Intelligent insulation diagnosis device

Abstract

An intelligent insulation diagnosis device disclosed by the present invention comprises a data acquisition module, an edge calculation module, an AI analysis module, a remote monitoring and alarm module and an intelligent optimization strategy module, the edge calculation module is connected with the data acquisition module and the AI analysis module, and the AI analysis module is connected with the remote monitoring and alarm module and the intelligent optimization strategy module. The intelligent optimization strategy module is connected with the remote monitoring and alarm module. By combining a high-precision sensor, edge calculation and AI intelligent analysis, real-time online monitoring and intelligent analysis can be carried out on the cable, so that the partial discharge condition can be found in time, sudden insulation faults can be found in time, AI prediction diagnosis can be carried out according to monitored data, potential fault points can be warned in advance, and the reliability of the cable is improved. Therefore, personnel can conveniently perform protection treatment in time, and the use requirement is met.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent insulation diagnosis, and particularly to an intelligent insulation diagnosis device. Background Art

[0002] An intelligent insulation diagnosis device is an intelligent detection device used for power systems, electrical machinery equipment, and high-voltage electrical appliances, mainly used for monitoring and evaluating the insulation status to predict and prevent faults caused by insulation aging, moisture ingress, partial discharge, etc. in electrical equipment;

[0003] The existing intelligent insulation diagnosis devices have the following deficiencies during use: 1. Most of them collect data periodically during diagnosis rather than real-time online monitoring. Especially for long cables, it is difficult to detect sudden insulation faults in a timely manner; 2. They lack the function of warning potential fault points based on the diagnosis results, which is inconvenient for personnel to make protective treatments in a timely manner. Considering the above situation, the present application proposes an intelligent insulation diagnosis device. Summary of the Invention

[0004] Based on the technical problems existing in the background art, the present invention proposes an intelligent insulation diagnosis device.

[0005] The intelligent insulation diagnosis device proposed by the present invention includes a data acquisition module, an edge computing module, an AI analysis module, a remote monitoring and warning module, and an intelligent optimization strategy module. The edge computing module is connected to the data acquisition module and the AI analysis module, the AI analysis module is connected to the remote monitoring and warning module and the intelligent optimization strategy module, and the intelligent optimization strategy module is connected to the remote monitoring and warning module;

[0006] The data acquisition module includes a distributed optical fiber sensor, a UHF partial discharge detector, an ultrasonic sensor, and a current / voltage / temperature sensor;

[0007] The edge computing module includes a high-frequency signal analysis module and an anomaly detection module;

[0008] The AI analysis module includes a big data storage management module, an AI diagnosis module, and a prediction module.

[0009] Preferably, the distributed optical fiber sensor includes a DTS sensor and a DFOS sensor. The DTS sensor is used to measure the temperature distribution along the cable to find abnormal heating points, and the DFOS sensor is used to monitor the change in optical fiber scattering caused by mechanical stress to predict the risk of insulation layer damage;

[0010] The UHF partial discharge detector uses ultra-high frequency (300 MHz - 3 GHz) electromagnetic wave detection to identify cable insulation defects;

[0011] The ultrasonic sensor is used to detect partial discharge inside the cable sheath and is applicable to oil-immersed cables and high-voltage cross-linked cables;

[0012] The current / voltage / temperature sensor is used to monitor external current, voltage, and temperature.

[0013] Preferably, the high-frequency signal analysis module uses short-time Fourier transform (STFT) to extract partial discharge (PD) characteristics, and its specific logical steps are as follows:

[0014] S101: Use discrete wavelet transform (DWT) to filter out power frequency interference (50Hz / 60Hz) and environmental noise in the data collected by the data acquisition module;

[0015] S102: Use parallel computing of DSP on the FPGA side to perform real-time short-time Fourier transform to obtain the time-frequency characteristics of the PD signal, and its calculation formula is:

[0016] Where X(m,k) is the STFT result of the m-th time window and the k-th frequency component, x(n) is the n-th discrete point of the sampled signal, w(n-m) is the window function, N is the number of FFT calculation points, k is the frequency index, corresponding to the frequency f s is the sampling frequency, e -j2πkn / N is the discrete Fourier transform (DFT) basis function;

[0017] S103: Calculate the partial discharge power spectral density (PSD), the number of discharges (NPD), the pulse amplitude, and the characteristic frequency distribution;

[0018] Their calculation formulas are as follows: Where P psd (k) is the power spectral density at the frequency index K, X(m,k) is the time-frequency signal calculated by STFT, M is the number of time windows of STFT, and T is the total duration of the signal;

[0019] Where NPD is the number of discharges, x(n) is the sampled partial discharge signal, V th is the set discharge threshold, 1(·) is the indicator function, which takes 1 when ∣x(n)∣>V th and 0 otherwise;

[0020] APD = max|x(n)|, APD is the maximum amplitude of the partial discharge pulse, and x(n) is the data point of the partial discharge signal;

[0021] Where FPD is the characteristic frequency of the partial discharge signal, f k is the frequency corresponding to the frequency index k, P psd(k) is the power spectral density at frequency K;

[0022] S104: Extract the PD key features from the calculation result of S103. The key features include time domain features, frequency domain features, time-frequency features, and statistical features.

[0023] Preferably, the anomaly detection module performs partial discharge pattern recognition through edge AI, screens abnormal data, and only uploads abnormal events;

[0024] The specific logical steps are as follows:

[0025] S201: Deploy the lightweight neural network on the edge device. Input the STFT spectrogram, power spectral features (PSD), and time domain parameters (NPD, amplitude) into the lightweight neural network, and then normal data and abnormal discharge data can be directly screened out. The abnormal discharge data includes corona discharge, surface discharge, internal discharge, and floating discharge;

[0026] Among them, when screening abnormal data, the screening logic used is:

[0027] ①. Calculate the current window features,

[0028] F current ={NPD,A max ,f peak ,P PSD , STFT};

[0029] ②. Calculate the historical average features,

[0030] ③. Calculate the feature change rate, D = ‖F current -F avg ||2. If D > the threshold, it is judged as abnormal partial discharge;

[0031] S202: Discard the normal data locally, and the abnormal data will trigger upload to the AI analysis module.

[0032] Preferably, the big data storage management module uses a time series database to store and manage the abnormal data uploaded by the edge computing module;

[0033] The AI diagnosis module is used to identify the PD type. The specific logical steps are as follows:

[0034] S301: Select the convolutional neural network (CNN) as the training model. Divide the abnormal data uploaded by the anomaly detection module into a training set, a validation set, and a test set, and input the divided data into the convolutional neural network (CNN) for training in turn;

[0035] S302: Evaluate the trained convolutional neural network (CNN) in S301, and evaluate its accuracy, precision, and recall rate;

[0036] S303: Deploy the evaluated convolutional neural network model to the edge AI device;

[0037] S304: Input the abnormally detected data in real time into the convolutional neural network model, and PD data can be directly output.

[0038] Preferably, the prediction module combines historical data to predict the insulation aging trend of the cable, and warns of potential fault points 3 to 6 months in advance. The specific logical steps are as follows:

[0039] S401: Collect partial discharge (PD) data, ambient temperature, and historical fault data;

[0040] S402: Construct features through the data collected in S401, calculate the time series features and trend change rates, and establish a complete insulation aging feature library. The calculation formula is: X t ={NPD t , PSD t , PEAKPD t , f peakt , T t}, where X t is the feature vector at time t, including multiple monitoring parameters, NPD t is the number of partial discharge (PD) pulses at time t, PSD t is the PD spectrum energy distribution at time t, PEAKPD t is the maximum PD pulse amplitude at time t, f peakt is the main frequency of PD at time t, and T t is the temperature of the cable environment at time t;

[0041] R t =X t -X t-1 , where R t represents the change of PD-related features from t-1 to t, X t is the feature vector at time t, and X t-1 is the feature vector at time t-1;

[0042] S403: Select the long short-term memory network model as the training model, and train the model with the data collected in S401;

[0043] S404: Construct a warning strategy in the trained model to issue an insulation aging warning 3 to 6 months in advance. The warning strategy is: If R t+T > 20% is short-term severe aging and a red warning needs to be issued for high risk. If 10% ≤ R t+T ≤ 20%, a yellow warning is issued for medium risk. If R t+T < 10%, a green signal is issued for safety;

[0044] S405: Deploy the trained long short-term memory network model to the edge AI device. By inputting real-time data into the long short-term memory network model, the prediction result is directly output.

[0045] Preferably, the remote monitoring and warning module is used to notify the maintenance personnel according to the abnormal level, and the notification method is one of SMS, email, and APP.

[0046] Preferably, the intelligent optimization strategy module is used to generate optimization strategies according to the abnormal data. When the predicted partial discharge anomaly increases, a strategy to reduce the load and cable stress is generated. When the temperature is too high, a strategy to optimize the cooling system and prevent accelerated insulation aging is generated, and insulation repair in the severely partial discharge area is generated to avoid greater damage;

[0047] The triggering conditions for generating the optimization strategy include: ①, a sharp increase in partial discharge (R' NPD > 20%); ②, T t > 80°C, and the relative temperature change rate R' T > 15%; ③, in the severely partial discharge area, the PSD signal exceeds the threshold.

[0048] Compared with the existing technology, the beneficial effects of the present invention are:

[0049] By combining high-precision sensors, edge computing, and AI intelligent analysis, the present invention can perform real-time online monitoring and intelligent analysis on the cable, so as to timely detect partial discharge conditions, timely detect sudden insulation faults, and perform AI prediction diagnosis based on the monitored data to early warn potential fault points, thus facilitating personnel to make protection treatments in a timely manner and meeting the usage requirements. Brief Description of the Drawings

[0050] Figure 1 It is the system block diagram of the intelligent insulation diagnosis device proposed by the present invention;

[0051] Figure 2 It is the flow chart of the intelligent insulation diagnosis device proposed by the present invention. Detailed Embodiments

[0052] The present invention will be further explained below with specific embodiments.

[0053] Embodiment

[0054] Reference Figure 1-2 In this embodiment, an intelligent insulation diagnosis device is proposed, which includes a data acquisition module, an edge computing module, an AI analysis module, a remote monitoring and warning module, and an intelligent optimization strategy module. The edge computing module is connected to the data acquisition module and the AI analysis module, the AI analysis module is connected to the remote monitoring and warning module and the intelligent optimization strategy module, and the intelligent optimization strategy module is connected to the remote monitoring and warning module. Among them, the remote monitoring and warning module is used to notify maintenance personnel according to the abnormal level, and the notification method is one of SMS, email, and APP;

[0055] The intelligent optimization strategy module is used to generate optimization strategies based on abnormal data. When it is predicted that the local discharge anomaly increases, a strategy of reducing the load and reducing the cable stress is generated. When the temperature is too high, a strategy of optimizing the cooling system to prevent the insulation from accelerating aging is generated, and insulation repair in the severely local discharge area is generated to avoid greater damage;

[0056] The triggering conditions for generating optimization strategies include ①, a sharp increase in local discharge (R′ NPD > 20%); ②, T t > 80 °C, and the relative temperature change rate R′ T > 15%; ③, in the severely local discharge area, the PSD signal exceeds the threshold;

[0057] The data acquisition module includes a distributed optical fiber sensor, a UHF partial discharge detector, an ultrasonic sensor, and a current / voltage / temperature sensor. Among them, the distributed optical fiber sensor includes a DTS sensor and a DFOS sensor. The DTS sensor is used to measure the temperature distribution along the cable to find abnormal heating points, and the DFOS sensor is used to monitor the change in optical fiber scattering caused by mechanical stress to predict the risk of insulation layer damage;

[0058] The UHF partial discharge detector uses ultra-high frequency (300 MHz - 3 GHz) electromagnetic waves to detect and identify cable insulation defects;

[0059] The ultrasonic sensor is used to detect partial discharge inside the cable sheath and is applicable to oil-immersed cables and high-voltage cross-linked cables;

[0060] The current / voltage / temperature sensor is used to monitor the external current, voltage, and temperature;

[0061] The edge computing module includes a high-frequency signal analysis module and an anomaly detection module;

[0062] Among them, the high-frequency signal analysis module uses short-time Fourier transform (STFT) to extract partial discharge (PD) characteristics, and its specific logical steps are as follows:

[0063] S101: Use discrete wavelet transform (DWT) to filter out power frequency interference (50Hz / 60Hz) and environmental noise in the data collected by the data acquisition module;

[0064] S102: Use parallel computing of DSP on the FPGA side to perform real-time short-time Fourier transform to obtain the time-frequency characteristics of PD signals. The calculation formula is as follows:

[0065] Where X(m,k) is the STFT result of the m-th time window and the k-th frequency component, x(n) is the n-th discrete point of the sampled signal, w(n - m) is the window function, N is the number of FFT calculation points, k is the frequency index, corresponding to the frequency f s is the sampling frequency, e -j2πkn / N is the discrete Fourier transform (DFT) basis function;

[0066] S103: Calculate the partial discharge power spectral density (PSD), the number of discharges (NPD), the pulse amplitude, and the characteristic frequency distribution;

[0067] The calculation formulas are as follows: Where P psd (k) is the power spectral density at the frequency index K, X(m,k) is the time-frequency signal calculated by STFT, M is the number of time windows of STFT, and T is the total duration of the signal;

[0068] Where NPD is the number of discharges, x(n) is the sampled partial discharge signal, V th is the set discharge threshold, 1(·) is the indicator function, which takes 1 when |x(n)| > V th and takes 0 otherwise;

[0069] APD = max|x(n)|, APD is the maximum amplitude of the partial discharge pulse, and x(n) is the data point of the partial discharge signal;

[0070] Where FPD is the characteristic frequency of the partial discharge signal, f k is the frequency corresponding to the frequency index k, and P psd (k) is the power spectral density at frequency K;

[0071] S104: Extract the key PD features from the calculation results of S103. The key features include time domain features, frequency domain features, time-frequency features, and statistical features;

[0072] The anomaly detection module performs partial discharge pattern recognition through edge AI, screens out abnormal data, and only uploads abnormal events;

[0073] The specific logical steps are as follows:

[0074] S201: Deploy the lightweight neural network on the edge device. Input the STFT spectrogram, power spectral density (PSD) features, and time-domain parameters (NPD, amplitude) into the lightweight neural network, and then normal data and abnormal discharge data will be directly screened out. The abnormal discharge data includes corona discharge, surface discharge, internal discharge, and floating discharge;

[0075] When screening abnormal data, the screening logic used is as follows:

[0076] ①. Calculate the current window features,

[0077] F current ={NPD, A max , f peak , P PSD , STFT};

[0078] ②. Calculate the historical average features,

[0079] ③. Calculate the feature change rate, D = ‖F current -F avg ||2. If D > the threshold, it is judged as abnormal partial discharge;

[0080] S202: Discard the normal data locally, and trigger the upload of abnormal data to the AI analysis module;

[0081] The AI analysis module includes a big data storage management module, an AI diagnosis module, and a prediction module. Among them, the big data storage management module uses a time-series database to store and manage the abnormal data uploaded by the edge computing module;

[0082] The AI diagnosis module is used to identify the PD type. The specific logical steps are as follows:

[0083] S301: Select the convolutional neural network (CNN) as the training model. Divide the abnormal data uploaded by the abnormal detection module into a training set, a validation set, and a test set, and input the divided data into the convolutional neural network (CNN) for training in sequence;

[0084] S302: Evaluate the trained convolutional neural network (CNN) in S301, and evaluate its accuracy, precision, and recall rate;

[0085] S303: Deploy the evaluated convolutional neural network model to the edge AI device;

[0086] S304: Input the abnormally detected data in real time into the convolutional neural network model, and the PD data can be directly output;

[0087] The prediction module combines historical data to predict the insulation aging trend of the cable and warns of potential fault points 3 to 6 months in advance. The specific logical steps are as follows:

[0088] S401: Collect partial discharge (PD) data, ambient temperature, and historical fault data;

[0089] S402: Construct features based on the data collected in S401, calculate time series features and trend change rates, and establish a complete insulation aging feature library. The calculation formula is: X t ={NPD t , PSD t , PEAKPD t , f peakt , T t}, where X t is the feature vector at time t, including multiple monitoring parameters. NPD t is the number of partial discharge (PD) pulses at time t. PSD t is the PD spectrum energy distribution at time t. PEAKPD t is the maximum PD pulse amplitude at time t. f peakt is the main frequency of PD at time t. T t is the temperature of the cable environment at time t;

[0090] R t =X t -X t-1 , where R t represents the change in PD-related features from time t - 1 to time t. X t is the feature vector at time t, and X t-1 is the feature vector at time t - 1;

[0091] S403: Select the long short-term memory network model as the training model and train the model with the data collected in S401;

[0092] S404: Construct a warning strategy in the trained model to issue an insulation aging warning 3 to 6 months in advance. The warning strategy is: If R t+T > 20% is short-term severe aging, and a red warning for high risk needs to be issued. If 10% ≤ R t+T ≤ 20%, a yellow warning for medium risk is issued. If R t+T < 10%, a green signal for safety is issued;

[0093] S405: Deploy the trained long short-term memory network model to the edge AI device. By inputting real-time data into the long short-term memory network model, the prediction result is directly output.

[0094] In this embodiment, a distributed optical fiber sensor, a UHF partial discharge detector, an ultrasonic sensor, and a current / voltage / temperature sensor are pre-deployed to the monitored area location. The deployed sensors monitor the temperature distribution along the cable, the change in optical fiber scattering caused by mechanical stress, cable insulation defects, partial discharge inside the cable sheath, and external current, voltage, and temperature values. The monitored data is transmitted to the high-frequency signal analysis module through 5G network communication. The high-frequency signal analysis module extracts the partial discharge (PD) characteristics of the transmitted data using the short-time Fourier transform (STFT). The extracted data is transmitted to the anomaly detection module. The anomaly detection module performs partial discharge pattern recognition through edge AI, screens out abnormal data, discards normal data, and only uploads abnormal events to the AI analysis module. The big data storage and management module stores and manages the uploaded abnormal data using a time-series database. At the same time, the AI diagnosis module uses a convolutional neural network (CNN) to identify the PD type, thereby determining the insulation fault area. By means of real-time monitoring data and timely processing, analysis, and diagnosis, the partial discharge situation can be detected in time, and thus the sudden insulation fault can be detected in time;

[0095] At the same time, the prediction module combines the historical data and real-time data in the big data storage and management module to predict the insulation aging trend of the cable, and warns of potential fault points 3 to 6 months in advance. At the same time, the intelligent optimization strategy module generates an optimization strategy according to the abnormal data and the triggering conditions of the optimization strategy. The remote monitoring and warning module notifies the maintenance personnel to perform maintenance processing in combination with the optimization strategy through text messages, emails, or the APP according to the abnormal level, so as to achieve the purpose of warning of potential fault points in advance and facilitating personnel to make protection measures in time.

[0096] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention shall cover within the protection scope of the present invention any equivalent replacement or change made according to the technical solution and inventive concept of the present invention.

Claims

1. Intelligent insulation diagnosis device, characterized in that, It includes a data acquisition module, an edge computing module, an AI analysis module, a remote monitoring and warning module, and an intelligent optimization strategy module. The edge computing module is connected to the data acquisition module and the AI analysis module. The AI analysis module is connected to the remote monitoring and warning module and the intelligent optimization strategy module. The intelligent optimization strategy module is connected to the remote monitoring and warning module; The data acquisition module includes a distributed optical fiber sensor, a UHF partial discharge detector, an ultrasonic sensor, and a current / voltage / temperature sensor; The edge computing module includes a high-frequency signal analysis module and an anomaly detection module; The AI analysis module includes a big data storage management module, an AI diagnosis module, and a prediction module.

2. The intelligent insulation diagnosis device according to claim 1, characterized in that The distributed optical fiber sensor includes a DTS sensor and a DFOS sensor. The DTS sensor is used to measure the temperature distribution along the cable and find abnormal heating points. The DFOS sensor is used to monitor the change in optical fiber scattering caused by mechanical stress and predict the risk of insulation layer damage; The UHF partial discharge detector uses ultra-high frequency electromagnetic waves to detect and identify cable insulation defects; The ultrasonic sensor is used to detect partial discharges inside the cable sheath and is applicable to oil-immersed cables and high-voltage cross-linked cables; The current / voltage / temperature sensor is used to monitor the external current, voltage, and temperature.

3. The intelligent insulation diagnosis device according to claim 1, characterized in that The high-frequency signal analysis module uses the short-time Fourier transform (STFT) to extract partial discharge (PD) characteristics. The specific logical steps are as follows: S101: Use the discrete wavelet transform (DWT) to filter out the power frequency interference and environmental noise in the data collected by the data acquisition module; S102: Perform real-time short-time Fourier transform using parallel computing of DSP on the FPGA side to obtain the time-frequency characteristics of the PD signal, and its calculation formula is: Where X(m,k) is the STFT result of the m-th time window and the k-th frequency component, x(n) is the n-th discrete point of the sampled signal, w(n-m) is the window function, N is the number of points for FFT calculation, k is the frequency index, corresponding to the frequency f s is the sampling frequency, e -j2πkn / N is the discrete Fourier transform (DFT) basis function; S103: Calculate the partial discharge power spectral density (PSD), the number of discharges (NPD), the pulse amplitude, and the characteristic frequency distribution; The calculation formulas are as follows respectively: where P psd (k) is the power spectral density at the frequency index K, X(m, k) is the time-frequency signal calculated by STFT, M is the number of time windows of STFT, and T is the total duration of the signal; where NPD is the number of discharges, x(n) is the sampled partial discharge signal, V th is the set discharge threshold, 1(·) is the indicator function, which takes 1 when |x(n)| > V th and 0 otherwise; APD = max|x(n)|, where APD is the maximum amplitude of the partial discharge pulse and x(n) is the partial discharge signal data point; where FPD is the characteristic frequency of the partial discharge signal, f k is the frequency corresponding to the frequency index k, P psd (k) is the power spectral density at frequency K; S104: Extract the key PD characteristics from the calculation results of S103. The key characteristics include time domain characteristics, frequency domain characteristics, time-frequency characteristics, and statistical characteristics.

4. The intelligent insulation diagnosis device according to claim 1, characterized in that The anomaly detection module uses edge AI to identify partial discharge patterns, screen out abnormal data, and only upload abnormal events; The specific logical steps are as follows: S201: Deploy a lightweight neural network on the edge device. Input the STFT spectrogram, power spectral characteristics (PSD), and time domain parameters (NPD, amplitude) into the lightweight neural network, and normal data and abnormal discharge data will be directly screened out. The abnormal discharge data includes corona discharge, surface discharge, internal discharge, and floating discharge; Among them, when screening abnormal data, the screening logic used is: ①. Calculate the current window characteristics, F current = {NPD, A max , f peak , P PsD , STFT}; ②. Calculate the historical average features, ③. Calculate the feature change rate, D = ||F current - F avg ||2. If D > the threshold value, it is judged as an abnormal partial discharge; S202: Discard the normal data locally, and the abnormal data will be triggered to be uploaded to the AI analysis module.

5. The intelligent insulation diagnosis device according to claim 1, characterized in that, The big data storage management module uses a time series database to store and manage the abnormal data uploaded by the edge computing module; The AI diagnosis module is used to identify the PD type. The specific logical steps are as follows: S301: Select the Convolutional Neural Network (CNN) as the training model. Divide the abnormal data uploaded by the anomaly detection module into a training set, a validation set, and a test set, and sequentially input the divided data into the Convolutional Neural Network (CNN) for training; S302: Evaluate the trained Convolutional Neural Network (CNN) in S301, and evaluate its accuracy, precision, and recall rate; S303: Deploy the evaluated Convolutional Neural Network model to the edge AI device; S304: Input the real-time detected abnormal data into the Convolutional Neural Network model, and the PD data can be directly output.

6. The intelligent insulation diagnosis device according to claim 1, characterized in that, The prediction module combines historical data to predict the insulation aging trend of the cable, and warns of potential fault points 3 - 6 months in advance. The specific logical steps are as follows: S401: Collect partial discharge (PD) data, ambient temperature, and historical fault data; S402: Construct features based on the data collected in S401, calculate the time-series features and the trend change rate, and establish a complete insulation aging feature library. The calculation formula is: X t = {NPD t , PSD t , PEAKPD t , f peakt , T t t, where X t is the feature vector at time t, including multiple monitoring parameters. NPD t is the number of partial discharge (PD) pulses at time t. PSD t is the PD spectrum energy distribution at time t. PEAKPD t is the maximum PD pulse amplitude at time t. f peakt is the main frequency of PD at time t. T t is the temperature of the cable environment at time t; R t = X t - X t-1 where R t represents the change in PD-related features from time t - 1 to time t, X y is the feature vector at time t, and X y-1 is the feature vector at time t - 1; S403: Select the Long Short-Term Memory network model as the training model, and train the model in combination with the data collected in S401; S404: Build an early warning strategy in the trained model to issue an insulation aging warning 3 - 6 months in advance. The early warning strategy is as follows: If R t+T > 20% is short-term severe aging, and a red warning needs to be issued for high risk. If 10% ≤ R t+T ≤ 20%, a yellow warning is issued for medium risk. If R t+T < 10%, a green signal is issued for safety; S405: Deploy the trained Long Short-Term Memory network model to the edge AI device. By inputting real-time data into the Long Short-Term Memory network model, the prediction result can be directly output.

7. The intelligent insulation diagnosis device according to claim 1, wherein The remote monitoring and alarm module is used to notify the maintenance personnel according to the anomaly level, and the notification method is one of SMS, email, and APP.

8. The intelligent insulation diagnosis device according to claim 1, characterized in that, The intelligent optimization strategy module is used to generate optimization strategies based on abnormal data. When it is predicted that the partial discharge anomaly increases, generate strategies to reduce the load and cable stress. When the temperature is too high, generate strategies to optimize the cooling system to prevent accelerated insulation aging, and generate insulation repair for severely partial discharge areas to avoid greater damage; The triggering conditions for generating optimization strategies include: ①, a sharp increase in partial discharge (R′ NPD > 20%); ②, T t > 80 °C, and the relative temperature change rate R′ T > 15%; ③, in the area with severe partial discharge, the PSD signal exceeds the threshold value.

Images