Partial discharge identification method and device, computer equipment, storage medium and program product

Abstract

The invention relates to a partial discharge identification method and device, computer equipment, a storage medium and a program product. The method comprises the following steps: acquiring test partial discharge data, and performing state quality grading on the test partial discharge data to obtain graded data; inputting the hierarchical data into a target convolutional neural network to obtain extracted deep implicit features, the target convolutional neural network being obtained by training an initial convolutional neural network based on the hierarchical data until convergence; establishing a random forest grading model based on the deep hidden features and original features corresponding to the grading data; and performing state quality grade identification on to-be-identified partial discharge data based on the random forest grading model to obtain an identified state quality grade result. By adopting the method, the accuracy of partial discharge identification can be improved.

Description

Technical Field

[0001] This application relates to the technical field of electric power, and particularly to a method, device, computer device, storage medium, and program product for partial discharge recognition. Background Art

[0002] At present, the demand for electricity is increasing, and more and more cables are being used. However, cable failures are also increasing. Moreover, since cables are generally buried underground, troubleshooting and repair are time-consuming when a failure occurs.

[0003] Partial discharge is a common fault type of cables. If partial discharge can be effectively identified, the reliability and safety of the power system can be improved. However, some current machine learning methods have low accuracy in identifying partial discharge. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method, device, computer device, storage medium, and program product for partial discharge recognition that can improve the accuracy.

[0005] In a first aspect, this application provides a method for partial discharge recognition, including:

[0006] Obtain experimental partial discharge data, and perform a division of the state quality level on the experimental partial discharge data to obtain classified data;

[0007] Input the classified data into a target convolutional neural network to obtain extracted deep implicit features, where the target convolutional neural network is obtained by training an initial convolutional neural network based on the classified data until convergence;

[0008] Based on the deep implicit features and the original features corresponding to the classified data, establish a random forest classification model;

[0009] Based on the random forest classification model, perform a state quality level recognition on the partial discharge data to be recognized to obtain the recognized state quality level result.

[0010] In one of the embodiments, the establishing a random forest classification model based on the deep implicit features and the original features corresponding to the classified data includes:

[0011] Combine the deep implicit features and the original features corresponding to the classified data to obtain a fused feature;

[0012] Perform independent random sampling on the fused feature to obtain multiple sampling sets;

[0013] Based on the multiple sampling sets, perform training to obtain multiple decision trees;

[0014] Combine the multiple decision trees to obtain the random forest classification model.

[0015] In one embodiment, the step of inputting the classification data into the target convolutional neural network to obtain the extracted deep implicit features includes:

[0016] Divide the classification data into a training set and a test set;

[0017] Input the training set into the target convolutional neural network to obtain the deep implicit features, where the target convolutional neural network is obtained by training the initial convolutional neural network based on the training set until convergence.

[0018] In one embodiment, before the step of identifying the state quality level of the local discharge data to be recognized based on the random forest classification model to obtain the recognized state quality level result, the method further includes:

[0019] Test the random forest classification model based on the test set to obtain evaluation metrics;

[0020] The step of identifying the state quality level of the local discharge data to be recognized based on the random forest classification model to obtain the recognized state quality level result includes:

[0021] When the evaluation metrics meet the preset metric requirements, identify the state quality level of the local discharge data to be recognized based on the random forest classification model to obtain the recognized state quality level result.

[0022] In one embodiment, the step of dividing the state quality level of the test local discharge data to obtain classification data includes:

[0023] Obtain the state parameters corresponding to the test local discharge data; the state parameters include at least one of withstand voltage value, insulation resistance value, absorption ratio, elongation at break, and dielectric constant;

[0024] Based on the state parameters corresponding to the test local discharge data, divide the local discharge data into multiple levels according to the preset cable state quality level to obtain the classification data.

[0025] In one embodiment, the step of dividing the local discharge data into multiple levels according to the preset cable state quality level based on the state parameters corresponding to the test local discharge data to obtain the classification data includes:

[0026] Based on the state parameters corresponding to the test local discharge data, use the fuzzy C-means clustering algorithm according to the preset cable state quality level to determine the classification result of each sample in the test local discharge data.

[0027] Using the test partial discharge data as eigenvalue and the grading result of each sample as label value, construct a data feature matrix to obtain the graded data.

[0028] In a second aspect, the present application further provides a partial discharge identification device, which includes:

[0029] A data partitioning module, configured to obtain test partial discharge data and perform a state quality grade partition on the test partial discharge data to obtain graded data;

[0030] A feature extraction module, configured to input the graded data into a target convolutional neural network to obtain extracted deep hidden features, where the target convolutional neural network is obtained by training an initial convolutional neural network based on the graded data until convergence;

[0031] A grading model establishment module, configured to establish a random forest grading model based on the deep hidden features and the original features corresponding to the graded data;

[0032] An identification module, configured to perform a state quality grade identification on the partial discharge data to be identified based on the random forest grading model to obtain an identified state quality grade result.

[0033] In a third aspect, the present application further provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0034] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0035] In a fifth aspect, the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0036] The above-mentioned partial discharge identification method, device, computer equipment, storage medium and computer program product acquire test partial discharge data, and perform state quality level classification on the test partial discharge data to obtain classified data; input the classified data into a target convolutional neural network to obtain extracted deep implicit features, where the target convolutional neural network is obtained by training an initial convolutional neural network based on the classified data until convergence; establish a random forest classification model based on the deep implicit features and the original features corresponding to the classified data; perform state quality level identification on the partial discharge data to be identified based on the random forest classification model to obtain the identified state quality level result. This solution acquires test partial discharge data and performs state quality level classification to obtain classified data. This processing process makes the data more regular and distinguishable, providing a high-quality data basis for subsequent model training. Secondly, the initial convolutional neural network is trained based on the classified data until convergence to obtain the target convolutional neural network. This targeted training method enables the network to deeply mine the feature information in the data, thereby effectively extracting deep implicit features. Finally, a random forest classification model is established based on the deep implicit features and the original features corresponding to the classified data, integrating feature information at different levels, giving full play to the powerful feature extraction ability of the convolutional neural network and the advantages of the random forest in classification. Through the collaborative action of multiple models, accurate state quality level identification of partial discharge data is achieved, effectively improving the accuracy of partial discharge identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0038] Figure 1 It is a schematic flowchart of the partial discharge identification method in an embodiment;

[0039] Figure 2 It is a schematic flowchart of the FCM algorithm used for classifying partial discharge data in an embodiment;

[0040] Figure 3 It is a schematic flowchart of the random forest algorithm in the random forest classification model in an embodiment;

[0041] Figure 4 It is a structural block diagram of the partial discharge identification device in an embodiment;

[0042] Figure 5 It is an internal structure diagram of the computer equipment in an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0044] The partial discharge recognition method provided by the embodiments of the present application can be applied to a variety of devices. In the data acquisition link, high-precision sensor devices are required, such as ultra-high frequency sensors, ultrasonic sensors, etc., to obtain test partial discharge data. These sensors can sensitively capture the electromagnetic signals or ultrasonic signals generated by the partial discharge of the cable. In the data processing and model training stage, it relies on high-performance computers, including ordinary workstations and server clusters with powerful computing capabilities. They have multi-core processors, large-capacity memories and high-speed storage devices, which can meet the computing requirements of massive data and the training of complex neural networks and random forest models. In the actual operation and monitoring scenarios, it can be integrated into intelligent power grid monitoring terminal devices to realize the real-time monitoring of the cable operation status and the recognition of partial discharges.

[0045] The application scenarios provided by the embodiments of the present application are very extensive. In the urban power grid, a large number of cables are buried underground for the transmission of residential and commercial electricity. Through this method, the cable status can be monitored in real time, potential partial discharge faults can be discovered in time, and the stability and reliability of urban power supply can be guaranteed. In the industrial field, the internal power transmission system of factories has a high dependence on cables. This method can be applied to the power monitoring system of factories to discover cable faults in advance, avoid production stagnation caused by power outages, and reduce economic losses. For large power generation stations and substations, their cable connections are complex and carry huge power transmission tasks. Using this method can effectively ensure the safe and stable operation of power equipment and maintain the normal operation of the entire power system.

[0046] In an exemplary embodiment, as Figure 1 shown, a partial discharge recognition method is provided, and this method includes the following steps 101 to step 105:

[0047] 101. Obtain test partial discharge data.

[0048] Among them, the above-mentioned test partial discharge data can be partial discharge data collected during the test.

[0049] In the embodiments of the present application, partial discharge data can be collected by building a test platform. The test platform can include defective cable specimens, oscilloscopes, partial discharge testers, consoles, etc.

[0050] Among them, typical defects such as scratches on the outer surface of the insulation, creepage of the outer semiconductor, air gaps inside the insulation, and contamination on the main surface of the insulation can be set in the defective cable specimen. The above-mentioned outer semiconductor creepage refers to the partial discharge phenomenon that occurs between the outer semiconductor shielding layer and the main insulation layer, which is usually caused by the inductive voltage exceeding the breakdown voltage of the dielectric due to the interface capacitance between the semiconductive layer and the main insulation layer. The main surface of the cable insulation usually refers to the layer of insulation material inside the cable close to the conductor (such as the copper core), and the contamination on the main surface of the insulation means that there is contamination on this layer of insulation material.

[0051] Among them, the above-mentioned partial discharge data may include but are not limited to at least one of the following data:

[0052] The maximum value corresponding to the partial discharge time-domain waveform, the minimum value corresponding to the partial discharge time-domain waveform, the skewness corresponding to the partial discharge time-domain waveform, the kurtosis corresponding to the partial discharge time-domain waveform, the amplitude factor corresponding to the partial discharge time-domain waveform, the waveform factor corresponding to the partial discharge time-domain waveform, the margin factor corresponding to the partial discharge time-domain waveform, the partial discharge quantity, the partial discharge frequency, the partial discharge intensity, and the partial discharge pulse number.

[0053] In some embodiments, after performing the above step 101 and before performing step 102, the test partial discharge data can be preprocessed. Among them, the preprocessing may include but is not limited to validity verification and normalization processing.

[0054] Among them, the validity verification can perform quality control on the collected test partial discharge data. Exemplarily, low-abnormal data with too large or too small numerical values in the test partial discharge data can be eliminated.

[0055] Among them, the normalization of the data is to control the variation range of the eigenvectors corresponding to different dimensions, and avoid overfitting of the network caused by large differences in the data fluctuation ranges.

[0056] The above elimination of abnormal data can avoid its interference with model training and improve the data quality; while the normalization processing can enable eigenvectors with different dimensions to be analyzed on a unified scale, preventing the model training from being unstable or overfitting due to large data fluctuations, thereby improving the training effect and generalization ability of the model.

[0057] 102. Classify the test partial discharge data into different levels of quality to obtain classified data.

[0058] In some embodiments, first, the state parameters corresponding to the test partial discharge data can be obtained; then, based on the state parameters corresponding to the test partial discharge data, the partial discharge data is classified into multiple levels according to the preset cable state quality classification to obtain the classified data.

[0059] Among them, the state parameters include at least one of a withstand voltage value, an insulation resistance value, an absorption ratio, an elongation at break, and a dielectric constant.

[0060] In some embodiments, based on the state parameters corresponding to the test partial discharge data, according to a preset cable state quality level, the Fuzzy C-Means (FCM) algorithm can be used to determine the classification result of each sample in the test partial discharge data; using the test partial discharge data as eigenvalue and the classification result of each sample as the label value, a data feature matrix is constructed to obtain the classified data.

[0061] Exemplarily, the above preset cable state quality levels may include four levels such as excellent, good, medium, and poor. By using the Fuzzy C-Means (FCM) algorithm, classification labels can be constructed.

[0062] The above FCM algorithm is a clustering algorithm that uses a membership function to determine the degree to which each sample (i.e., data point) belongs to a certain cluster. The FCM algorithm allows a sample to belong to multiple cluster centers. FCM represents the degree to which a sample belongs to different clusters by assigning a membership degree of each sample to each cluster center. Exemplarily, Figure 2 FIG. 11 is a schematic flowchart of the FCM algorithm for grading partial discharge data in an embodiment, and the steps of the specific implementation process include but are not limited to the following steps 201 to step 205:

[0063] 201. Select the number C of cluster centers, divide them into C fuzzy clusters (such as four clusters of excellent, good, medium, and poor), and randomly initialize the membership degree of each sample to each cluster center.

[0064] Among them, the cost function of FCM is shown in the following formula (1):

[0065]

[0066] In formula (1), C j represents the cluster center of the j-th fuzzy cluster; u ij represents the membership degree of the j-th sample belonging to the i-th fuzzy cluster, u ij ∈[0,1], d ij represents the Euclidean distance between the i-th data point and the j-th cluster center, d ij =||C j -x i ||; m represents the weighting exponent, which can be a specified value, for example, it can be 2; n represents the number of samples; x i represents the i-th data point.

[0067] Among them, the steps to find the minimum partition of the value function of FCM are as follows:

[0068] 202. Initialize the partition matrix and calculate the centers of all C classes for each step.

[0069] Among them, calculating the centers of all C classes for each step is shown in the following formula (2):

[0070]

[0071] 203. Given the discrimination accuracy, update the partition matrix, calculate the updated partition matrix, and calculate the Euclidean distance between the updated partition matrix and the partition matrix before the update.

[0072] Among them, updating the partition matrix U (k) , is as shown in the following formula (3):

[0073]

[0074] 204. Determine whether the Euclidean distance is less than the given discrimination accuracy.

[0075] The discrimination accuracy ε>0 for convergence can be given in advance. If the Euclidean distance is ||U (k+1) -U (k) ||<ε, stop the iteration and output the vectors of the centers of C clusters and a fuzzy partition matrix; otherwise, perform the next iteration and return to continue executing the above step 203.

[0076] 205. Output the vectors of the centers of C clusters and a fuzzy partition matrix.

[0077] In the case where the Euclidean distance is less than the given discrimination accuracy, output the vectors of the centers of C clusters and a fuzzy partition matrix; in the case where the Euclidean distance is not less than the given discrimination accuracy, return to execute the above step 203.

[0078] For each sample in the test partial discharge data, the output of the FCM algorithm is the vectors of the centers of C clusters (as shown in the above formula (2)) and a fuzzy partition matrix (as shown in the above formula (3)). This fuzzy partition matrix represents the membership degree of each sample belonging to each class. In the embodiments of the present application, according to the fuzzy partition matrix output by the FCM algorithm, the classification of each sample point is determined according to the maximum membership principle. Finally, the test partial discharge data is used as the eigenvalue, and the classification result of each sample point is used as the label value to construct a data feature matrix to obtain the classified data.

[0079] The above-mentioned FCM algorithm allows samples to belong to multiple cluster centers, which can more flexibly and accurately reflect the complex distribution of cable partial discharge data. Compared with traditional clustering algorithms, it can more accurately classify data levels, provide more reasonable classification labels for subsequent model training, and enhance the model's ability to distinguish partial discharge data in different states.

[0080] 103. Input the classified data into the target convolutional neural network to obtain the extracted deep hidden features.

[0081] Among them, the above-mentioned target convolutional neural network is obtained by training the initial convolutional neural network based on the classified data until convergence.

[0082] In some embodiments, the above-mentioned inputting the classified data into the target convolutional neural network to obtain the extracted deep hidden features includes: dividing the classified data into a training set and a test set; inputting the training set into the target convolutional neural network to obtain the deep hidden features, and the target convolutional neural network can be obtained by training the initial convolutional neural network based on the training set until convergence.

[0083] In some embodiments, the classified data can be divided into a training set and a test set. Among them, the target convolutional neural network can be obtained by training the initial convolutional neural network based on the training set until convergence.

[0084] The above-mentioned inputting the classified data into the target convolutional neural network to obtain the extracted deep hidden features can be: inputting the training set into the target convolutional neural network to obtain the extracted deep hidden features.

[0085] Among them, the classified data can be converted into a data feature matrix and then input into the target convolutional neural network to obtain the extracted deep hidden features.

[0086] The form of the above-mentioned data feature matrix is shown in the following formula (4):

[0087]

[0088] Among them, in the above formula (4), P n×m represents the data feature matrix, n represents the number of samples, and m represents different local amplification data, that is, the maximum value corresponding to the partial discharge time-domain waveform, the minimum value corresponding to the partial discharge time-domain waveform, the skewness corresponding to the partial discharge time-domain waveform, the kurtosis corresponding to the partial discharge time-domain waveform, the amplitude factor corresponding to the partial discharge time-domain waveform, the waveform factor corresponding to the partial discharge time-domain waveform, the margin factor corresponding to the partial discharge time-domain waveform, the partial discharge quantity, the partial discharge frequency, the partial discharge intensity, the number of partial discharge pulses, etc.

[0089] In the above preprocessing process, the step of normalizing the test partial discharge data can be executed after obtaining the above data feature matrix, that is, normalizing the data feature matrix.

[0090] In the embodiments of the present application, since the above hierarchical data can be divided into a training set and a test set, the above data feature matrix can also be divided into a data feature matrix corresponding to the training set and a data feature matrix corresponding to the test set.

[0091] In the embodiments of the present application, the above target convolutional neural network includes: an input layer, several convolutional layers, several pooling layers, an activation layer, and a fully connected layer.

[0092] Among them, the convolutional layer is used to perform a convolution operation on the data feature matrix corresponding to the input training set.

[0093] For the discrete one-dimensional sequences {an} and {bm}, the convolution operation of the two can be defined by the following formula (5):

[0094] c n = ∑(a k ·b i-k ) (5)

[0095] Among them, {an} is called the sequence to be convolved, and {bm} is called the convolution kernel.

[0096] The above pooling layer is used to reduce the output dimension of the convolutional layer and prevent overfitting to a certain extent. In the embodiments of the present application, max pooling is adopted. Max pooling selects the maximum value in the local area as the output value, thereby retaining the most significant features. Its calculation process is as follows formula (6):

[0097] f(x) = max(x ij ) (6)

[0098] x ij is the input of the pooling layer, which is the selected local area, and f(x) represents the output result after max pooling.

[0099] After passing through the convolutional layer and the pooling layer, the features extracted by the previous layer are mapped to a one-dimensional vector through the fully connected layer without losing information, and then the final vector is input into the activation layer for operation learning and classification. In the embodiments of the present application, the activation function adopted by the activation layer can be the Softmax function. The activation layer can be expressed as shown in the following formula (7):

[0100] y l = softmax(w l ×x l -l + b l ) (7)

[0101] Among them, x is the input of the fully connected layer; y is the output of the fully connected layer; w is the weight coefficient, b is the bias term, and l is the number of network layers. The above softmax function is also called the normalization function.

[0102] In some embodiments, the convolutional neural network in the embodiments of the present application (i.e., the above-mentioned target convolutional neural network) can select the Adam optimizer. The final value of the training accuracy of Adam is relatively high, and it can reach stability when the number of iterations is small.

[0103] In some embodiments, a residual structure (Res) can be introduced into the convolutional neural network in the embodiments of the present application (i.e., the above-mentioned target convolutional neural network). The residual structure (Res) can enable information to be propagated arbitrarily between the shallow layer and the deep layer, thereby solving the problem of network degradation that may occur in the neural network.

[0104] The relevant formulas of the residual structure (Res) are shown in the following formulas (8) and (9):

[0105] x l+1 = x l + F(x l , W l ) (8)

[0106] In the above formula (8), x l is the input of the l-th convolutional layer, x l+1 is the input of the (l + 1)-th convolutional layer, F represents the convolutional operation, and W l represents the weight of the l-th convolutional kernel. By performing recursive operations on formula (8), the input features in any deep layer L in the neural network can be obtained, that is, as shown in formula 9:

[0107]

[0108] Introducing the residual structure (Res) can ensure that the information in the shallow layer input is transmitted to the deep layer without loss.

[0109] The above residual structure can solve the problem of network degradation in the neural network, enable the model to train deeper levels, learn more complex feature representations, and improve the ability to extract partial discharge data features; while the above Adam optimizer has high training accuracy and fast convergence, can shorten the model training time, improve training efficiency, and at the same time ensure that the model finally reaches a high training accuracy and improves the model performance.

[0110] 104. Establish a random forest classification model based on the original features corresponding to the deep hidden features and hierarchical data.

[0111] In some embodiments, first, the deep implicit features can be combined with the original features corresponding to the hierarchical data to obtain fused features, forming a new feature set; then, independent random sampling is performed on the fused features to obtain multiple sampling sets, and multiple decision trees are trained based on the multiple sampling sets. Finally, the multiple decision trees are combined to obtain the random forest hierarchical model.

[0112] Exemplarily, as Figure 3 shown, it is a schematic flowchart of the random forest algorithm in a random forest hierarchical model. This process may include the following steps 301 to 306:

[0113] 301. Combine the deep implicit features with the original features corresponding to the hierarchical data to obtain fused features, forming a new feature set.

[0114] 302. Perform independent random sampling according to the new feature set to generate multiple decision trees.

[0115] Among them, the Bagging strategy can be adopted, and the bootstrap sampling method can be directly used to obtain different sampling sets. The Bagging strategy is essentially a method of constructing different data sets through multiple sampling with replacement. In this way, the diversity of model training data can be increased, the dependence of the model on specific data can be reduced, thereby reducing the variance of the model and improving the generalization ability of the model, enabling the model to have better performance when facing different data.

[0116] Among them, for each sampling set generated by sampling, when constructing a decision tree, each time a node is split, a part of the features are randomly selected from all the features for splitting. The purpose of this operation is to avoid the decision tree being overly dependent on certain features. In this way, each decision tree can learn to judge from different angles, reducing the dependence on a single feature, increasing the diversity of the model, and making the random forest model combined by multiple decision trees more powerful and accurate.

[0117] 303. Combine multiple decision trees to obtain the random forest hierarchical model.

[0118] After completing the training of multiple sampling sets to obtain multiple decision trees, these decision trees are combined together to form the random forest hierarchical model. Each decision tree is trained based on different sampling sets and feature selections, learning different patterns and rules of the data, and analyzing and judging the data from different angles. The random forest hierarchical model can integrate information from multiple aspects, make up for the limitations of a single decision tree, enabling the model to comprehensively consider various factors when facing complex and diverse partial discharge data, make more comprehensive and accurate judgments, and thus improve the model's ability to identify the quality level of the cable partial discharge state.

[0119] 304. Each decision tree is used for decision-making to obtain the voting results of the decisions of all decision trees.

[0120] When the random forest classification model receives the partial discharge data to be recognized, each decision tree in the model will make an independent judgment on the data according to the knowledge and rules it has learned, and give its own decision result. For example, it judges that the cable is in one of the states of "excellent", "good", "medium", or "poor". Then, the decision results of all decision trees are collected and statistically analyzed by voting. One vote of each decision tree represents its judgment tendency on the category to which the data belongs, and thus the voting results of the decisions of all decision trees are obtained. This method makes full use of the collective wisdom of multiple decision trees and avoids the one-sidedness and uncertainty of a single decision tree.

[0121] 305. Obtain the final cable state recognition result through the voting results.

[0122] After obtaining the voting results of all decision trees, count the number of votes obtained for each category, and the category with the most votes is determined as the final cable state recognition result. This is like an election. Numerous decision trees are voters, and they each cast a vote for what they think is correct. Finally, the "candidate" with the most votes, that is, the category with the most votes, becomes the final determination of the partial discharge state of the cable. This method based on majority voting can effectively integrate the opinions of multiple decision trees, reduce the impact of incorrect judgments of individual decision trees, and make the final recognition result more reliable and stable, providing a basis for the power system to accurately grasp the cable state.

[0123] 306. Evaluate the decision results of the model through the test set.

[0124] The test set is a data set reserved specifically during the previous data processing stage and not involved in model training. Input the partial discharge data in the test set into the already constructed random forest classification model, and the model will give corresponding decision results. Then, compare these decision results with the actual categories (i.e., the true cable states) of the data in the test set, and measure the performance of the model by calculating evaluation indicators such as accuracy, precision, recall, and F1 score. These indicators can reflect the advantages and disadvantages of the model from different perspectives. For example, accuracy reflects the overall correctness of the model classification, precision reflects the accuracy of the model's judgment on a certain category, recall shows the ability of the model to capture positive samples, and F1 score comprehensively considers precision and recall to comprehensively evaluate the comprehensive performance of the model. By evaluating the decision results of the model, the reliability and effectiveness of the model in actual applications can be understood, providing a basis for further optimizing the model.

[0125] Exemplarily, the above-mentioned newly composed feature set can be expressed as shown in the following formula (10):

[0126] F = {f1, f2, f3, ..., f j-1 , f j} (1)

[0127] Among them, f1 to f j-1 represent the original features corresponding to the hierarchical data, and f j represents the deep hidden features.

[0128] The above original features are the basic information directly obtained from the cable partial discharge data, such as the maximum and minimum values of the partial discharge time-domain waveform, etc., which can intuitively display the basic characteristics of the data. The deep hidden features are obtained by deeply mining the hierarchical data through a convolutional neural network, and it contains more abstract and deep-level information of the data. Combining the two is like summarizing the information observed from different angles, which can provide a more comprehensive and rich data basis for subsequent model training, enable the model to learn more complete knowledge, and improve the accuracy of recognition.

[0129] 105. Identify the state quality level of the partial discharge data to be recognized based on the random forest classification model to obtain the recognized state quality level result.

[0130] In some embodiments, before identifying the state quality level of the partial discharge data to be recognized based on the random forest classification model to obtain the recognized state quality level result, the random forest classification model can also be tested based on the test set to obtain evaluation indicators; when the evaluation indicators meet the preset indicator requirements, identify the state quality level of the partial discharge data to be recognized based on the random forest classification model to obtain the recognized state quality level result.

[0131] Among them, the random forest classification model can be tested based on the test set first to obtain evaluation indicators, such as accuracy, precision, recall, F1 score, etc. When the evaluation indicators meet the preset indicator requirements, input the partial discharge data to be recognized into the random forest classification model. The final decision result of the random forest classification model is determined by the majority vote of all decision tree results, and the category with the most votes is the recognized state quality level result.

[0132] Among them, the F1 score comprehensively considers precision and recall, and is an indicator for comprehensively evaluating the performance of a classification model. The value range of the F1 score is between 0 and 1.

[0133] (1) Accuracy is one of the indicators used to evaluate a classification model, referring to the correct rate of the random forest classification model in classifying samples. The calculation formula is:

[0134]

[0135] Among them, TP represents the number of samples that are actually positive examples and are correctly classified as positive examples; FP represents the number of samples that are actually negative examples but are misclassified as positive examples; TN represents the number of samples that are actually negative examples and are correctly classified as negative examples; FN represents the number of samples that are actually positive examples but are misclassified as negative examples.

[0136] The higher the above accuracy rate, the closer the classification result is to the actual result, and the better the performance of the random forest classification model.

[0137] (2) Precision refers to the proportion of the number of samples that truly belong to a certain category among all the samples classified as that category. The calculation formula is:

[0138]

[0139] The above precision measures the accuracy of the random forest classification model when determining a certain category, that is, for the samples predicted as positive examples, how many truly belong to that category.

[0140] (3) Recall rate is the proportion of positive samples predicted as positive samples among the total positive samples. The calculation formula is:

[0141]

[0142] The larger the above recall rate value, the better the performance of the random forest classification model.

[0143] (4) The higher the F1 score value, the better the comprehensive performance. The calculation formula is:

[0144]

[0145] Among them, the F1 score value is related to precision and recall rate.

[0146] Only when the evaluation indicators meet the preset indicator requirements can the random forest classification model be used to identify the state quality level of the local discharge data to be recognized. The preset indicator requirements are set in advance according to the actual application requirements and project standards. When the evaluation indicators of the model reach or exceed these requirements, the reliability and effectiveness of the model in actual applications can be guaranteed. Input the local discharge data to be recognized into the random forest classification model. The final decision result of the random forest classification model is determined by the majority vote of all decision tree results. The category with the most votes is the recognized state quality level result, so as to realize the accurate judgment of the cable local discharge state and provide strong support for the safe and stable operation of the power system.

[0147] The random forest classification model integrates multiple decision trees. These multiple decision trees are trained based on different sampling sets and feature selections, and can fully learn different patterns and rules in the data. The final result is determined by majority voting, which can effectively reduce the error and overfitting risk of a single decision tree, enhance the stability and accuracy of the model, and improve the reliability of identifying the quality level of partial discharge states.

[0148] When testing the random forest classification model and evaluating indicators based on the test set, these evaluation indicators can comprehensively measure the model performance. Meeting the preset indicator requirements is very important for ensuring the accurate judgment of the cable partial discharge state in actual applications. For example, a high accuracy means that the model can correctly classify most samples. The precision and recall rate respectively reflect the accuracy and completeness of the model's judgment of positive example samples from different perspectives. The F1 score comprehensively reflects the comprehensive performance of the model. Only when these indicators meet the standards can it be ensured that the model operates reliably in the power system and the power supply is stable.

[0149] It should be understood that although the steps in the flowcharts involved in the above embodiments are shown in sequence according to the arrows, these steps do not necessarily need to be executed in the order indicated by the arrows. Unless there is a clear description in this article, there is no strict order limit for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages. These steps or stages do not necessarily need to be executed at the same time, but can be executed at different times. The execution order of these steps or stages does not necessarily need to be sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0150] Based on the same inventive concept, the embodiments of the present application also provide a partial discharge identification device for implementing the partial discharge identification method described above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the partial discharge identification device provided below can refer to the limitations on the partial discharge identification method in the above text, and will not be repeated here.

[0151] In an exemplary embodiment, as Figure 4 shown, a partial discharge identification device is provided, including:

[0152] A data partitioning module 401, configured to obtain test partial discharge data and perform a quality level partitioning on the test partial discharge data to obtain partitioned data;

[0153] A feature extraction module 402 for inputting hierarchical data into a target convolutional neural network to obtain extracted deep implicit features, where the target convolutional neural network is obtained by training an initial convolutional neural network based on the hierarchical data until convergence;

[0154] A hierarchical model establishment module 403 for establishing a random forest hierarchical model based on the deep implicit features and the original features corresponding to the hierarchical data;

[0155] An identification module 404 for identifying the state quality level of the local discharge data to be identified based on the random forest hierarchical model to obtain the identified state quality level result.

[0156] In some embodiments, the hierarchical model establishment module 403 is configured to:

[0157] Establish a random forest hierarchical model based on the deep implicit features and the original features corresponding to the hierarchical data, including:

[0158] Combine the deep implicit features and the original features corresponding to the hierarchical data to obtain a fused feature;

[0159] Perform independent random sampling on the fused feature to obtain multiple sampling sets;

[0160] Train based on the multiple sampling sets to obtain multiple decision trees;

[0161] Combine the multiple decision trees to obtain the random forest hierarchical model.

[0162] In some embodiments, the feature extraction module 402 is configured to:

[0163] The step of inputting the hierarchical data into the target convolutional neural network to obtain the extracted deep implicit features includes:

[0164] Divide the hierarchical data into a training set and a test set;

[0165] Input the training set into the target convolutional neural network to obtain the deep implicit features, where the target convolutional neural network is obtained by training an initial convolutional neural network based on the training set until convergence.

[0166] In some embodiments, before the identification module 404 further identifies the state quality level of the local discharge data to be identified based on the random forest hierarchical model to obtain the identified state quality level result, it can also test the random forest hierarchical model based on the test set to obtain an evaluation index; the identification module 404 is specifically configured to:

[0167] Based on the random forest classification model, identify the state quality level of the partial discharge data to be recognized to obtain the recognized state quality level result, including: when the evaluation index meets the preset index requirements, based on the random forest classification model, identify the state quality level of the partial discharge data to be recognized to obtain the recognized state quality level result.

[0168] In some embodiments, the data partitioning module 401 is specifically configured to:

[0169] The partitioning of the test partial discharge data into state quality levels to obtain classified data includes:

[0170] Obtain the state parameters corresponding to the test partial discharge data; the state parameters include at least one of withstand voltage value, insulation resistance value, absorption ratio, elongation at break, and dielectric constant;

[0171] Based on the state parameters corresponding to the test partial discharge data, divide the partial discharge data into multiple levels according to the preset cable state quality levels to obtain the classified data.

[0172] In some embodiments, the data partitioning module 401 is specifically configured to:

[0173] The dividing the partial discharge data corresponding to the test partial discharge data into multiple levels according to the preset cable state quality levels to obtain the classified data includes:

[0174] Based on the state parameters corresponding to the test partial discharge data, according to the preset cable state quality levels, use the fuzzy C-means clustering algorithm to determine the classification result of each sample in the test partial discharge data;

[0175] Take the test partial discharge data as the eigenvalue and the classification result of each sample as the label value to construct a data feature matrix to obtain the classified data.

[0176] Each module in the above partial discharge recognition device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0177] In an exemplary embodiment, a computer device is provided. The computer device can be an intelligent power grid monitoring terminal device, and its internal structure diagram can be as Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a partial discharge identification method.

[0178] Those skilled in the art can understand that Figure 5 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0179] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements the partial discharge identification method described in the above method embodiment.

[0180] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, it implements the partial discharge identification method described in the above method embodiment.

[0181] In an embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, it implements the partial discharge identification method described in the above method embodiment.

[0182] Those of ordinary skill in the art can understand that all or part of the processes in the above-described method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the method embodiments as described above. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0183] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0184] The above-described embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for identifying partial discharge, characterized in that: The method comprises: Acquiring test partial discharge data, and classifying the test partial discharge data into good and bad grades to obtain graded data; Inputting the hierarchical data into a target convolutional neural network to obtain extracted deep implicit features, wherein the target convolutional neural network is obtained by training an initial convolutional neural network based on the hierarchical data until convergence; Establishing a random forest classification model based on the deep hidden features and the original features corresponding to the classification data; The state quality level of the partial discharge data to be identified is identified based on the random forest classification model to obtain an identified state quality level result.

2. The method according to claim 1, characterized in that The establishing of a random forest classification model based on the deep implicit features and the original features corresponding to the classification data comprises: Combining the deep implicit features with the original features corresponding to the hierarchical data to obtain fused features; Performing independent random sampling on the fusion features to obtain multiple sampling sets; Perform training based on the multiple sampling sets to obtain multiple decision trees; The multiple decision trees are combined to obtain the random forest classification model.

3. The method according to claim 1 or 2, characterized in that: The step of inputting the hierarchical data into a target convolutional neural network to obtain extracted deep implicit features comprises: Dividing the classified data into a training set and a test set; The training set is input into the target convolutional neural network to obtain the deep implicit features, and the target convolutional neural network is obtained by training the initial convolutional neural network based on the training set until convergence.

4. The method according to claim 3, characterized in that Before the state quality level identification of the partial discharge data to be identified based on the random forest classification model to obtain the identified state quality level result, the method further includes: Testing the random forest classification model based on the test set to obtain an evaluation index; The identifying the state quality of the partial discharge data to be identified based on the random forest classification model to obtain the identified state quality level result includes: When the evaluation index meets the preset index requirements, the state quality level of the partial discharge data to be identified is identified based on the random forest classification model to obtain an identified state quality level result.

5. The method according to claim 1, characterized in that The step of classifying the test partial discharge data into good or bad states to obtain graded data includes: Acquire state parameters corresponding to the test partial discharge data; the state parameters include: at least one of withstand voltage value, insulation resistance value, absorption ratio, elongation at break, and dielectric constant; Based on the state parameters corresponding to the test partial discharge data, the partial discharge data is divided into multiple levels according to the preset cable state quality levels to obtain the graded data.

6. The method according to claim 5, characterized in that The state parameters corresponding to the test partial discharge data are divided into multiple levels of partial discharge data according to preset cable state quality levels to obtain the graded data, including: Based on the state parameters corresponding to the test partial discharge data, according to the preset cable state quality level, a fuzzy C-means clustering algorithm is used to determine the classification result of each sample in the test partial discharge data; The experimental partial discharge data is used as a feature value, and the classification result of each sample is used as a label value to construct a data feature matrix to obtain the classification data.

7. A partial discharge identification device, characterized in that: The device comprises: A data classification module is used to obtain test partial discharge data and classify the test partial discharge data into good or bad grades to obtain graded data; A feature extraction module, used for inputting the hierarchical data into a target convolutional neural network to obtain extracted deep implicit features, wherein the target convolutional neural network is obtained by training an initial convolutional neural network based on the hierarchical data until convergence; A classification model building module, used to build a random forest classification model based on the deep hidden features and the original features corresponding to the classification data; The identification module is used to identify the state quality of the partial discharge data to be identified based on the random forest classification model to obtain the identified state quality grade result.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

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

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images