermal stress monitoring system for a voltage transformer based on an LSTM neural network. Background Art

[0002] In the power system, voltage transformers are electrical equipment that connect the primary and secondary systems of the power grid and are installed at various nodes of the power grid. Voltage transformers convert high voltage into low voltage and provide the voltage information of the primary system to the secondary system. This not only ensures the safety of operation, but also provides necessary voltage signals for each link of the power system for operation monitoring. They are an indispensable part of the power system.

[0003] In recent years, the number of electrical equipment in the power distribution network has continued to increase. As the voltage transformer's operating life and operating pressure increase, the equipment itself will have various internal and external faults. In severe cases, the fault may even cause the voltage transformer to burn out or explode. Various faults have seriously affected the accuracy of electrical measurements in the power system, the safety and reliability of the power system, and the personal safety of personnel. Therefore, there is an urgent need for a detection system that can detect voltage transformer faults in real time, so as to promptly detect and eliminate the faults of the transformer to ensure the safe and stable operation of the system.

[0004] When a transformer fails, it is often accompanied by changes in temperature and thermal stress. This patent will establish a thermal stress monitoring system based on the above two points. Traditional methods for monitoring thermal stress of voltage transformers include infrared thermal imaging technology, equivalent resistance detection method, resistance measurement method, and temperature monitoring based on the rate of change of collector current at IGBT shutdown. Most of these methods focus on non-contact monitoring, and there are certain limitations in measurement accuracy and cost. For example, infrared thermal imaging technology belongs to offline monitoring and cannot continuously reflect the working status of the voltage transformer. In addition, the cost of infrared thermal imagers is relatively high and requires personnel to operate on site.

[0005] To solve the above problems, an innovative method for monitoring the thermal stress and temperature of the voltage transformer is proposed. The thermal stress and temperature of the voltage transformer are monitored in real time. The fault type is reflected by the monitoring data, and the fault is sent to the staff in time to ensure the operation reliability of the equipment. Summary of the invention

[0006] The purpose of the present invention is to overcome the above problems or at least partially solve the above problems, and to propose an online thermal stress monitoring system for voltage transformer based on LSTM neural network.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an online thermal stress monitoring system for a voltage transformer, wherein the monitoring system can obtain real-time data according to a data communication interface of a DAQ data acquisition card, perform iterative calculations on the data required by the system read from the real-time data, save the results of the analysis and calculation in a server, and display them through a graphical interface, and can retrieve historical data for analysis and research at any time, and according to the operation requirements, the thermal stress monitoring system of the voltage transformer is started by operating the thermal stress button of the monitoring system, the thermal stress condition of the monitoring unit is retrieved and relevant instructions are issued;

[0008] The monitoring system includes a control unit, a status display unit, a thermal stress monitoring unit, a voice alarm unit and a data storage unit, wherein the control unit is respectively connected to the status display unit, the thermal stress monitoring unit, the voice alarm unit and the data storage unit, and is used to receive and process data information transmitted by the above-mentioned units, and send command data information to each unit to control each unit to complete corresponding functions.

[0009] In a preferred embodiment, the status display unit is connected to the control unit and has the function of real-time monitoring of the operating status and actual parameters of the equipment. It can display the thermal stress and temperature conditions of the voltage transformer and immediately issue an alarm to notify relevant operators once an abnormality or potential fault is detected.

[0010] In a preferred embodiment, the thermal stress monitoring unit is connected to the control unit and the status display unit, and monitors the temperature changes of the voltage transformer during operation and the thermal stress generated by thermal expansion through sensors. When the monitored thermal stress exceeds a preset safety threshold, a fault alarm signal is sent to the system.

[0011] In a preferred embodiment, the voice alarm unit is connected to the control unit and the thermal stress monitoring unit, and is used to receive the voltage transformer fault signal, monitor various signal parameters of thermal stress, and play a real-time voice alarm once a fault signal is received.

[0012] In a preferred embodiment, the data storage unit is connected to the control unit, has the function of collecting and saving various temperature and pressure data obtained from the sensor device, supports the query and analysis function of historical data, can quickly retrieve and analyze historical data in multiple dimensions, and reflect the historical data and related results of the voltage transformer operation in the form of trend charts and reports. It also has an offline analysis function and can record and store key parameters of thermal stress of the voltage transformer for relevant technical personnel to analyze and study.

[0013] A method for predicting thermal stress of a voltage transformer based on an LSTM neural network, based on the above-mentioned online thermal stress monitoring system for a voltage transformer, performs working condition identification to classify different operating states to improve calculation accuracy, and then performs data processing, including cleaning data to remove outliers and noise, and normalizing data of different dimensions; the preprocessed sample data is input into the LSTM neural network, the input layer receives the data, the hidden layer performs feature extraction and learning processing of long-term dependencies, the output layer iteratively predicts and denormalizes the predicted data, and derives the remaining useful life (RUL) of the equipment based on the thermal stress prediction, and predicts the normal operation time of the equipment under the current working condition by analyzing the trend of thermal stress changes over time.

[0014] In a preferred embodiment, the goal of the LSTM neural network is to predict the value of thermal stress by directly establishing a mapping relationship between raw data and thermal stress:

[0015] y RUL =f RUL (x1,x2,...,x n )

[0016] Where: x1, x2, ..., x n Is the monitoring data of the equipment status; RUL is the value of the predicted thermal stress; f RUL is the mapping function of the representation;

[0017] Let the set of devices in the training set be U. For each device, there is u∈U. Then the multiple sets of sequences are Perform Z-score normalization on the sequence to prevent the influence of different value ranges on the prediction results:

[0018]

[0019] Where: x i represents the i-th data point in the sequence X; μ represents the mean of the data set; σ represents the standard deviation of the data set; z represents the data point after Z-score normalization.

[0020] In a preferred embodiment, the LSTM neural network uses the Adam optimization algorithm to adjust the network parameters. The formula of the Adam optimization algorithm is:

[0021]

[0022] In order to update the parameters normally, deviation correction needs to be done according to the following formula:

[0023]

[0024] The formula for updating the neural network parameters by the Adam optimization algorithm is:

[0025]

[0026] Where: m t is the gradient first moment estimate; β1 and β2 are hyperparameters; v t is the second moment estimate; α is the learning rate of the neural network model; ε is a constant to prevent zero division error.

[0027] In a preferred embodiment, the voltage transformer state recognition based on the LSTM neural network is performed according to the following steps:

[0028] Firstly, the historical data of voltage transformer under different working conditions are collected for analysis and processing, and used as the input feature quantity of LSTM network;

[0029] Secondly, the data is normalized to meet the input requirements of the LSTM network. At the same time, the data is divided into a training set and a test set. The LSTM network structure is designed, and the LSTM model is trained using the training set data. The network parameters are adjusted through the back propagation algorithm to minimize the error between the predicted thermal stress and the actual thermal stress.

[0030] Then, the trained model is verified using the test set data to evaluate the prediction accuracy and generalization ability of the model. The trained LSTM model is applied to real-time data to predict the thermal stress of the voltage transformer under different working conditions, providing decision support for the operation and maintenance of the voltage transformer. The prediction results of the LSTM model are analyzed and compared with the actual measured values ​​to evaluate the accuracy of the model, and further model adjustments are made as needed.

[0031] Finally, the trained model is deployed to the production environment to perform state recognition and classification output on the input feature sample to be tested, determine the current operating status of the voltage transformer, and perform real-time monitoring and prediction.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. The present invention performs data acquisition based on a DAQ data acquisition card, which has high acquisition accuracy, good data processing capability, and is flexible and expandable. It is connected to a sensor to collect data such as temperature and pressure of a voltage transformer in real time, and transmits the data to a control system via a communication network to comprehensively monitor and control its operation process;

[0034] 2. The LSTM network of the present invention is good at processing sequence data, capturing the time dependency of thermal stress, reducing reliance on traditional methods, and improving prediction efficiency. Its parallel computing capability can quickly process data, calculate thermal stress, and issue warnings to prevent equipment failures caused by excessive thermal stress.

[0035] 3. The monitoring modules constructed by the present invention cooperate with each other for real-time monitoring. The control unit adjusts the strategy in response to equipment changes. The status display unit updates data in real time. The thermal stress monitoring unit alarms abnormally. The data storage unit ensures data security. The voice alarm unit facilitates staff to receive alarms. The overall operation is simple, user experience is emphasized, and work efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the monitoring system of the present invention;

[0037] Figure 2 It is a module diagram of the monitoring system of the present invention;

[0038] Figure 3 This is a flow chart of thermal stress prediction based on LSTM neural network in the present invention. DETAILED DESCRIPTION

[0039] 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.

[0040] See also Figure 1-3 The present invention provides a technical solution: an online thermal stress monitoring system for a voltage transformer, wherein the monitoring system can obtain real-time data according to a data communication interface of a DAQ data acquisition card, perform iterative calculations on the data required by the system read from the real-time data, save the results of the analysis and calculation in a server, and display them through a graphical interface, and can retrieve historical data for analysis and research at any time, and according to the operation requirements, the thermal stress monitoring system of the voltage transformer is started by operating the thermal stress button of the monitoring system, the thermal stress situation of the monitoring unit is retrieved and relevant instructions are issued; the thermal stress can be monitored in real time through the online system, and the over-limit alarm parameters can be displayed, and the related parameters can be automatically saved, and the relevant personnel can also perform analysis through the saved parameters, thereby ensuring the normal operation of the equipment;

[0041] The monitoring system includes a control unit, a status display unit, a thermal stress monitoring unit, a voice alarm unit and a data storage unit, wherein the control unit is respectively connected to the status display unit, the thermal stress monitoring unit, the voice alarm unit and the data storage unit, and is used to receive and process data information transmitted by the above-mentioned units, and send command data information to each unit to control each unit to complete the corresponding function.

[0042] In specific implementation, the status display unit is connected to the control unit and has the function of real-time monitoring of the operating status and actual parameters of the equipment. It can display the thermal stress and temperature conditions of the voltage transformer and immediately issue an alarm to notify relevant operators once an abnormality or potential fault is detected.

[0043] In specific implementation, the thermal stress monitoring unit is connected to the control unit and the status display unit, and monitors the temperature changes of the voltage transformer during operation and the thermal stress generated by thermal expansion through sensors. When the monitored thermal stress exceeds the preset safety threshold, a fault alarm signal is sent to the system.

[0044] During specific implementation, the voice alarm unit is connected to the control unit and the thermal stress monitoring unit to receive the voltage transformer fault signal and monitor various signal parameters of thermal stress. Once the fault signal is received, real-time voice playback is performed to give an alarm.

[0045] During specific implementation, the data storage unit is connected to the control unit, and has the function of collecting and saving various temperature and pressure data obtained from the sensor equipment, supports the query and analysis of historical data, can quickly retrieve and analyze historical data in multiple dimensions, and reflect the historical data and related results of the voltage transformer operation in the form of trend charts and reports. It also has offline analysis functions and can record and store key parameters of thermal stress of the voltage transformer for relevant technical personnel to conduct analysis and research.

[0046] Neural networks can be thought of as black boxes that can fit any function. As long as there is enough training data and the network structure is complex enough, it can approximate any continuous function with arbitrary precision. However, traditional neural networks are powerless when faced with sequence data. They can only process data one by one, and the previous input and the next input are completely unrelated. To predict thermal stress, it is necessary to understand the correlation between different time series and find the optimal network model. For example, the temperature data of a voltage transformer is time series data. For this type of problem, recurrent neural networks can handle it with ease. By introducing the concept of implicit states, considering the relationship between the sequence data before and after, and "memorizing" key information, it can effectively mine sequence features.

[0047] When the temperature of an object is uneven, thermal stress will be generated. Although thermal stress cannot be measured directly, it can be obtained through parameters such as temperature and pressure, which are within the range of sensor monitoring. The change of thermal stress is closely related to time, and the LSTM network can effectively process and analyze time series data. Its design allows it to learn long-term dependencies in the data, which helps to predict the changing trend of thermal stress. In addition, the LSTM network is highly adaptable and can adapt to different voltage transformer models and operating conditions, and has good generalization ability. In short, the LSTM neural network can well solve the problem of long-term learning of traditional neural networks. It can remember information for a long time and can choose to forget or remember information in the neuron state.

[0048] Therefore, a voltage transformer thermal stress prediction method based on LSTM neural network is proposed. It is very important to identify the working condition in the initial stage of thermal stress prediction, and different working conditions will have different effects on thermal stress. On the basis of the above-mentioned voltage transformer online thermal stress monitoring system, working condition identification is carried out to classify different operating states to improve calculation accuracy, and then data processing is performed, including cleaning data to remove outliers and noise, and normalizing data of different dimensions. The preprocessed sample data is input into the LSTM neural network, the input layer receives the data, the hidden layer performs feature extraction and learning to process long-term dependencies, and the output layer iteratively predicts and de-normalizes the predicted data. The remaining service life (RUL) of the equipment is derived based on the thermal stress prediction, and the normal operation time of the equipment under the current working conditions is predicted by analyzing the trend of thermal stress changes over time.

[0049] In general, this prediction flowchart can accurately predict the thermal stress value and further deduce the remaining service life of the equipment through the steps of working condition identification, data preprocessing, LSTM neural network training and prediction.

[0050] In specific implementation, the goal of the LSTM neural network is to predict the value of thermal stress by directly establishing a mapping relationship between the original data and thermal stress:

[0051] y RUL =f RUL (x1,x2,...,x n )

[0052] Where: x1, x2, ..., x n Is the monitoring data of the equipment status; RUL is the value of the predicted thermal stress; f RUL is the mapping function of the representation;

[0053] Let the set of devices in the training set be U. For each device, there is u∈U. Then the multiple sets of sequences are Perform Z-score normalization on the sequence to prevent the influence of different value ranges on the prediction results:

[0054]

[0055] Where: x i represents the i-th data point in the sequence X; μ represents the mean of the data set; σ represents the standard deviation of the data set; z represents the data point after Z-score normalization.

[0056] In specific implementation, the LSTM neural network uses the Adam optimization algorithm to adjust the network parameters. The core of the optimization is to obtain model parameters that can minimize the objective function through iteration. Adam combines the advantages of adaptive gradient (AdaGrad) and RMSProp. The updated step size is obtained by the first moment estimate of the gradient (the average value of the gradient) and the second moment estimate (the variance of the centerless gradient). The Adam optimizer is essentially an implementation method of time back propagation BPTT. The formula of the Adam optimization algorithm is:

[0057]

[0058] In order to update the parameters normally, deviation correction needs to be done according to the following formula:

[0059]

[0060] The formula for updating the neural network parameters by the Adam optimization algorithm is:

[0061]

[0062] Where: m t is the gradient first moment estimate; β1 and β2 are hyperparameters; v t is the second moment estimate; α is the learning rate of the neural network model; ε is a constant to prevent zero division error.

[0063] The finite element input and output data are processed according to the sliding time window method to construct feature samples, which are input into the already built LSTM network model. The model is trained and the model accuracy is tested with a test set, and then the performance of the prediction algorithm is evaluated.

[0064] In specific implementation, the voltage transformer state recognition based on LSTM neural network is carried out in the following four steps:

[0065] (1) The historical data of the voltage transformer under different working conditions are collected and analyzed as the input feature quantity of the LSTM network.

[0066] (2) The data is normalized to meet the input requirements of the LSTM network. At the same time, the data is divided into a training set and a test set. The LSTM network structure is designed, the LSTM model is trained using the training set data, and the network parameters are adjusted through the back propagation algorithm to minimize the error between the predicted thermal stress and the actual thermal stress.

[0067] (3) Use the test set data to verify the trained model and evaluate the prediction accuracy and generalization ability of the model. Apply the trained LSTM model to real-time data to predict the thermal stress of the voltage transformer under different working conditions and provide decision support for the operation and maintenance of the voltage transformer. Analyze the prediction results of the LSTM model and compare them with the actual measurement values ​​to evaluate the accuracy of the model, and make further model adjustments as needed.

[0068] (4) Deploy the trained model to the production environment, identify the state of the input feature samples to be tested, classify the output, determine the current operating status of the voltage transformer, and perform real-time monitoring and prediction.

[0069] The present invention proposes an online thermal stress monitoring system for voltage transformer based on LSTM neural network. The DAQ data acquisition card has high-precision data acquisition capability, which provides high flexibility and scalability for data acquisition. The present invention collects data from each measuring point through the DAQ data acquisition card. The DAQ data acquisition card is connected to the sensor to collect key data such as temperature and pressure during the operation of the voltage transformer in real time. The collected data is transmitted to the control system through the communication network, and the entire process of the voltage transformer operation is comprehensively monitored and controlled.

[0070] LSTM networks are good at processing sequence data and are good at capturing the time dependency of data, which is of great significance for thermal stress calculation. During the operation of voltage transformers, thermal stress accumulates over time, which is affected by factors such as working status, load and ambient temperature changes at different time points. LSTM networks can effectively capture these long-term effects, reduce reliance on traditional finite element analysis methods, and improve prediction efficiency; the parallel computing capability of LSTM networks enables them to quickly process real-time collected data, calculate thermal stress values ​​in a timely manner, and issue warnings based on preset thresholds, so that staff can respond in a timely manner to prevent excessive thermal stress from causing equipment failures;

[0071] A complete voltage transformer online monitoring module has been built, and each unit works together to achieve real-time monitoring. The control unit can quickly respond to changes in equipment status and adjust the monitoring strategy; the status display unit updates data in real time, allowing users to easily grasp the latest status of the equipment; the thermal stress monitoring unit immediately alarms when thermal stress is abnormal, ensuring that staff can handle it in time to avoid equipment damage and accidents; the data storage unit ensures that data is safe and reliable, and historical data will not be lost even in the event of a fault or power outage, providing strong support for fault diagnosis and maintenance; the voice alarm unit allows staff to receive alarm information and respond in time nearby without having to stare at the screen; the status display unit intuitively displays the equipment status without the need for complex operations and data analysis. The entire monitoring module focuses on user experience, is easy to operate, and improves work efficiency.

[0072] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A voltage transformer online thermal stress monitoring system, characterized in that: The monitoring system can obtain real-time data according to the data communication interface of the DAQ data acquisition card, iteratively calculate the data required by the system read from the real-time data, save the results of the analysis and calculation in the server, and display them through a graphical interface. It can retrieve historical data at any time for analysis and research, and according to the operation requirements, the thermal stress monitoring system of the voltage transformer is started by operating the thermal stress button of the monitoring system, the thermal stress situation of the monitoring unit is retrieved and relevant instructions are issued; The monitoring system includes a control unit, a status display unit, a thermal stress monitoring unit, a voice alarm unit and a data storage unit, wherein the control unit is respectively connected to the status display unit, the thermal stress monitoring unit, the voice alarm unit and the data storage unit, and is used to receive and process data information transmitted by the above-mentioned units, and send command data information to each unit to control each unit to complete corresponding functions.

2. The voltage transformer online thermal stress monitoring system according to claim 1, characterized in that: The status display unit is connected to the control unit and has the function of real-time monitoring of the operating status and actual parameters of the equipment. It can display the thermal stress and temperature conditions of the voltage transformer and immediately issue an alarm to notify relevant operators once an abnormality or potential fault is detected.

3. The voltage transformer online thermal stress monitoring system according to claim 2, characterized in that: The thermal stress monitoring unit is connected to the control unit and the status display unit, and monitors the temperature change of the voltage transformer during operation and the thermal stress generated by thermal expansion through sensors. When the monitored thermal stress exceeds a preset safety threshold, a fault alarm signal is sent to the system.

4. The voltage transformer online thermal stress monitoring system according to claim 3, characterized in that: The voice alarm unit is connected with the control unit and the thermal stress monitoring unit, and is used to receive the voltage transformer fault signal, monitor various signal parameters of thermal stress, and play the voice in real time to alarm once the fault signal is received.

5. The voltage transformer online thermal stress monitoring system according to claim 4, characterized in that: The data storage unit is connected to the control unit, and has the function of collecting and saving various temperature and pressure data obtained from the sensor device, supports the query and analysis function of historical data, can quickly retrieve and analyze historical data in multiple dimensions, and reflect the historical data and related results of the voltage transformer operation in the form of trend charts and reports. It also has an offline analysis function and can record and store key parameters of thermal stress of the voltage transformer for analysis and research by relevant technical personnel.

6. A voltage transformer thermal stress prediction method based on LSTM neural network, characterized in that: Based on the voltage transformer online thermal stress monitoring system as described in any one of claims 1 to 5, operating condition identification is performed to classify different operating states to improve calculation accuracy, and then data processing is performed, including cleaning data to remove outliers and noise, and normalizing data of different dimensions; The preprocessed sample data is input into the LSTM neural network. The input layer receives the data, the hidden layer extracts features and learns to process long-term dependencies, and the output layer iteratively predicts and de-normalizes the predicted data. The remaining useful life (RUL) of the equipment is derived based on the thermal stress prediction, and the normal operation time of the equipment under the current working conditions is predicted by analyzing the trend of thermal stress changes over time.

7. The method for predicting thermal stress of a voltage transformer based on an LSTM neural network according to claim 6 is characterized in that: The goal of the LSTM neural network is to predict the value of thermal stress by directly establishing a mapping relationship between raw data and thermal stress: y RUL =f RUL (x1,x2,…,x n ) Where: x1, x2, ..., x n Is the monitoring data of the equipment status; RUL is the value of the predicted thermal stress; f RUL is the mapping function of the representation; Let the set of devices in the training set be U. For each device, there is u∈U. Then the multiple sets of sequences are Perform Z-score normalization on the sequence to prevent the influence of different value ranges on the prediction results: Where: x i represents the i-th data point in the sequence X; μ represents the mean of the data set; σ represents the standard deviation of the data set; z represents the data point after Z-score normalization.

8. The method for predicting thermal stress of a voltage transformer based on an LSTM neural network according to claim 6 is characterized in that: The LSTM neural network uses the Adam optimization algorithm to adjust the network parameters. The formula of the Adam optimization algorithm is: In order to update the parameters normally, deviation correction needs to be done according to the following formula: The formula for updating the neural network parameters by the Adam optimization algorithm is: Where: m t is the gradient first moment estimate; β1 and β2 are hyperparameters; v t is the second moment estimate; α is the learning rate of the neural network model; ε is a constant to prevent zero division error.

9. The method for predicting thermal stress of a voltage transformer based on an LSTM neural network according to claim 6, characterized in that: The voltage transformer status recognition based on LSTM neural network is carried out according to the following steps: Firstly, the historical data of voltage transformer under different working conditions are collected for analysis and processing, and used as the input feature quantity of LSTM network; Secondly, the data is normalized to meet the input requirements of the LSTM network. At the same time, the data is divided into a training set and a test set. The LSTM network structure is designed, and the LSTM model is trained using the training set data. The network parameters are adjusted through the back propagation algorithm to minimize the error between the predicted thermal stress and the actual thermal stress. Then, the trained model is verified using the test set data to evaluate the prediction accuracy and generalization ability of the model. The trained LSTM model is applied to real-time data to predict the thermal stress of the voltage transformer under different working conditions, providing decision support for the operation and maintenance of the voltage transformer. The prediction results of the LSTM model are analyzed and compared with the actual measured values ​​to evaluate the accuracy of the model, and further model adjustments are made as needed. Finally, the trained model is deployed to the production environment to perform state recognition and classification output on the input feature sample to be tested, determine the current operating status of the voltage transformer, and perform real-time monitoring and prediction.

Images