Method and system for monitoring insulation aging of distribution cable based on leakage current

Abstract

The invention discloses a method and system for monitoring insulation aging of a distribution cable based on leakage current, and belongs to the technical field of power system monitoring. According to the method, data are collected through a leakage current sensor, abnormal data are screened after preprocessing, time-frequency domain analysis is carried out through multi-scale wavelet transform, and features are extracted and input into a deep learning model to recognize insulation aging features. And calculating an aging index based on the identification result, determining an aging level and generating a maintenance suggestion. And the result is sent to a management center and stored in a database for subsequent analysis and strategy making. According to the method, real-time monitoring, intelligent diagnosis and predictive maintenance of the insulation state of the power distribution cable are realized, the reliability of a power distribution system is improved, and the maintenance cost is reduced.

Description

Technical Field

[0001] The present invention relates to detection technology, and in particular to a method and system for monitoring insulation aging of distribution cables based on leakage current. Background Art

[0002] Distribution cables are key equipment in power systems, and their insulation performance directly affects the safety and reliability of distribution systems. At present, the insulation aging monitoring of distribution cables mainly adopts the methods of regular detection and offline diagnosis, which has the following defects:

[0003] Discontinuous monitoring method: Traditional periodic detection methods cannot achieve real-time monitoring of the insulation status of distribution cables, making it difficult to detect potential insulation problems in a timely manner, which may lead to sudden failures.

[0004] Insufficient data analysis: Existing technologies usually only focus on a single electrical parameter, such as insulation resistance or dielectric loss factor, and lack a comprehensive analysis of leakage current signals, making it difficult to accurately reflect the complex process of insulation aging.

[0005] Low level of intelligent diagnosis: Existing insulation aging assessment methods mostly rely on empirical rules and simple threshold judgments, lack the application of intelligent data processing and deep learning technology, and are difficult to adapt to the identification of insulation aging characteristics under different operating environments and load conditions.

[0006] These defects limit the accuracy and timeliness of distribution cable insulation aging monitoring, affecting the preventive maintenance and reliable operation of the distribution system. Therefore, there is an urgent need for a distribution cable insulation aging monitoring method that can achieve real-time monitoring, intelligent diagnosis and predictive maintenance. Summary of the invention

[0007] The purpose of the present invention is to solve the problems of discontinuous monitoring mode, insufficient data analysis and low intelligent diagnosis in the prior art.

[0008] In order to solve the above problems, the present invention provides a method for monitoring the insulation aging of a distribution cable based on leakage current, comprising:

[0009] The leakage current data of the distribution cable is collected in real time by a leakage current sensor arranged on the distribution cable; the leakage current data is input into a data preprocessing module, and the data preprocessing module performs filtering, denoising and normalization on the leakage current data to obtain preprocessed leakage current data; the preprocessed leakage current data is compared with the pre-stored historical leakage current data to screen out abnormal leakage current data;

[0010] The abnormal leakage current data is analyzed in the time and frequency domain by using a multi-scale wavelet transform to extract the spectrum characteristics, amplitude characteristics and phase characteristics of the leakage current signal; the extracted characteristics are input into a pre-trained deep learning model, the deep learning model includes a combination structure of a convolutional neural network and a long short-term memory network, and is used to identify the insulation aging characteristics in the leakage current signal; based on the identified insulation aging characteristics, combined with a pre-established insulation aging evaluation standard, the insulation aging index of the distribution cable is calculated;

[0011] The insulation aging index is compared with the preset multi-level warning threshold to determine the aging level of the distribution cable; according to the aging level, a preset decision rule library is called to generate corresponding maintenance suggestions; the maintenance suggestions are sent to the distribution system management center through a remote communication module; at the same time, the insulation aging index, aging level and maintenance suggestions are stored in a historical database for subsequent trend analysis and predictive maintenance strategy formulation; the deep learning model and decision rule library are regularly updated and optimized.

[0012] In the preferred mode,

[0013] The abnormal leakage current data is analyzed in the time and frequency domain by using multi-scale wavelet transform to extract the spectrum characteristics, amplitude characteristics and phase characteristics of the leakage current signal, including:

[0014] The Daubechies wavelet is selected as the basis function, and the input abnormal leakage current signal is decomposed by 5 layers of wavelet to obtain a low-frequency approximation coefficient and five detail coefficients of different scales;

[0015] Performing a fast Fourier transform on the wavelet coefficients of each scale to obtain a corresponding spectrum; calculating the energy distribution of each frequency band and extracting the main frequency components where the energy is concentrated; calculating the spectrum entropy corresponding to each scale coefficient to characterize the complexity of the signal; combining the main frequency components, energy distribution and spectrum entropy to form a spectrum feature vector;

[0016] Calculate the statistical characteristics of the wavelet coefficients of each scale, including mean, variance, skewness and kurtosis; extract the maximum value, minimum value and root mean square value of each scale coefficient; calculate the wavelet energy ratio, that is, the percentage of each scale energy to the total energy; combine the statistical characteristics, maximum value, minimum value, root mean square value and wavelet energy ratio to form an amplitude feature vector;

[0017] The instantaneous phase of the wavelet coefficients of each scale is calculated by using Hilbert transform; the statistical characteristics of the phase are extracted, including the phase mean and the phase variance; the phase synchronization index is calculated to characterize the phase relationship between different scales; the phase difference between different scales is calculated; the phase statistical characteristics, the phase synchronization index and the phase difference are combined to form a phase feature vector;

[0018] The frequency spectrum feature vector, the amplitude feature vector and the phase feature vector are combined to form a comprehensive feature vector, which fully reflects the characteristics of the abnormal leakage current signal in the time domain and the frequency domain.

[0019] In the preferred mode,

[0020] The extracted features are input into a pre-trained deep learning model, which includes a combination of a convolutional neural network and a long short-term memory network, and is used to identify insulation aging features in leakage current signals, including:

[0021] Constructing a combined structure integrating a convolutional neural network and a long short-term memory network, the combined structure comprising an input layer, a convolutional neural network part, a long short-term memory network part and a fully connected layer; wherein the input layer receives a comprehensive feature vector, and the comprehensive feature vector comprises a spectrum feature, an amplitude feature and a phase feature;

[0022] The convolutional neural network part includes a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer and a flattening layer; the first convolution layer uses a plurality of three-by-three convolution kernels to perform convolution operations, the first pooling layer and the second pooling layer both use maximum pooling operations, and the flattening layer converts a two-dimensional feature map into a one-dimensional vector;

[0023] The long short-term memory network part includes a first long short-term memory layer and a second long short-term memory layer, the first long short-term memory layer returns a complete sequence, and the second long short-term memory layer only returns the output of the last time step;

[0024] The fully connected layer includes a first fully connected layer, a random dropout layer, a second fully connected layer and an output layer, and the number of neurons in the output layer is equal to the number of insulation aging levels;

[0025] The historical leakage current data and the corresponding insulation aging level labels are divided into a training set, a validation set and a test set; the training set is subjected to data enhancement processing, wherein the data enhancement processing includes adding Gaussian noise, time shift and amplitude scaling; the combined structure is trained using a cross entropy loss function and an Adam optimizer, and a learning rate scheduling strategy and an early stopping strategy are adopted during the training process; the model performance is evaluated on the test set, and the confusion matrix, precision, recall rate and F1 score are calculated; the model structure and hyperparameters are fine-tuned according to the evaluation results; an integration method or a gradient method is used to analyze the contribution of different features to model prediction; SHapley Additive exPlanations values ​​are used to interpret the prediction results of the model, and the most critical features for insulation aging judgment are identified.

[0026] In the preferred mode,

[0027] The insulation aging index is compared with a preset multi-level warning threshold to determine the aging level of the distribution cable; according to the aging level, a preset decision rule library is called to generate corresponding maintenance suggestions including:

[0028] Obtaining an insulation aging index of a distribution cable; comparing the insulation aging index with a preset multi-level warning threshold to determine an aging level of the distribution cable, wherein the multi-level warning threshold is set by the following steps:

[0029] Collect cable sample data with known aging degree; use deep learning model to calculate insulation aging index of the cable sample data; use clustering algorithm to group the calculated insulation aging index; fine-tune the clustering results based on the experience of power system experts to determine the threshold of each level;

[0030] A decision rule base is constructed, wherein the decision rule base includes aging level, cable importance, cable service life, cable current carrying capacity utilization, environmental factors and historical fault records, wherein the construction process of the decision rule base includes:

[0031] Gather the knowledge and experience of power system experts; organize historical maintenance records and analyze maintenance strategies and their effects under different circumstances; use decision tree algorithms to build preliminary decision rules; optimize and adjust decision rules through expert review and field verification; encode the optimized decision rules into a form that can be processed by computers;

[0032] Generating maintenance suggestions according to the aging level and the decision rule base specifically includes:

[0033] Input the aging level and other relevant parameters of the distribution cable; match the rules that meet the conditions in the decision rule library; when multiple rules are matched, sort them according to the priority of the rules; select the rule with the highest priority and generate corresponding maintenance suggestions.

[0034] In the preferred mode,

[0035] The method further comprises:

[0036] The maintenance suggestion is dynamically updated, including:

[0037] Record each maintenance operation and its effect; regularly analyze maintenance effect data; adjust the corresponding decision rules when it is found that some maintenance suggestions are not effective; use reinforcement learning algorithms to continuously optimize decision rules;

[0038] The aging level judgment and maintenance suggestion generation module is integrated into the power distribution network management system, and the power distribution network management system includes:

[0039] A distribution network topology diagram that uses different colors to identify the aging level of each cable; a cable information page that displays the aging index, aging level, and maintenance recommendations; a maintenance plan generator that automatically generates an overall maintenance plan based on the maintenance recommendations for each cable; and a historical data trend chart that displays the changing trend of the cable aging index.

[0040] In the preferred mode,

[0041] Based on the identified insulation aging characteristics and combined with the pre-established insulation aging assessment standards, the insulation aging index of the distribution cable is calculated to include:

[0042] Acquiring insulation aging characteristic data of a distribution cable; and performing quantitative processing on the insulation aging characteristic data, including:

[0043] Calculate the amplitude ratio and total harmonic distortion rate of the main harmonic components in the spectrum characteristics; extract the peak factor, rise time, fall time and asymmetry of the leakage current waveform in the time domain characteristics; calculate the mean, variance, skewness and kurtosis of the leakage current in the statistical characteristics;

[0044] Establish insulation aging assessment standards, including:

[0045] Determine an evaluation factor set, the evaluation factor set includes different insulation aging characteristics; determine a comment set, the comment set includes multiple levels; establish a single factor evaluation matrix, the single factor evaluation matrix represents the membership of each factor to the comment set; use the hierarchical analysis method to determine the weight of each evaluation factor to obtain a weight vector; perform fuzzy comprehensive evaluation to obtain a fuzzy evaluation result;

[0046] The characteristic data after quantization is normalized; the normalized characteristic data is multiplied and summed with the corresponding weight to obtain a preliminary insulation aging index; the preliminary insulation aging index is adjusted using the fuzzy comprehensive evaluation result; the adjusted insulation aging index is nonlinearly mapped to obtain the final insulation aging index.

[0047] The system for monitoring the insulation aging of distribution cables based on leakage current includes:

[0048] The first unit is used to collect leakage current data of the distribution cable in real time through a leakage current sensor arranged on the distribution cable; input the leakage current data into a data preprocessing module, and the data preprocessing module performs filtering, denoising and normalization on the leakage current data to obtain preprocessed leakage current data; compare the preprocessed leakage current data with pre-stored historical leakage current data to screen out abnormal leakage current data;

[0049] The second unit is used to perform time-frequency domain analysis on the abnormal leakage current data using multi-scale wavelet transform to extract the spectrum characteristics, amplitude characteristics and phase characteristics of the leakage current signal; input the extracted characteristics into a pre-trained deep learning model, the deep learning model includes a combination structure of a convolutional neural network and a long short-term memory network, and is used to identify insulation aging characteristics in the leakage current signal; based on the identified insulation aging characteristics, combined with a pre-established insulation aging evaluation standard, the insulation aging index of the distribution cable is calculated;

[0050] The third unit is used to compare the insulation aging index with the preset multi-level warning threshold to determine the aging level of the distribution cable; according to the aging level, call the preset decision rule library to generate corresponding maintenance suggestions; send the maintenance suggestions to the distribution system management center through the remote communication module; at the same time, store the insulation aging index, aging level and maintenance suggestions in the historical database for subsequent trend analysis and predictive maintenance strategy formulation; and regularly update and optimize the deep learning model and decision rule library.

[0051] An electronic device, comprising:

[0052] processor;

[0053] a memory for storing processor-executable instructions;

[0054] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0055] A computer-readable storage medium stores computer program instructions, which implement the aforementioned method when executed by a processor.

[0056] Beneficial effects of the present invention:

[0057] Real-time monitoring and intelligent diagnosis of distribution cable insulation aging are realized:

[0058] The present invention can detect insulation aging problems in a timely manner by setting leakage current sensors on distribution cables, collecting leakage current data in real time, and analyzing them in combination with multi-scale wavelet transform and deep learning models. This method overcomes the limitations of traditional periodic detection, greatly improves the timeliness and accuracy of insulation aging monitoring, and effectively reduces the risk of sudden failures in the distribution system.

[0059] Improved accuracy and reliability of insulation aging assessment:

[0060] The present invention uses multi-scale wavelet transform to analyze abnormal leakage current data in the time and frequency domain, extracts multi-dimensional features such as spectrum, amplitude and phase, and uses a combination of convolutional neural network and long short-term memory network for feature recognition. This method fully considers the complexity and timing characteristics of leakage current signals, and greatly improves the accuracy and reliability of insulation aging assessment compared to traditional single parameter evaluation methods.

[0061] Realize intelligent predictive maintenance of power distribution system:

[0062] The present invention can automatically generate maintenance suggestions based on the insulation aging index by establishing a multi-level warning threshold and decision rule library, and store relevant information in a historical database. This method realizes the transition from passive maintenance to active prevention, helps to formulate scientific and reasonable maintenance strategies, optimize resource allocation, significantly reduce maintenance costs, and improve the overall reliability and operating efficiency of the distribution system. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of a method for monitoring insulation aging of a power distribution cable based on leakage current according to an embodiment of the present invention;

[0064] Figure 2 It is a schematic diagram of the structure of a system for monitoring insulation aging of distribution cables based on leakage current according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0066] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0067] Figure 1 FIG. 1 is a flow chart of a method for monitoring insulation aging of a power distribution cable based on leakage current according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0068] S101. Collect leakage current data of the distribution cable in real time through a leakage current sensor installed on the distribution cable; input the leakage current data into a data preprocessing module, and the data preprocessing module performs filtering, denoising and normalization on the leakage current data to obtain preprocessed leakage current data; compare the preprocessed leakage current data with the pre-stored historical leakage current data to screen out abnormal leakage current data;

[0069] S102. Perform time-frequency domain analysis on the abnormal leakage current data using multi-scale wavelet transform to extract the spectrum characteristics, amplitude characteristics and phase characteristics of the leakage current signal; input the extracted characteristics into a pre-trained deep learning model, wherein the deep learning model includes a combination structure of a convolutional neural network and a long short-term memory network, and is used to identify insulation aging characteristics in the leakage current signal; based on the identified insulation aging characteristics, combined with a pre-established insulation aging evaluation standard, calculate the insulation aging index of the distribution cable;

[0070] S103. Compare the insulation aging index with the preset multi-level warning threshold to determine the aging level of the distribution cable; according to the aging level, call the preset decision rule library to generate corresponding maintenance suggestions; send the maintenance suggestions to the distribution system management center through the remote communication module; at the same time, store the insulation aging index, aging level and maintenance suggestions in the historical database for subsequent trend analysis and predictive maintenance strategy formulation; regularly update and optimize the deep learning model and decision rule library.

[0071] The method for monitoring the insulation aging of distribution cables based on leakage current is specifically implemented as follows:

[0072] First, a leakage current sensor is installed on the distribution cable to collect leakage current data in real time. A high-precision Hall sensor is used with a measurement range of 0-100mA and a sampling frequency of 1kHz. For example, 60,000 leakage current data points are collected in one minute for a certain cable section.

[0073] Next, the collected raw leakage current data is input into the data preprocessing module. This module uses a Butterworth low-pass filter to remove high-frequency noise above 50Hz, and then uses the wavelet threshold denoising method to further eliminate random noise. Finally, the data is normalized to the range of 0-1. The processed data is smoother and the noise is significantly reduced.

[0074] The preprocessed data is then compared with the normal leakage current data stored in the historical database. Using the 3σ criterion, data points that exceed the normal range by 3 times the standard deviation are marked as abnormal data. For example, if the leakage current is detected to suddenly increase from 5mA to 50mA, it is judged as abnormal.

[0075] Perform multi-scale wavelet transform on the screened abnormal data to obtain wavelet coefficients of different frequency bands. Select db4 wavelet and decompose it into 5 scales. Extract the energy, entropy and other characteristics of wavelet coefficients as spectrum characteristics; calculate the mean, variance and other statistics of each frequency band coefficient as amplitude characteristics; analyze the phase relationship between adjacent scale coefficients to obtain phase characteristics.

[0076] The extracted feature vector is input into a pre-trained deep learning model. The model is composed of a 3-layer convolutional neural network and a 2-layer long short-term memory network in series. The convolution layer is used to extract local features, and the long short-term memory network captures the temporal relationship. The model outputs the probability of insulation aging.

[0077] Based on the aging probability output by the model, the insulation aging index is calculated in combination with the pre-established evaluation criteria. The evaluation criteria take into account factors such as cable type and service life. For example, if the aging probability of a cable is 0.7, the corresponding aging index is 75.

[0078] The calculated aging index is compared with the preset multi-level warning thresholds to determine the aging level. Assuming that the warning thresholds are 60, 75, and 90 respectively, the aging level of the cable is moderate aging.

[0079] According to the aging level, the decision rule base is called to generate maintenance recommendations. For moderate aging, it is recommended to increase the monitoring frequency and arrange maintenance within 3 months.

[0080] The aging level and maintenance suggestions are sent to the power distribution system management center through the remote communication module. 4G wireless communication is used to achieve real-time alarm.

[0081] At the same time, the aging index, grade and recommendations are stored in the historical database. The database adopts a distributed architecture to support large-scale data storage and fast retrieval. These data are used for subsequent trend analysis and predictive maintenance.

[0082] Finally, the deep learning model and decision rule library are regularly updated and optimized. The model is retrained with new data every quarter to improve accuracy. According to the actual maintenance effect, the threshold and strategy of the decision rule are adjusted to continuously improve the system.

[0083] Through the above steps, the insulation aging monitoring of distribution cables based on leakage current is realized. This method uses deep learning technology to detect abnormal situations in time and provide strong support for the reliable operation of the distribution system.

[0084] In an optional implementation, the abnormal leakage current data is analyzed in the time-frequency domain using a multi-scale wavelet transform to extract the spectrum characteristics, amplitude characteristics and phase characteristics of the leakage current signal, including:

[0085] The Daubechies wavelet is selected as the basis function, and the input abnormal leakage current signal is decomposed by 5 layers of wavelet to obtain a low-frequency approximation coefficient and five detail coefficients of different scales;

[0086] Performing a fast Fourier transform on the wavelet coefficients of each scale to obtain a corresponding spectrum; calculating the energy distribution of each frequency band and extracting the main frequency components where the energy is concentrated; calculating the spectrum entropy corresponding to each scale coefficient to characterize the complexity of the signal; combining the main frequency components, energy distribution and spectrum entropy to form a spectrum feature vector;

[0087] Calculate the statistical characteristics of the wavelet coefficients of each scale, including mean, variance, skewness and kurtosis; extract the maximum value, minimum value and root mean square value of each scale coefficient; calculate the wavelet energy ratio, that is, the percentage of each scale energy to the total energy; combine the statistical characteristics, maximum value, minimum value, root mean square value and wavelet energy ratio to form an amplitude feature vector;

[0088] The instantaneous phase of the wavelet coefficients of each scale is calculated by using Hilbert transform; the statistical characteristics of the phase are extracted, including the phase mean and the phase variance; the phase synchronization index is calculated to characterize the phase relationship between different scales; the phase difference between different scales is calculated; the phase statistical characteristics, the phase synchronization index and the phase difference are combined to form a phase feature vector;

[0089] The frequency spectrum feature vector, the amplitude feature vector and the phase feature vector are combined to form a comprehensive feature vector, which fully reflects the characteristics of the abnormal leakage current signal in the time domain and the frequency domain.

[0090] According to the method, Daubechies wavelet is first selected as the basis function to perform 5-layer wavelet decomposition on the input abnormal leakage current signal. Specifically, the db4 wavelet can be used to decompose the leakage current signal with a sampling frequency of 10kHz and a duration of 1s to obtain a low-frequency approximation coefficient a5 and five detail coefficients of different scales d1, d2, d3, d4, and d5.

[0091] Next, perform a fast Fourier transform on the wavelet coefficients of each scale to obtain the corresponding spectrum. For example, perform FFT on the d1 coefficient to obtain a spectrum with a frequency range of 2500-5000Hz, and on the d2 coefficient to obtain a spectrum with a frequency range of 1250-2500Hz, and so on. Then calculate the energy distribution of each frequency band, and extract the frequency components with an energy share of more than 5% as the main frequency. At the same time, calculate the spectral entropy corresponding to each scale coefficient to characterize the complexity of the signal. The extracted main frequency, energy distribution percentage, and spectral entropy value are combined to form a spectral feature vector.

[0092] In terms of amplitude feature extraction, the statistical characteristics of the wavelet coefficients of each scale are first calculated, including mean, variance, skewness and kurtosis. Taking the d1 coefficient as an example, the mean may be 0.02, variance 1.5, skewness 0.1, kurtosis 3.2, etc. Then the maximum, minimum and root mean square value of each scale coefficient are extracted. Then the wavelet energy ratio is calculated, that is, the percentage of each scale energy to the total energy. These statistical features, extreme value features and energy features are combined to form an amplitude feature vector.

[0093] For phase features, the Hilbert transform is used to calculate the instantaneous phase of the wavelet coefficients at each scale. The statistical features of the phase are extracted, including the phase mean and phase variance. The phase synchronization index is calculated to characterize the phase relationship between different scales. Specifically, the phase locking value method can be used to calculate the variance of the phase difference between the two scale coefficients. The smaller the value, the better the synchronization. At the same time, the phase difference between different scales is calculated, such as the mean phase difference between d1 and d2, d2 and d3, etc. These phase features are combined to form a phase feature vector.

[0094] Finally, the obtained spectrum feature vector, amplitude feature vector and phase feature vector are combined to form a comprehensive feature vector. This vector may contain 50-100 feature components, which fully reflects the characteristics of abnormal leakage current signal in time domain and frequency domain. These features can be used as input for subsequent identification and classification of leakage fault types.

[0095] In practical applications, a large amount of sample data can be collected according to different types of leakage faults, the above features can be extracted and a feature library can be established. When an unknown leakage fault needs to be diagnosed, it is only necessary to extract the feature vector of the signal and match it with the samples in the feature library to identify the fault type. This feature extraction method based on multi-scale wavelet transform can fully exploit the time-frequency characteristics of the leakage current signal and improve the accuracy and reliability of fault diagnosis.

[0096] In an optional embodiment, the extracted features are input into a pre-trained deep learning model, wherein the deep learning model includes a combination structure of a convolutional neural network and a long short-term memory network, and the method for identifying insulation aging features in the leakage current signal includes:

[0097] Constructing a combined structure integrating a convolutional neural network and a long short-term memory network, the combined structure comprising an input layer, a convolutional neural network part, a long short-term memory network part and a fully connected layer; wherein the input layer receives a comprehensive feature vector, and the comprehensive feature vector comprises a spectrum feature, an amplitude feature and a phase feature;

[0098] The convolutional neural network part includes a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer and a flattening layer; the first convolution layer uses a plurality of three-by-three convolution kernels to perform convolution operations, the first pooling layer and the second pooling layer both use maximum pooling operations, and the flattening layer converts a two-dimensional feature map into a one-dimensional vector;

[0099] The long short-term memory network part includes a first long short-term memory layer and a second long short-term memory layer, the first long short-term memory layer returns a complete sequence, and the second long short-term memory layer only returns the output of the last time step;

[0100] The fully connected layer includes a first fully connected layer, a random dropout layer, a second fully connected layer and an output layer, and the number of neurons in the output layer is equal to the number of insulation aging levels;

[0101] The historical leakage current data and the corresponding insulation aging level labels are divided into a training set, a validation set and a test set; the training set is subjected to data enhancement processing, wherein the data enhancement processing includes adding Gaussian noise, time shift and amplitude scaling; the combined structure is trained using a cross entropy loss function and an Adam optimizer, and a learning rate scheduling strategy and an early stopping strategy are adopted during the training process; the model performance is evaluated on the test set, and the confusion matrix, precision, recall rate and F1 score are calculated; the model structure and hyperparameters are fine-tuned according to the evaluation results; an integration method or a gradient method is used to analyze the contribution of different features to model prediction; SHapley Additive exPlanations values ​​are used to interpret the prediction results of the model, and the most critical features for insulation aging judgment are identified.

[0102] In a specific implementation, the process of inputting the extracted features into a pre-trained deep learning model for insulation aging feature recognition is as follows:

[0103] First, a combined structure integrating a convolutional neural network and a long short-term memory network is constructed. The combined structure includes an input layer, a convolutional neural network part, a long short-term memory network part, and a fully connected layer. The input layer receives a comprehensive feature vector, which includes a spectrum feature, an amplitude feature, and a phase feature. For example, the spectrum feature may include the peak and valley of the power spectrum density, the amplitude feature may include the maximum value, the minimum value, and the root mean square value, and the phase feature may include the statistics of the phase angle.

[0104] The convolutional neural network part includes two convolutional layers, two pooling layers, and a flattening layer. The first convolutional layer uses 32 3×3 convolutional kernels, with a stride of 1, a padding method of "same", and a ReLU activation function. The first pooling layer uses 2×2 max pooling with a stride of 2. The second convolutional layer uses 64 3×3 convolutional kernels, and the other parameters are the same as those of the first convolutional layer. The parameters of the second pooling layer are the same as those of the first pooling layer. The flattening layer converts the two-dimensional feature map into a one-dimensional vector.

[0105] The long short-term memory network part includes two long short-term memory layers. The first long short-term memory layer contains 128 units and returns the complete sequence. The second long short-term memory layer contains 64 units and only returns the output of the last time step. Both long short-term memory layers use tanh as the activation function and sigmoid as the recurrent activation function.

[0106] The fully connected layer includes two fully connected layers, a dropout layer, and an output layer. The first fully connected layer contains 128 neurons and uses the ReLU activation function. The dropout rate of the dropout layer is set to 0.5. The second fully connected layer contains 64 neurons and uses the ReLU activation function. The number of neurons in the output layer is equal to the number of insulation aging levels, and the softmax activation function is used. For example, if the insulation aging is divided into 5 levels, the output layer contains 5 neurons.

[0107] Next, prepare the dataset. Randomly divide the historical leakage current data and their corresponding insulation aging level labels into a training set, a validation set, and a test set according to a ratio of 7:2:1. Perform data augmentation on the training set, including: 1) adding Gaussian noise with a mean of 0 and a standard deviation of 0.01; 2) randomly translating in the range of -10% to 10% on the time axis; 3) randomly scaling the signal amplitude by a factor of 0.9 to 1.1. This can expand the training samples and improve the generalization ability of the model.

[0108] Then start training the model. Use cross-entropy as the loss function, Adam as the optimizer, and set the initial learning rate to 0.001. Adopt a learning rate scheduling strategy. When the validation set loss does not decrease for 5 consecutive epochs, halve the learning rate. At the same time, adopt an early stopping strategy. When the validation set loss does not decrease for 10 consecutive epochs, stop training. After each epoch, evaluate the model performance on the validation set and record the loss and accuracy.

[0109] After training is completed, evaluate the model performance on the test set. Calculate the confusion matrix to obtain the precision, recall, and F1-score for each category, as well as the overall macro-average and weighted-average metrics. For example, for a 5-classification problem, the following results may be obtained: precision is 0.92, recall is 0.91, and F1-score is 0.915.

[0110] Fine-tune the model structure and hyperparameters based on the evaluation results. You can try adjusting the number and number of units in the convolutional layers and long short-term memory layers, changing the structure of the fully connected layers, or adjusting hyperparameters such as the learning rate and batch size. Retrain and evaluate the model after each adjustment until the performance no longer improves significantly.

[0111] In order to analyze the contribution of different features to model prediction, an ensemble method or a gradient method can be used. The ensemble method evaluates feature importance by training multiple sub-models, each of which uses a different subset of features, and comparing the performance of the sub-models. The gradient method calculates the gradient of the prediction result to the input feature. The larger the absolute value of the gradient, the greater the influence of the feature on the prediction result.

[0112] Finally, the SHAP (SHapley Additive exPlanations) value is used to explain the prediction results of the model. The SHAP value is based on the Shapley value in game theory and can quantify the contribution of each feature to the prediction results. By analyzing the SHAP values ​​of a large number of samples, the most critical features for insulation aging judgment can be identified. For example, it may be found that a specific frequency band or phase feature in the spectrum has the greatest impact on the prediction results.

[0113] Through the above steps, an efficient deep learning model can be constructed to identify the insulation aging characteristics in the leakage current signal, and the prediction results of the model can be interpreted to provide a reliable basis for insulation status assessment.

[0114] In an optional implementation, the insulation aging index is compared with a preset multi-level warning threshold to determine the aging level of the distribution cable; according to the aging level, a preset decision rule library is called to generate corresponding maintenance suggestions including:

[0115] Obtaining an insulation aging index of a distribution cable; comparing the insulation aging index with a preset multi-level warning threshold to determine an aging level of the distribution cable, wherein the multi-level warning threshold is set by the following steps:

[0116] Collect cable sample data with known aging degree; use deep learning model to calculate insulation aging index of the cable sample data; use clustering algorithm to group the calculated insulation aging index; fine-tune the clustering results based on the experience of power system experts to determine the threshold of each level;

[0117] A decision rule base is constructed, wherein the decision rule base includes aging level, cable importance, cable service life, cable current carrying capacity utilization, environmental factors and historical fault records, wherein the construction process of the decision rule base includes:

[0118] Gather the knowledge and experience of power system experts; organize historical maintenance records and analyze maintenance strategies and their effects under different circumstances; use decision tree algorithms to build preliminary decision rules; optimize and adjust decision rules through expert review and field verification; encode the optimized decision rules into a form that can be processed by computers;

[0119] Generating maintenance suggestions according to the aging level and the decision rule base specifically includes:

[0120] Input the aging level and other relevant parameters of the distribution cable; match the rules that meet the conditions in the decision rule library; when multiple rules are matched, sort them according to the priority of the rules; select the rule with the highest priority and generate corresponding maintenance suggestions.

[0121] The specific implementation is as follows:

[0122] First, obtain the insulation aging index of the distribution cable. This can be achieved in a variety of ways, such as using online monitoring equipment to collect cable parameters such as partial discharge and dielectric loss, or obtaining relevant data through regular offline testing. Taking a 10kV distribution cable as an example, the partial discharge amplitude collected by the online monitoring system is 25pC, and the dielectric loss tanδ is 0.008.

[0123] Next, the insulation aging index is compared with the preset multi-level warning threshold to determine the aging level of the distribution cable. The multi-level warning threshold is set by the following steps:

[0124] First, collect cable sample data with known aging degree. For example, collect 100 10kV cable sample data with different service life and different operating environment from a regional power company, including parameters such as partial discharge and dielectric loss as well as actual aging status.

[0125] The deep learning model is then used to calculate the insulation aging index of these cable sample data. A convolutional neural network (CNN) model can be used to take various cable parameters as input and output an insulation aging index between 0 and 1. After training and verification, the accuracy of the model on the test set reached 95%.

[0126] Then, a clustering algorithm is used to group the calculated insulation aging index. Here, the K-means clustering algorithm can be used to divide the sample data into 5 categories, corresponding to different aging levels.

[0127] Finally, the clustering results were fine-tuned based on the experience of power system experts to determine the thresholds of each level. For example, after expert discussion, the thresholds of the five levels were finally determined to be 0.2, 0.4, 0.6, and 0.8.

[0128] For the distribution cable in the previous example, the calculated insulation aging index is 0.65, and the corresponding aging level is 4 (severe aging).

[0129] The next step is to build a decision rule base. The decision rule base contains information such as aging level, cable importance, cable age, cable current utilization, environmental factors, and historical fault records. The construction process is as follows:

[0130] First, the knowledge and experience of power system experts were collected. By holding expert seminars, the key factors and their weights that affect cable maintenance decisions were sorted out.

[0131] Then, we sort out the historical maintenance records and analyze the maintenance strategies and their effects under different circumstances. For example, we collect the maintenance records of 100 cables in the past five years and analyze the maintenance measures and their effects for cables of different aging levels and importance.

[0132] Then, the decision tree algorithm is used to construct preliminary decision rules. Using the C4.5 decision tree algorithm, historical maintenance records are used as training data to build a decision tree model containing 50 decision nodes.

[0133] The decision rules were then optimized and adjusted through expert review and field verification. Five senior power experts were invited to review the decision tree model and conduct field verification on 10 typical cables. The model was optimized based on the feedback results.

[0134] Finally, the optimized decision rules are encoded into a computer-processable form. The decision tree is converted into an IF-THEN rule set and stored in a relational database for fast query and matching.

[0135] With the decision rule base, maintenance recommendations can be generated based on the cable aging level and other relevant parameters. The specific steps are as follows:

[0136] First, enter the aging level and other relevant parameters of the distribution cable. For the cable in the previous example, enter aging level 4, importance as medium, service life as 15 years, current carrying capacity utilization rate as 80%, environmental factor as humidity, and historical fault record as 2 times.

[0137] Then, the decision rule base matches the rules that meet the conditions. The system searches the database for all rules that meet the input conditions, and may match multiple rules.

[0138] When multiple rules are matched, they are sorted according to their priority. The priority can be calculated based on indicators such as the credibility and support of the rule. For example, if three rules are matched, the priorities are 0.9, 0.8, and 0.7 respectively.

[0139] Finally, the rule with the highest priority is selected to generate the corresponding maintenance recommendation. For this example, the maintenance recommendation generated by the system may be: "It is recommended to replace this cable within 3 months, and at the same time increase the frequency of daily inspections and perform infrared temperature checks once a week."

[0140] Through the above steps, a distribution cable maintenance decision method based on insulation aging index is realized. This method combines data-driven and expert experience, can provide scientific and reasonable decision support for power system operation and maintenance personnel, and help improve the reliability and economy of the distribution network.

[0141] In an optional implementation, the method further includes:

[0142] The maintenance suggestion is dynamically updated, including:

[0143] Record each maintenance operation and its effect; regularly analyze maintenance effect data; adjust the corresponding decision rules when it is found that some maintenance suggestions are not effective; use reinforcement learning algorithms to continuously optimize decision rules;

[0144] The aging level judgment and maintenance suggestion generation module is integrated into the power distribution network management system, and the power distribution network management system includes:

[0145] A distribution network topology diagram that uses different colors to identify the aging level of each cable; a cable information page that displays the aging index, aging level, and maintenance recommendations; a maintenance plan generator that automatically generates an overall maintenance plan based on the maintenance recommendations for each cable; and a historical data trend chart that displays the changing trend of the cable aging index.

[0146] According to the method, dynamic updating of maintenance recommendations is a continuous optimization process. First, the specific content of each maintenance operation needs to be recorded in detail, including the maintenance time, location, operator, specific measures taken, etc. At the same time, the changes in cable performance after maintenance should be recorded, such as the increase in insulation resistance value, the reduction in partial discharge level, etc. These records can be stored in the database of the distribution network management system for subsequent analysis.

[0147] Regular analysis of maintenance effect data is a key step in dynamic updating. A comprehensive analysis can be set up every quarter or every six months, or the analysis cycle can be flexibly adjusted according to actual conditions. When analyzing, it is necessary to compare the changes in various indicators before and after maintenance to evaluate the effects of different types of maintenance measures. For example, the average increase in insulation resistance after cleaning insulators, or the average reduction in partial discharge levels after replacing joints can be counted. These data can be used to quantify the effects of different maintenance measures.

[0148] When it is found that some maintenance suggestions are not effective, the corresponding decision rules need to be adjusted in time. For example, if it is found that the effect of replacing a certain type of joint is not obvious, you can consider adjusting the trigger conditions for joint replacement and raising the replacement threshold. For another example, if a certain cleaning method is found to be effective, the condition threshold for adopting this method can be lowered accordingly. These adjustments can be achieved by modifying the decision rules in the distribution network management system.

[0149] Using reinforcement learning algorithms to continuously optimize decision rules is an advanced dynamic update method. Each maintenance operation can be regarded as an "action", and the performance improvement after maintenance can be regarded as a "reward". By constantly trying different maintenance strategies, the optimal decision rule can be gradually found. For example, an initial maintenance decision rule can be set, and then various parameters such as maintenance trigger conditions and priority sorting can be continuously adjusted in actual applications. These parameters can be optimized based on the feedback of maintenance effects, and finally a more ideal decision rule can be obtained.

[0150] Integrating the aging level judgment and maintenance suggestion generation modules into the distribution network management system can realize the visual management of cable status. In the distribution network topology diagram, different colors can be used to identify the aging level of each cable, such as green for normal, yellow for mild aging, orange for moderate aging, and red for severe aging. This can intuitively display the aging status of the entire network, making it easier for managers to quickly identify areas that need to be focused on.

[0151] In the cable information page, the aging index, aging level and maintenance recommendations of each cable can be displayed in detail. The aging index can be accurate to two decimal places, and the aging level can be described in words, such as "mild aging", "moderate aging", etc. The maintenance recommendations can include specific measures recommended, the recommended implementation time, and the expected results. For example, for a cable with an aging index of 0.75 and an aging level of "moderate aging", the system may give the following maintenance recommendations: "It is recommended to clean the insulators and conduct partial discharge tests within 3 months, which is expected to reduce the aging index to below 0.6".

[0152] The maintenance plan generator can automatically generate an overall maintenance plan based on the maintenance recommendations for each cable. It will consider factors such as the ageing degree, maintenance urgency, and geographical location of each cable to reasonably arrange the maintenance sequence and time. For example, cables with a higher degree of ageing can be prioritized, and cables with similar geographical locations can be maintained in the same time period to improve maintenance efficiency. The generated maintenance plan can include specific timetables, required manpower and material resources, and expected completion time.

[0153] The historical data trend chart can intuitively show the changing trend of the cable aging index. You can choose different time scales, such as monthly, quarterly or annual, to draw the changing curve of the aging index. This can clearly show the acceleration or slowdown trend of cable aging, which is helpful for predicting future aging conditions and formulating long-term maintenance strategies. For example, if the aging index of a cable is observed to rise rapidly in the near future, it may be necessary to conduct more frequent inspections or consider replacing it in advance.

[0154] Through the above method, the dynamic update and visual management of cable maintenance suggestions can be realized, and the operation and maintenance efficiency and reliability of the distribution network can be improved. This method can continuously optimize the decision-making rules according to the actual maintenance effect, adapt to the characteristics of different regions and different types of cables, and provide strong support for distribution network management.

[0155] In an optional embodiment, based on the identified insulation aging characteristics and in combination with a pre-established insulation aging evaluation standard, the insulation aging index of the distribution cable is calculated to include:

[0156] Acquiring insulation aging characteristic data of a distribution cable; and performing quantitative processing on the insulation aging characteristic data, including:

[0157] Calculate the amplitude ratio and total harmonic distortion rate of the main harmonic components in the spectrum characteristics; extract the peak factor, rise time, fall time and asymmetry of the leakage current waveform in the time domain characteristics; calculate the mean, variance, skewness and kurtosis of the leakage current in the statistical characteristics;

[0158] Establish insulation aging assessment standards, including:

[0159] Determine an evaluation factor set, the evaluation factor set includes different insulation aging characteristics; determine a comment set, the comment set includes multiple levels; establish a single factor evaluation matrix, the single factor evaluation matrix represents the membership of each factor to the comment set; use the hierarchical analysis method to determine the weight of each evaluation factor to obtain a weight vector; perform fuzzy comprehensive evaluation to obtain a fuzzy evaluation result;

[0160] The characteristic data after quantization is normalized; the normalized characteristic data is multiplied and summed with the corresponding weight to obtain a preliminary insulation aging index; the preliminary insulation aging index is adjusted using the fuzzy comprehensive evaluation result; the adjusted insulation aging index is nonlinearly mapped to obtain the final insulation aging index.

[0161] In practical applications, based on the identified insulation aging characteristics and combined with the pre-established insulation aging evaluation standards, the specific implementation method for calculating the insulation aging index of the distribution cable is as follows:

[0162] First, the insulation aging characteristic data of the distribution cable is obtained. These data usually include spectrum characteristics, time domain characteristics and statistical characteristics. The spectrum characteristics reflect the frequency distribution of the leakage current signal, the time domain characteristics reflect the morphological characteristics of the leakage current waveform, and the statistical characteristics reflect the overall statistical laws of the leakage current data.

[0163] The obtained insulation aging characteristic data is quantified. For spectral characteristics, the amplitude ratio of the main harmonic components (such as fundamental wave, 3rd harmonic, 5th harmonic, etc.) and the total harmonic distortion rate are calculated. For example, the amplitude ratio of the 3rd harmonic of a distribution cable is 0.15, the amplitude ratio of the 5th harmonic is 0.08, and the total harmonic distortion rate is 18%. For time domain characteristics, the peak factor, rise time, fall time and asymmetry of the leakage current waveform are extracted. For example, the peak factor is 1.8, the rise time is 2ms, the fall time is 3ms, and the asymmetry is 0.12. For statistical characteristics, the mean, variance, skewness and kurtosis of the leakage current are calculated. For example, the mean is 5mA, the variance is 0.8, the skewness is 0.3, and the kurtosis is 3.2.

[0164] Next, establish the insulation aging evaluation standard. First, determine the evaluation factor set, which includes the various insulation aging characteristics obtained by the above quantitative processing. Then determine the comment set, which can be divided into five levels, such as "excellent", "good", "general", "poor", and "serious". Establish a single factor evaluation matrix to represent the membership of each factor to the comment set. For example, for the third harmonic amplitude ratio, its membership to the five levels may be [0.1, 0.3, 0.4, 0.2, 0]. Use the hierarchical analysis method to determine the weight of each evaluation factor and obtain the weight vector. For example, the weights of spectral features, time domain features, and statistical features may be 0.4, 0.35, and 0.25, respectively. Perform fuzzy comprehensive evaluation to obtain fuzzy evaluation results. For example, the final fuzzy evaluation result may be [0.15, 0.25, 0.35, 0.2, 0.05].

[0165] The quantized feature data is normalized so that data of different dimensions can be compared. For example, the minimum-maximum normalization method can be used to map each feature value to the interval [0,1]. The normalized feature data is multiplied by the corresponding weight and summed to obtain a preliminary insulation aging index. For example, assuming that the normalized feature data is [0.6, 0.7, 0.5, 0.8, 0.4] and the corresponding weights are [0.2, 0.25, 0.15, 0.3, 0.1], the preliminary insulation aging index is 0.63.

[0166] Adjust the preliminary insulation aging index using the fuzzy comprehensive evaluation results. The preliminary index can be weighted and adjusted according to the membership degrees of each level in the fuzzy evaluation results. For example, if the membership degree of the "general" level is the highest in the fuzzy evaluation results, the preliminary index can be adjusted closer to the intermediate value of 0.5.

[0167] Finally, perform a non-linear mapping on the adjusted insulation aging index to obtain the final insulation aging index. The non-linear mapping can make the index distribution more reasonable and highlight the differences in the key intervals. For example, an S-shaped function can be used for mapping to make the changes in the intermediate region more sensitive. Suppose the adjusted index is 0.58, after non-linear mapping, the final insulation aging index may be 0.65.

[0168] Through the above steps, the complex insulation aging characteristics can be synthesized into an intuitive index, providing strong support for evaluating the insulation condition of the distribution cable. This method combines multiple characteristics and evaluation techniques, can comprehensively reflect the insulation aging degree of the cable, and provides an important basis for subsequent maintenance decisions.

[0169] Figure 2 The following is a schematic structural diagram of the system for monitoring the insulation aging of a distribution cable based on leakage current in an embodiment of the present invention. As Figure 2 shown, the system includes:

[0170] A first unit for real-time collecting the leakage current data of the distribution cable through a leakage current sensor arranged on the distribution cable; inputting the leakage current data into a data preprocessing module, and the data preprocessing module filters, denoises, and normalizes the leakage current data to obtain preprocessed leakage current data; comparing the preprocessed leakage current data with the pre-stored historical leakage current data to screen out abnormal leakage current data;

[0171] A second unit for performing time-frequency domain analysis on the abnormal leakage current data using multi-scale wavelet transform to extract the spectral characteristics, amplitude characteristics, and phase characteristics of the leakage current signal; inputting the extracted characteristics into a pre-trained deep learning model, the deep learning model includes a combined structure of a convolutional neural network and a long short-term memory network, and is used to identify the insulation aging characteristics in the leakage current signal; based on the identified insulation aging characteristics, combining the pre-established insulation aging evaluation criteria, calculating the insulation aging index of the distribution cable;

[0172] The third unit is used to compare the insulation aging index with the preset multi-level warning threshold to determine the aging level of the distribution cable; according to the aging level, call the preset decision rule library to generate corresponding maintenance suggestions; send the maintenance suggestions to the distribution system management center through the remote communication module; at the same time, store the insulation aging index, aging level and maintenance suggestions in the historical database for subsequent trend analysis and predictive maintenance strategy formulation; and regularly update and optimize the deep learning model and decision rule library.

[0173] According to a third aspect of the embodiments of the present invention,

[0174] An electronic device is provided, comprising:

[0175] processor;

[0176] a memory for storing processor-executable instructions;

[0177] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0178] A fourth aspect of the embodiments of the present invention is:

[0179] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0180] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring insulation aging of distribution cables based on leakage current, characterized in that: include: The leakage current data of the distribution cable is collected in real time by means of the leakage current sensor arranged on the distribution cable; Inputting the leakage current data into a data preprocessing module, the data preprocessing module performs filtering, denoising and normalization processing on the leakage current data to obtain preprocessed leakage current data; Comparing the pre-processed leakage current data with pre-stored historical leakage current data to screen out abnormal leakage current data; Performing time-frequency domain analysis on the abnormal leakage current data by using multi-scale wavelet transform to extract the frequency spectrum characteristics, amplitude characteristics and phase characteristics of the leakage current signal; Inputting the extracted features into a pre-trained deep learning model, wherein the deep learning model includes a combination structure of a convolutional neural network and a long short-term memory network, for identifying insulation aging features in the leakage current signal; Based on the identified insulation aging characteristics and in combination with the pre-established insulation aging assessment standards, the insulation aging index of the distribution cable is calculated; Comparing the insulation aging index with a preset multi-level warning threshold to determine the aging level of the distribution cable; According to the aging level, a preset decision rule library is called to generate corresponding maintenance suggestions; the maintenance suggestions are sent to the distribution system management center through a remote communication module; at the same time, the insulation aging index, aging level and maintenance suggestions are stored in a historical database for subsequent trend analysis and predictive maintenance strategy formulation; the deep learning model and decision rule library are regularly updated and optimized.

2. The method for monitoring insulation aging of distribution cables based on leakage current according to claim 1 is characterized in that: The abnormal leakage current data is analyzed in the time and frequency domain by using multi-scale wavelet transform to extract the spectrum characteristics, amplitude characteristics and phase characteristics of the leakage current signal, including: The Daubechies wavelet is selected as the basis function, and the input abnormal leakage current signal is decomposed by 5 layers of wavelet to obtain a low-frequency approximation coefficient and five detail coefficients of different scales; Performing a fast Fourier transform on the wavelet coefficients of each scale to obtain a corresponding spectrum; calculating the energy distribution of each frequency band and extracting the main frequency components where the energy is concentrated; calculating the spectrum entropy corresponding to each scale coefficient to characterize the complexity of the signal; combining the main frequency components, energy distribution and spectrum entropy to form a spectrum feature vector; Calculate the statistical characteristics of the wavelet coefficients of each scale, including mean, variance, skewness and kurtosis; extract the maximum value, minimum value and root mean square value of each scale coefficient; calculate the wavelet energy ratio, that is, the percentage of each scale energy to the total energy; combine the statistical characteristics, maximum value, minimum value, root mean square value and wavelet energy ratio to form an amplitude feature vector; The instantaneous phase of the wavelet coefficients of each scale is calculated by using Hilbert transform; the statistical characteristics of the phase are extracted, including the phase mean and the phase variance; the phase synchronization index is calculated to characterize the phase relationship between different scales; the phase difference between different scales is calculated; the phase statistical characteristics, the phase synchronization index and the phase difference are combined to form a phase feature vector; The frequency spectrum feature vector, the amplitude feature vector and the phase feature vector are combined to form a comprehensive feature vector, which fully reflects the characteristics of the abnormal leakage current signal in the time domain and the frequency domain.

3. The method for monitoring insulation aging of distribution cables based on leakage current according to claim 1 is characterized in that: The extracted features are input into a pre-trained deep learning model, which includes a combination of a convolutional neural network and a long short-term memory network, and is used to identify insulation aging features in leakage current signals, including: Constructing a combined structure integrating a convolutional neural network and a long short-term memory network, the combined structure comprising an input layer, a convolutional neural network part, a long short-term memory network part and a fully connected layer; wherein the input layer receives a comprehensive feature vector, and the comprehensive feature vector comprises a spectrum feature, an amplitude feature and a phase feature; The convolutional neural network part includes a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer and a flattening layer; the first convolution layer uses a plurality of three-by-three convolution kernels to perform convolution operations, the first pooling layer and the second pooling layer both use maximum pooling operations, and the flattening layer converts a two-dimensional feature map into a one-dimensional vector; The long short-term memory network part includes a first long short-term memory layer and a second long short-term memory layer, the first long short-term memory layer returns a complete sequence, and the second long short-term memory layer only returns the output of the last time step; The fully connected layer includes a first fully connected layer, a random dropout layer, a second fully connected layer and an output layer, and the number of neurons in the output layer is equal to the number of insulation aging levels; The historical leakage current data and the corresponding insulation aging level labels are divided into a training set, a validation set and a test set; the training set is subjected to data enhancement processing, wherein the data enhancement processing includes adding Gaussian noise, time shift and amplitude scaling; the combined structure is trained using a cross entropy loss function and an Adam optimizer, and a learning rate scheduling strategy and an early stopping strategy are adopted during the training process; the model performance is evaluated on the test set, and the confusion matrix, precision, recall rate and F1 score are calculated; the model structure and hyperparameters are fine-tuned according to the evaluation results; an integration method or a gradient method is used to analyze the contribution of different features to model prediction; SHapley Additive exPlanations values ​​are used to interpret the prediction results of the model, and the most critical features for insulation aging judgment are identified.

4. The method for monitoring insulation aging of distribution cables based on leakage current according to claim 1 is characterized in that: Comparing the insulation aging index with a preset multi-level warning threshold to determine the aging level of the distribution cable; According to the aging level, a pre-set decision rule library is called to generate corresponding maintenance suggestions including: Obtaining an insulation aging index of a distribution cable; comparing the insulation aging index with a preset multi-level warning threshold to determine an aging level of the distribution cable, wherein the multi-level warning threshold is set by the following steps: Collect cable sample data with known aging degree; use deep learning model to calculate insulation aging index of the cable sample data; use clustering algorithm to group the calculated insulation aging index; fine-tune the clustering results based on the experience of power system experts to determine the threshold of each level; A decision rule base is constructed, wherein the decision rule base includes aging level, cable importance, cable service life, cable current carrying capacity utilization, environmental factors and historical fault records, wherein the construction process of the decision rule base includes: Gather the knowledge and experience of power system experts; organize historical maintenance records and analyze maintenance strategies and their effects under different circumstances; use decision tree algorithms to build preliminary decision rules; optimize and adjust decision rules through expert review and field verification; encode the optimized decision rules into a form that can be processed by computers; Generating maintenance suggestions according to the aging level and the decision rule base specifically includes: Input the aging level and other relevant parameters of the distribution cable; match the qualified rules in the decision rule library; when multiple rules are matched, sort them according to the priority of the rules; select the rule with the highest priority and generate maintenance suggestions.

5. The method for monitoring insulation aging of distribution cables based on leakage current according to claim 4 is characterized in that: The method further comprises: The maintenance suggestion is dynamically updated, including: Record each maintenance operation and its effect; regularly analyze maintenance effect data; adjust the corresponding decision rules when it is found that some maintenance suggestions are not effective; use reinforcement learning algorithms to continuously optimize decision rules; The aging level judgment and maintenance suggestion generation module is integrated into the power distribution network management system, and the power distribution network management system includes: A distribution network topology diagram that uses different colors to identify the aging level of each cable; a cable information page that displays the aging index, aging level, and maintenance recommendations; a maintenance plan generator that automatically generates an overall maintenance plan based on the maintenance recommendations for each cable; and a historical data trend chart that displays the changing trend of the cable aging index.

6. The method for monitoring insulation aging of distribution cables based on leakage current according to claim 1 is characterized in that: Based on the identified insulation aging characteristics and combined with the pre-established insulation aging assessment standards, the insulation aging index of the distribution cable is calculated to include: Acquiring insulation aging characteristic data of a distribution cable; and performing quantitative processing on the insulation aging characteristic data, including: Calculate the amplitude ratio and total harmonic distortion rate of the main harmonic components in the spectrum characteristics; extract the peak factor, rise time, fall time and asymmetry of the leakage current waveform in the time domain characteristics; calculate the mean, variance, skewness and kurtosis of the leakage current in the statistical characteristics; Establish insulation aging assessment standards, including: Determine an evaluation factor set, the evaluation factor set includes different insulation aging characteristics; determine a comment set, the comment set includes multiple levels; establish a single factor evaluation matrix, the single factor evaluation matrix represents the membership of each factor to the comment set; use the hierarchical analysis method to determine the weight of each evaluation factor to obtain a weight vector; perform fuzzy comprehensive evaluation to obtain a fuzzy evaluation result; The characteristic data after quantization is normalized; the normalized characteristic data is multiplied and summed with the corresponding weight to obtain a preliminary insulation aging index; the preliminary insulation aging index is adjusted using the fuzzy comprehensive evaluation result; the adjusted insulation aging index is nonlinearly mapped to obtain the final insulation aging index.

7. A system for monitoring insulation aging of distribution cables based on leakage current, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to collect leakage current data of the distribution cable in real time through a leakage current sensor arranged on the distribution cable; Inputting the leakage current data into a data preprocessing module, the data preprocessing module performs filtering, denoising and normalization processing on the leakage current data to obtain preprocessed leakage current data; Comparing the pre-processed leakage current data with pre-stored historical leakage current data to screen out abnormal leakage current data; The second unit is used to perform time-frequency domain analysis on the abnormal leakage current data by using multi-scale wavelet transform to extract the frequency spectrum characteristics, amplitude characteristics and phase characteristics of the leakage current signal; Inputting the extracted features into a pre-trained deep learning model, wherein the deep learning model includes a combination structure of a convolutional neural network and a long short-term memory network, for identifying insulation aging features in the leakage current signal; Based on the identified insulation aging characteristics and in combination with the pre-established insulation aging assessment standards, the insulation aging index of the distribution cable is calculated; A third unit is used to compare the insulation aging index with a preset multi-level warning threshold to determine the aging level of the distribution cable; According to the aging level, a preset decision rule library is called to generate corresponding maintenance suggestions; the maintenance suggestions are sent to the distribution system management center through a remote communication module; at the same time, the insulation aging index, aging level and maintenance suggestions are stored in a historical database for subsequent trend analysis and predictive maintenance strategy formulation; the deep learning model and decision rule library are regularly updated and optimized.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images