Power transmission line discharge voiceprint feature recognition method based on WPE-BA-CNN optimization model

Abstract

The invention discloses a power transmission line discharge voiceprint feature recognition method based on a WPE-BA-CNN optimization model. The method comprises the following steps: S1, carrying out preprocessing and noise reduction on a fault data set; s2, constructing a convolutional neural network (CNN) model; three convolutional layers and a maximum pooling layer are used for feature extraction, and then classification is carried out through a full connection layer to train a model; s3, the filtered signals are imported into a BA-CNN algorithm for voiceprint feature recognition and classification, the BA algorithm is used for CNN convolution kernel number parameter optimization, population parameters of the BA algorithm are initialized, the BA-CNN algorithm adopts a bat algorithm to optimize hyper-parameters of a convolutional neural network, and the number of three layers of convolution kernels serves as position parameters of bat individuals for optimization; according to the power transmission line discharge voiceprint feature recognition method based on the WPE-BA-CNN optimization model, on the basis of adopting the CNN model, wavelet packet transformation is combined to carry out frequency band energy analysis; and optimizing the hyper-parameters of the convolutional neural network by using a bat algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring of power systems, and in particular to a method for recognizing discharge voiceprint features of power transmission lines based on a WPE-BA-CNN optimization model. Background Art

[0002] Transmission line discharge fault refers to the phenomenon that in high-voltage transmission lines, current is released through abnormal paths (such as air or other media) due to aging of insulation materials, external pollution, weather factors or mechanical damage. This phenomenon is common in high-voltage transmission lines and equipment, especially under the influence of insulator aging, humidity changes or other external environmental influences. Discharge voiceprint feature recognition has important practical significance in the operation and maintenance of power systems. The insulation system of transmission lines is usually composed of a variety of materials, including gas, liquid and solid insulators. Over time, these insulating materials will be affected by various factors such as voltage, temperature, humidity and mechanical stress, resulting in their performance degradation. Usually, when the voltage of the sharp edge or protruding part of the high-voltage equipment exceeds a certain critical value, instantaneous discharge occurs in the air or other gases, resulting in partial discharge.

[0003] The mechanism of partial discharge is affected by many factors, such as: (1) Electric field strength: In high electric field areas, sharp edges or protrusions of insulating materials are prone to electrical breakdown. (2) Material defects: Defects such as bubbles, cracks, and impurities can cause partial discharge. (3) Environmental factors: Temperature, humidity, and pollutants can all promote the occurrence of partial discharge.

[0004] Traditional partial discharge detection technology mainly relies on regular manual inspections and simple detection equipment, and has problems such as low detection sensitivity, poor real-time performance, and inability to accurately locate the discharge source. Nowadays, with the development of artificial intelligence technology, intelligent and automated analysis of various sensor data is conducive to effectively reducing labor costs and experience errors. The application of deep learning technology in the recognition of partial discharge soundprint features of transmission lines demonstrates its powerful feature extraction and pattern recognition capabilities. Summary of the invention

[0005] The purpose of the present invention is to provide a method for recognizing the soundprint features of power transmission line discharge based on the WPE-BA-CNN optimization model, so as to recognize the soundprint features of partial discharge of power lines.

[0006] The present invention provides a method for identifying transmission line discharge voiceprint features based on a WPE-BA-CNN optimization model, comprising the following steps: S1, preprocessing and denoising a fault data set;

[0007] Construct a fault data set of partial discharge equipment; perform spectrum analysis and wavelet packet transform on the fault data set; perform spectrum analysis on the fault data set, and use wavelet packet transform to decompose the partial discharge signal into four layers when the fault characteristics of the partial discharge signal are not obvious, decompose the partial discharge signal into frequency bands, and obtain the energy proportion of each frequency band;

[0008] De-noise the fault data set; use the noise measurement index to test the de-noising effect;

[0009] By comparing the energy proportion of the partial discharge signal frequency band without noise and the energy proportion of the partial discharge signal frequency band with wind and rain environmental noise, the energy proportion of the abnormal frequency band is found and the environmental noise interference is filtered out through a bandpass filter to find the fault characteristics of the partial discharge signal;

[0010] The denoising effect is tested using noise measurement indicators;

[0011] S2. Construct a convolutional neural network (CNN) model; use three convolutional layers and a maximum pooling layer as feature extraction, and then perform classification through a fully connected layer to train the model;

[0012] S3. Import the filtered partial discharge signal into the BA-CNN algorithm for voiceprint feature recognition and classification. Use the BA algorithm to optimize the number of convolution kernels of the convolutional neural network CNN and initialize the population parameters of the BA algorithm. The BA-CNN algorithm uses the bat algorithm to optimize the hyperparameters of the convolutional neural network CNN and optimizes the number of three-layer convolution kernels as the position parameters of individual bats.

[0013] Initialize the parameters of the bat population; calculate the fitness of each initial solution through the parameters of each individual bat, and find the solution with the best fitness to judge the quality of the solution; update the position and speed of the bat in each iteration and generate a new frequency for each bat.

[0014] Preferably, in step S1, the fault data is collected by simulating the fault data using a partial discharge generator at a voltage of 20 kV on a partial discharge test bench, and then sampling is performed using a signal acquisition card in conjunction with an acoustic wave sensor, and the environmental noise from different weather conditions is added with noise.

[0015] Preferably, in step S1, the fault data set includes wind environment noise data, rain noise data, wind and rain environment noise data, partial discharge fault data, wind noise partial discharge fault data, rain noise partial discharge fault data and partial discharge fault data under wind and rain noise.

[0016] Preferably, in step S1, the indicators for measuring noise include signal-to-noise ratio, waveform similarity coefficient, mean square error, and root mean square error.

[0017] Preferably, in step S2, the convolutional neural network (CNN) model consists of a convolutional layer, a maximum pooling layer and a fully connected layer.

[0018] Preferably, in step S2, the convolutional neural network (CNN) model parameters are initialized; the convolutional neural network (CNN) model is trained; data fitting is performed through a fully connected layer; the classification result is returned through an output layer; and partial discharge faults are identified.

[0019] Therefore, the present invention adopts the above-mentioned transmission line discharge soundprint feature recognition method based on the WPE-BA-CNN optimization model, high-precision feature extraction: based on the CNN model, frequency band energy analysis is performed in combination with wavelet packet transform; hyperparameter optimization and convergence speed: the hyperparameters of the convolutional neural network are optimized using the bat algorithm. This optimization technology improves the accuracy and convergence speed of the algorithm by simulating the behavior of bats in echolocation. Result visualization and enhanced interpretability: Dimensionality reduction is achieved through principal component analysis, and the classification results of high-dimensional data are visualized. This process not only makes the classification results more intuitive, but also enhances the interpretability of the model.

[0020] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is an overall flow chart of a method for identifying the discharge voiceprint features of a power transmission line based on the WPE-BA-CNN optimization model of the present invention;

[0022] Figure 2 This is an overall structure diagram of a CNN model of a method for recognizing the characteristics of power transmission line discharge voiceprints based on a WPE-BA-CNN optimization model according to the present invention;

[0023] Figure 3 It is a flowchart of a bat algorithm of a method for recognizing the characteristics of power transmission line discharge soundprint based on a WPE-BA-CNN optimization model of the present invention;

[0024] Figure 4 It is a wind noise time domain diagram of a method for identifying the discharge soundprint features of a power transmission line based on the WPE-BA-CNN optimization model of the present invention;

[0025] Figure 5 It is a wind noise spectrum diagram of a method for identifying the discharge soundprint features of a power transmission line based on a WPE-BA-CNN optimization model of the present invention;

[0026] Figure 6 It is a wind noise spectrum diagram of a method for identifying the discharge soundprint features of a power transmission line based on a WPE-BA-CNN optimization model of the present invention;

[0027] Figure 7 It is a rain noise spectrum diagram of a transmission line discharge soundprint feature recognition method based on the WPE-BA-CNN optimization model of the present invention;

[0028] Figure 8 It is a partial discharge time domain diagram of a transmission line discharge soundprint feature recognition method based on the WPE-BA-CNN optimization model of the present invention;

[0029] Fig. 9 It is a partial discharge time domain diagram of a transmission line discharge soundprint feature recognition method based on the WPE-BA-CNN optimization model of the present invention;

[0030] Fig.10 It is a time domain diagram of partial discharge with environmental noise added in a method for recognizing the discharge soundprint features of a power transmission line based on the WPE-BA-CNN optimization model of the present invention;

[0031] Fig.11 It is a partial discharge spectrum diagram with environmental noise added for a method for recognizing the discharge soundprint features of a power transmission line based on the WPE-BA-CNN optimization model of the present invention;

[0032] Fig.12 It is a partial discharge wavelet packet decomposition frequency band energy diagram of a transmission line discharge soundprint feature recognition method based on a WPE-BA-CNN optimization model of the present invention;

[0033] Fig.13 It is a frequency band energy diagram of partial discharge wavelet packet decomposition with ambient noise of a transmission line discharge soundprint feature recognition method based on the WPE-BA-CNN optimization model of the present invention;

[0034] Fig.14 It is a time domain diagram of the signal after noise reduction of a method for identifying the discharge soundprint features of a power transmission line based on the WPE-BA-CNN optimization model of the present invention;

[0035] Fig.15 It is a signal spectrum diagram after noise reduction of a method for identifying the discharge voiceprint features of a power transmission line based on the WPE-BA-CNN optimization model of the present invention;

[0036] Fig.16 It is a training set and test set accuracy curve diagram of a transmission line discharge voiceprint feature recognition method based on the WPE-BA-CNN optimization model of the present invention;

[0037] Fig.17 It is a training set and test set loss curve diagram of a transmission line discharge voiceprint feature recognition method based on the WPE-BA-CNN optimization model of the present invention;

[0038] Fig.18It is a confusion matrix diagram of a method for recognizing the characteristics of power transmission line discharge voiceprint based on the WPE-BA-CNN optimization model of the present invention;

[0039] Fig.19 It is a pre-training visualization scatter plot of a method for recognizing the discharge voiceprint features of a power transmission line based on the WPE-BA-CNN optimization model of the present invention;

[0040] Fig. 20 It is a post-training visualization scatter plot of a method for recognizing the characteristics of power transmission line discharge voiceprints based on the WPE-BA-CNN optimization model of the present invention;

[0041] Fig.21 This is a curve chart showing the accuracy comparison between this model and other deep learning models in a method for recognizing the characteristics of power transmission line discharge voiceprints based on a WPE-BA-CNN optimization model of the present invention;

[0042] Fig. 22 It is a specific schematic diagram of a method for recognizing the discharge voiceprint features of a power transmission line based on the WPE-BA-CNN optimization model of the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0044] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0045] The words "first", "second" and similar terms used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Words such as "include" or "comprises" and similar terms mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Words such as "connect" or "connected" and similar terms are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0046] Embodiment 1

[0047] like Figure 1-Figure 22 As shown, the present invention provides a method for recognizing the characteristics of power transmission line discharge voiceprint based on the WPE-BA-CNN optimization model, comprising the following steps:

[0048] S1, preprocessing and denoising the fault data set;

[0049] Construct a fault data set of partial discharge equipment; the fault data set includes wind environment noise data, rain noise data, wind and rain environment noise data, partial discharge fault data, wind noise partial discharge fault data, rain noise partial discharge fault data and partial discharge fault data under wind and rain noise;

[0050] The experimental data uses a partial discharge test bench, a partial discharge generator is used at a voltage of 20kV to simulate fault data, and a signal acquisition card is used with an acoustic sensor for sampling. The sampling frequency is 48000Hz, and the environmental noise from different weather conditions is added with noise. Each sample is labeled, and each sample is divided into 400 data points, with 600 groups of samples for each category. The data set type and sample number are shown in Table 1.

[0051] Table 1 Data types and sample sizes

[0052]

[0053] Perform spectral analysis and wavelet packet transform on fault data sets;

[0054] The fault data set is analyzed by spectrum analysis. When the fault characteristics of the local discharge signal are not obvious, the partial discharge signal is decomposed into four layers using wavelet packet transform. The partial discharge signal is decomposed into frequency bands to obtain the energy proportion of each frequency band.

[0055] The time domain diagram of outdoor environmental noise and wind noise collected by experimental equipment in different weather conditions is as follows: Figure 4 As shown, the frequency domain diagram is Figure 5 As shown in Figure 1, the frequency distribution of wind noise is in the 0-3kHz frequency band, concentrated around 640Hz. Figure 6 As shown, the frequency domain diagram is Figure 7 As shown in Figure 1, the frequency distribution of rain noise is between 0 and 4 kHz, with the frequency concentrated around 3 kHz. The time domain diagram of the noise-free partial discharge signal is shown in Figure 1. Figure 8 As shown, its spectrum is as follows Fig. 9 As shown, it is found that the frequency distribution of partial discharge signals is between 5k-12kHz. The time domain diagram of partial discharge characteristics under the background noise of transmission lines in outdoor environments is shown in Fig.10 As shown in the figure, the time domain characteristics of partial discharge are manifested as periodic impulse signals, and the time domain characteristics of partial discharge faults are submerged under environmental noise.

[0056] The spectrum diagram is as follows Fig.11 As shown in the figure, the energy is concentrated in the low-frequency area, and the partial discharge fault characteristics are no longer obvious. The signal is decomposed into four layers using wavelet packet transform to obtain the energy proportion of each frequency band. The energy proportion of the pure partial discharge signal frequency band is Fig.12 As shown in the figure, the energy of the frequency band of the partial discharge acoustic signal containing wind and rain environmental noise accounts for Fig.13As shown, it can be seen that the energy of the first three frequency bands is very large. The environmental noise interference of the first three frequency bands is filtered out using a bandpass filter. The denoised time domain and frequency domain diagrams are as follows Fig.14 , 15 As shown, the time domain diagram and frequency domain diagram show the fault characteristics of partial discharge.

[0057] De-noise the fault data set; compare the energy ratio of the frequency band of the partial discharge signal without noise with the energy ratio of the frequency band of the partial discharge signal with wind and rain environmental noise, find out the energy ratio of the abnormal frequency band and filter out the environmental noise interference through a bandpass filter to find the fault characteristics of the partial discharge signal; use the index of measuring noise to test the denoising effect; use the index of measuring noise to test the denoising effect, such as signal-to-noise ratio, waveform similarity coefficient, mean square error, and root mean square error.

[0058] The denoising effect is shown in Table 3. It can be seen that the signal-to-noise ratio and waveform similarity coefficient of the signal have been improved, indicating that the proportion of environmental noise in the denoised signal has been reduced, and it is more similar to the pure signal characteristics of partial discharge. Data denoising and analysis are calculated using Matlab.

[0059] Indicators for measuring noise include signal-to-noise ratio, waveform similarity coefficient, mean square error, and root mean square error;

[0060] Table 2 Noise index comparison

[0061]

[0062] S2, build a convolutional neural network CNN model;

[0063] Table 3 CNN model parameter settings

[0064]

[0065] Note: cd1, cd2, cd3 are the optimal parameters after BA algorithm iteration

[0066] Initialize the convolutional neural network (CNN) model parameters; train the convolutional neural network (CNN) model; perform data fitting through the fully connected layer; and return the classification results through the output layer.

[0067] Partial discharge fault identification;

[0068] The optimized parameter is the number of convolution kernels of three layers of convolution. The convolutional neural network consists of convolution layer, maximum pooling layer and fully connected layer. The convolution layer can extract the features of the data to improve the accuracy of the model. The maximum pooling layer can extract the data features of the window and significantly reduce the amount of data, thereby accelerating the training of the model. The fully connected layer plays a classification role in the model, thereby identifying whether the data has faults. The structure of the CNN algorithm is as follows: First, a convolution kernel of size 16×16 is used for extraction, followed by a maximum pooling layer with a kernel size of (3×3) and a strides of 2. The second convolution uses a convolution kernel of size 7×7 for extraction, followed by a maximum pooling layer with a kernel size of (3×3) and a strides of 2. The third convolution uses a convolution kernel of size 3×3 for extraction, followed by a maximum pooling layer with a kernel size of (3×3) and a strides of 2. After the convolutional layer is extracted, the fully connected layer is classified. The fully connected layer has 64 input nodes, 7 output nodes, and a classifier with Softmax as the activation function.

[0069] S3. Import the filtered partial discharge signal into the BA-CNN algorithm for voiceprint feature recognition and classification. Use the BA algorithm to optimize the number of convolution kernel parameters of the convolution neural network CNN and initialize the population parameters of the BA algorithm. The algorithm uses the bat algorithm to optimize the hyperparameters of the convolution neural network. The number of three-layer convolution kernels is used as the position parameter of the individual bat to optimize the number of three-layer convolution kernels. The fitness evaluation function of the BA algorithm selects the error rate of the convolution neural network CNN model after training (1 minus Fig.15 The accuracy rate in the model is calculated by continuous small batch iteration to find the optimal number of CNN convolution kernel parameters as the specific structural parameters of the final model. Finally, the sample data is imported into the final CNN model to return the output result and complete the discharge voiceprint feature recognition.

[0070] Initialize the parameters of the bat population; calculate the fitness of each initial solution through the parameters of each individual bat, and find the solution with the best fitness to judge the quality of the solution; update the position and speed of the bat in each iteration, and generate a new frequency for each bat;

[0071] If the random frequency is greater than the pulse rate, the bat will generate a new solution around the current best solution. If the fitness of the new solution is better than the current solution and the random number is less than the loudness, the new solution is accepted. The new solution is evaluated for fitness. If the fitness of the new solution is better than the current best solution, the current best solution and the minimum fitness value are updated; when the maximum number of iterations is reached, the best solution found and the corresponding minimum fitness value are returned.

[0072] The steps of the bat algorithm are as follows Figure 3As shown. First, the parameters of the population, such as position, speed, frequency, loudness, and population size, must be initialized. The first position value is a random value, and the initial solution (population) is randomly generated using a uniform distribution to ensure that the solution is within the parameter boundary. The fitness of each initial solution is calculated for the parameters of each individual bat, and the solution with the best fitness (that is, the solution with the smallest objective function value, the objective function is the training error rate of the CNN model) is found to judge the quality of the solution. The position and speed of the bat are updated in each iteration, and a new frequency is generated for each bat. If the random number is greater than the pulse rate, the bat will generate a new solution around the current best solution. If the fitness of the new solution is better than the current solution, and the random number is less than the loudness, the new solution is accepted. The fitness of the new solution is evaluated. If the fitness of the new solution is better than the current best solution, the current best solution and the minimum fitness value are updated. When the maximum number of iterations is reached, the best solution found and the corresponding minimum fitness value are returned.

[0073] The position and velocity update formula of the BA algorithm is as follows:

[0074] f i =f min +(f max -f min )·rand;

[0075]

[0076]

[0077] In the formula, f i is the frequency, rand is the random value, v i is the speed of the bat, x * is the current optimal solution, x i It's the bat's location.

[0078] Table 4 BA initialization parameter settings

[0079] Parameter upper bound Parameter lower bound Iterations Population size Parameter Dimensions frequency Loudness (100,100,100) (1,1,1) 3 2 3 0.5 0.5

[0080] Training test curve and confusion matrix:

[0081] The model training accuracy curve is as follows Fig.16 As shown, the loss curve is Fig.17 As shown, the confusion matrix is Fig.18 As shown in the experimental results, the training set accuracy of the model algorithm of the present invention has basically converged after 5 iterations, with an accuracy rate of 99.60%, and the test set accuracy rate has reached 98.8%, which has basically converged after 20 iterations. From the confusion matrix, it can be seen that the accurate number of sample recognition for each category has reached a very high level.

[0082] Visualization of dimensionality reduction of the overall classification process:

[0083] Fig.19 , Fig. 20 The data distribution before model training and the PCA dimension reduction visualization after model training are shown in the figure. From the comparison of the two figures, we can see that the data distribution before model training is mixed, and the sample data are stacked together, making it difficult to classify effectively. However, the data distribution after model classification is very scattered and distributed according to the label type. The visualization effect is very good, which shows the excellent performance of the BA-CNN algorithm for partial discharge voiceprint feature recognition and classification.

[0084] Comparison of test accuracy and loss of different voiceprint feature recognition methods:

[0085] In order to ensure the performance of the algorithm in the present invention, other algorithm results are compared with it, such as CNN, AlexNet, and LeNet5. The same number of iterations is set to 50 during the training process, and the accuracy of the training results is visually compared. Fig.21 As shown in the figure, the BA-CNN algorithm has a higher convergence speed and accuracy. Among the other three methods, the LeNet5 algorithm has a lower accuracy. The accuracy of the CNN and AlexNet algorithms is slightly lower than that of the BA-CNN algorithm, but the convergence speed during iteration is not as good as that of the BA-CNN algorithm. Therefore, the BA-CNN algorithm has better overall performance.

[0086] Therefore, the present invention adopts the above-mentioned transmission line discharge soundprint feature recognition method based on the WPE-BA-CNN optimization model, high-precision feature extraction: based on the CNN model, frequency band energy analysis is performed in combination with wavelet packet transform; hyperparameter optimization and convergence speed: the hyperparameters of the convolutional neural network are optimized using the bat algorithm. This optimization technology improves the accuracy and convergence speed of the algorithm by simulating the behavior of bats in echolocation. Result visualization and enhanced interpretability: Dimensionality reduction is achieved through principal component analysis, and the classification results of high-dimensional data are visualized. This process not only makes the classification results more intuitive, but also enhances the interpretability of the model.

[0087] The above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can still be modified or replaced by equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for identifying transmission line discharge voiceprint features based on the WPE-BA-CNN optimization model, characterized in that: The following steps are involved: S1. Preprocess and reduce noise of fault data set; construct fault data set of partial discharge equipment; Perform spectrum analysis and wavelet packet transform on the fault data set; perform spectrum analysis on the fault data set, and use wavelet packet transform to decompose the partial discharge signal into four layers when the fault characteristics of the partial discharge signal are not obvious, decompose the partial discharge signal into frequency bands, and obtain the energy proportion of each frequency band; De-noise the fault data set; use noise measurement indicators to test the denoising effect; compare the energy proportion of the partial discharge signal frequency band without noise with the energy proportion of the partial discharge signal frequency band with wind and rain environmental noise, find out the energy proportion of the abnormal frequency band and filter out the environmental noise interference through a bandpass filter to find the fault characteristics of the partial discharge signal; The denoising effect is tested using noise measurement indicators; S2. Construct a convolutional neural network (CNN) model; use three convolutional layers and a maximum pooling layer as feature extraction, and then perform classification through a fully connected layer to train the model; S3. Import the filtered partial discharge signal into the BA-CNN algorithm for voiceprint feature recognition and classification. Use the BA algorithm to optimize the number of convolution kernels of the convolutional neural network CNN and initialize the population parameters of the BA algorithm. The BA-CNN algorithm uses the bat algorithm to optimize the hyperparameters of the convolutional neural network CNN and optimizes the number of three-layer convolution kernels as the position parameters of individual bats. Initialize the parameters of the bat population; calculate the fitness of each initial solution through the parameters of each individual bat, and find the solution with the best fitness; To judge the quality of the solution; The bat's position and velocity are updated in each iteration, generating a new frequency for each bat.

2. According to claim 1, a method for identifying the discharge voiceprint features of a power transmission line based on the WPE-BA-CNN optimization model is characterized in that: In step S1, the fault data is collected by simulating the fault data using a partial discharge generator at a voltage of 20 kV on a partial discharge test bench, and then sampled by a signal acquisition card in conjunction with an acoustic wave sensor, and the environmental noise from different weather conditions is added.

3. According to claim 1, a method for identifying the discharge voiceprint features of a power transmission line based on the WPE-BA-CNN optimization model is characterized in that: In step S1, the fault data set includes wind environment noise data, rain noise data, wind and rain environment noise data, partial discharge fault data, wind noise partial discharge fault data, rain noise partial discharge fault data and partial discharge fault data under wind and rain noise.

4. According to claim 1, a method for identifying the discharge voiceprint features of a power transmission line based on the WPE-BA-CNN optimization model is characterized in that: In step S1, the indicators for measuring noise include signal-to-noise ratio, waveform similarity coefficient, mean square error, and root mean square error.

5. According to claim 1, a method for identifying the discharge voiceprint features of a power transmission line based on the WPE-BA-CNN optimization model is characterized in that: In step S2, the convolutional neural network (CNN) model consists of a convolutional layer, a maximum pooling layer, and a fully connected layer.

6. According to claim 1, a method for identifying power line discharge voiceprint features based on the WPE-BA-CNN optimization model is characterized by: In step S2, the convolutional neural network (CNN) model parameters are initialized; the convolutional neural network (CNN) model is trained; data fitting is performed through a fully connected layer; the classification result is returned through an output layer; and partial discharge faults are identified.

Images