VMD-GWO-CNN-based overhead high-voltage transmission line discharge signal identification method

Abstract

The invention discloses an overhead high-voltage transmission line discharge signal identification method based on VMD-GWO-CNN. The method comprises the following steps: S1, collecting partial discharge signals not containing noise and partial discharge signals containing noise; s2, carrying out variational mode decomposition (VMD) on the partial discharge signal without noise; s3, carrying out variational mode decomposition (VMD) on the partial discharge signal containing the noise; s4, reading a partial discharge signal data set of various noise types; s5, building a convolutional neural network CNN model; S6, performing parameter optimization on the number of convolution kernels of a convolution layer in the convolutional neural network CNN model by using a grey wolf optimization algorithm GWO; s7, importing the preprocessed training set and test set into a parameter-optimized convolutional neural network (CNN) model; and S8, performing PCA (principal component analysis). According to the overhead high-voltage transmission line discharge signal identification method based on the VMD-GWO-CNN, noise reduction of the discharge signal is completed through VMD variational mode decomposition; and a grey wolf optimization algorithm GWO is used to optimize the hyper-parameters of the CNN.

Description

Technical Field

[0001] The invention relates to the technical field of overhead high-voltage transmission line discharge signal recognition, and in particular to an overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN. Background Art

[0002] With the development of science and technology and the increase in electricity demand, overhead high-voltage transmission lines are playing an increasingly important role in urban and rural power supply systems. In power systems, overhead high-voltage transmission lines are important facilities for transmitting electric energy. However, due to environmental factors, material aging, and improper operation, these lines may experience discharge during operation. Partial discharge (PD) refers to a small-scale electrical discharge phenomenon inside or on the surface of insulating materials. It usually indicates aging or damage to insulating materials, which may lead to line failures, power outages, and serious safety hazards. Partial discharge signals have multiple characteristics, including frequency, amplitude, duration, and waveform. Due to the complexity of partial discharge, a single voiceprint feature may not be sufficient to accurately identify different types of discharge sources.

[0003] On high-voltage transmission lines, discharge signals are usually collected through sensors (such as current sensors and voltage sensors). These acoustic signals are usually affected by noise and interference, so effective signal processing technology is required to extract useful information. Traditional discharge signal recognition methods include threshold method, fuzzy logic, statistical methods, etc. These methods are more effective when processing simple signals, but when faced with complex and changeable discharge signals, there are often problems with low recognition rate and insufficient accuracy. Nowadays, with the improvement of the level of intelligence, the acquisition speed and order of magnitude of various types of data have been significantly improved, which has laid a good foundation for the widespread application of deep learning methods in the field of partial discharge signal recognition.

[0004] As a widely used intelligent research method, deep learning can automatically extract and learn complex features in acoustic signals without manually designing features, which significantly improves the accuracy and efficiency of detection. On this basis, it can also process large amounts of data, has strong adaptability, and is suitable for applications in real-time monitoring and dynamic environments. In addition, deep learning also performs very well in pattern recognition and classification, and can better identify different types of partial discharge signals, thereby improving fault warning capabilities and reducing false alarm rates. These advantages make deep learning a more effective and advanced technical means in the field of partial discharge signal recognition. Therefore, it is gradually being applied by more and more researchers in the field of partial discharge signal recognition. Summary of the invention

[0005] The purpose of the present invention is to provide a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN, noise reduction, identification and noise classification of discharge signals of overhead high-voltage transmission lines under noise interference conditions.

[0006] The present invention provides a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN, comprising the following steps:

[0007] S1, collecting partial discharge signals without noise and with noise;

[0008] S2, performing variational mode decomposition (VMD) on the noise-free partial discharge signal;

[0009] S3, performing variational mode decomposition (VMD) on the noisy partial discharge signal; reconstructing the IMF and residual of the modal component dominated by the screened partial discharge signal;

[0010] S4, reading partial discharge signal data sets of various noise types, creating labels for the partial discharge signal data sets and converting the labels into one-hot encoding; dividing the partial discharge signal data sets into a training set and a test set and performing normalization processing;

[0011] S5. Convolutional neural network CNN model construction:

[0012] S6. Use the Grey Wolf Optimization Algorithm (GWO) to optimize the number of convolution kernels in the convolutional layer of the convolutional neural network (CNN) model.

[0013] S7, importing the preprocessed training set and test set into the convolutional neural network (CNN) model after parameter optimization and performing training and pattern recognition on the convolutional neural network (CNN) model;

[0014] S8. Use principal component analysis (PCA) and matplotlib to visualize the results of convolutional neural network (CNN) model pattern recognition.

[0015] Preferably, in step S1, the sound sensor and the signal acquisition card are placed on a discharge test bench to collect 20 kV local discharge signals at a sampling frequency of 48000 Hz, and a local discharge signal containing noise is obtained by playing wind, rain and other noises recorded in advance.

[0016] Preferably, in step S2, the decomposition of the modal component IMFs at different modal decomposition levels is calculated, and the center frequency of each modal component IMF is obtained; the optimal modal decomposition level is selected and the center frequency of the corresponding modal component IMF is recorded;

[0017] Preferably, in step S3, a modal component IMF closest to the center frequency of the optimal modal decomposition layer is obtained, and this modal component IMF is used as a threshold to filter the IMF components dominated by the partial discharge signal, and the remaining modal component IMFs are regarded as noise-dominated IMF components.

[0018] Preferably, in step S3, the obtained reconstructed signal is used as the local discharge signal after noise reduction, and the reconstructed signal and various local discharge signals containing different noises are combined into a local discharge signal data set.

[0019] Preferably, in step S4, local discharge signal data of multiple noise types are read, 240,000 data points are intercepted for the discharge signal data of different noise types, and the intercepted data of different noise types are divided into 600 samples according to the standard of 400 data points for a single sample; labels are created for the local discharge signal data set and converted into unique hot encodings, and each type of sample is divided into a training set and a test set in a ratio of 7:3, and normalized.

[0020] Preferably, in step S5, the convolutional neural network (CNN) model includes an input layer, a convolution layer, a normalization layer, a maximum pooling layer, a fully connected layer and an output layer.

[0021] Therefore, the present invention adopts the above-mentioned overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN. The variational mode decomposition VMD can effectively decompose complex signals into multiple modes with different frequency characteristics, and the denoising of the discharge signal is completed through VMD variational mode decomposition; the hyperparameters of CNN are optimized using the gray wolf optimization algorithm GWO; and principal component analysis PCA is used to visualize the results of pattern recognition.

[0022] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is an overall flow chart of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0024] Figure 2 A grey wolf level graph of a grey wolf optimization algorithm (GWO) of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to the present invention;

[0025] Figure 3 It is a schematic diagram of the gray wolf position update of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0026] Figure 4A GWO optimization algorithm flow chart of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to the present invention;

[0027] Figure 5 This is a time domain diagram of a noise-free partial discharge signal of a discharge signal recognition method for overhead high-voltage transmission lines based on VMD-GWO-CNN in the present invention.

[0028] Figure 6 It is a time domain diagram of a noise-free partial discharge signal of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0029] Figure 7 It is a time domain diagram of the mode decomposition of the noise-free discharge signal when K=3 of the overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN of the present invention;

[0030] Figure 8 This is a modal decomposition spectrum diagram of a noise-free discharge signal when K=3 of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0031] Fig. 9 It is an energy spectrum of the IMF component of the noise-free discharge signal when K=3 of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0032] Fig.10 It is a time domain diagram of the mode decomposition of the noise-free discharge signal when K=4 of the overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN of the present invention;

[0033] Fig.11 It is a modal decomposition spectrum diagram of a noise-free discharge signal when K=4 of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0034] Fig.12 It is an energy spectrum of the IMF component of the noise-free discharge signal when K=4 of a discharge signal recognition method for overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0035] Fig.13 A time domain diagram of wind and rain sound of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0036] Fig.14 A wind and rain sound spectrum diagram of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0037] Fig.15It is a time domain diagram of a discharge signal containing noise of a mixed wind and rain sound of a method for identifying discharge signals of an overhead high-voltage transmission line based on VMD-GWO-CNN of the present invention;

[0038] Fig.16 A spectrum diagram of a discharge signal containing noise of a mixed wind and rain sound of a method for identifying discharge signals of an overhead high-voltage transmission line based on VMD-GWO-CNN of the present invention;

[0039] Fig.17 It is a time domain diagram of the mode decomposition of the discharge signal containing noise when K=7 of the discharge signal recognition method of the overhead high-voltage transmission line based on VMD-GWO-CNN of the present invention;

[0040] Fig.18 It is a mode decomposition spectrum diagram of a discharge signal containing noise when K=7 of a discharge signal recognition method for overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0041] Fig.19 It is a time domain diagram of the reconstructed signal after noise reduction of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0042] Fig. 20 It is a reconstructed signal spectrum diagram after noise reduction of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0043] Fig.21 The overall structure of the GWO-CNN model of the overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN of the present invention;

[0044] Fig. 22 It is an iterative convergence curve of a GWO optimization algorithm of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN of the present invention;

[0045] Fig.23 It is a graph of the accuracy of noise-containing pattern recognition of the GWO-CNN model of the GWO optimization algorithm of the overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN of the present invention;

[0046] Fig.24 A loss diagram of noise-type pattern recognition of a GWO-CNN model of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to the present invention;

[0047] Fig.25A confusion matrix of noise type pattern recognition of a GWO-CNN model of a GWO optimization algorithm of a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN in the present invention;

[0048] Fig.26 It is a scatter plot of the GWO-CNN model of the overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN of the present invention before the noise type pattern recognition;

[0049] Fig. 27 It is a scatter plot of the noise type pattern recognition of the GWO-CNN model of the overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN of the present invention;

[0050] Fig.28 This is a comparison chart of the accuracy of the GWO-CNN model, CNN, Alex Net, and LSTM models under the same data set in a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN in the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0052] 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.

[0053] 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.

[0054] Embodiment 1

[0055] like Figure 1-Figure 28 As shown, the present invention provides a method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN, comprising the following steps:

[0056] S1. Collect the noise-free and noise-containing partial discharge signals:

[0057] The sound sensor and signal acquisition card are placed on the discharge test bench, collecting 20kv partial discharge signals at a sampling frequency of 48000Hz, and obtaining partial discharge signals containing noise by playing pre-recorded wind, rain and other noises;

[0058] S2. Use the VMD function in MATLAB to export the time domain diagram and spectrum diagram of the noise-free discharge signal, perform variational modal decomposition VMD on the noise-free partial discharge signal, and calculate the decomposition of the modal component IMFs at different modal decomposition levels when the modal decomposition level is K=3, and find the center frequency of each modal component IMF; select the optimal modal decomposition level and record the center frequency of the corresponding modal component IMF;

[0059] And generate modal component time domain diagram, spectrum diagram and energy spectrum such as Figure 7 , Figure 8 , Fig. 9 . The data graphs when K=4 are as follows Fig.10 , Fig.11 , Fig.12 The comparison shows that when K=3, the energy spectrum has a relatively uniform distribution, and when the value of K increases to 4, such as Fig.11 As shown in , there is some frequency band overlap between IMF1 and IMF2, and there may be over-decomposition. Therefore, K=3 can be set as the optimal modal decomposition layer number, and its center frequency is used as the condition for the next filtering.

[0060] Table 1 Center frequency of different K values

[0061]

[0062] Then the time domain diagram and spectrum diagram of wind and rain sound and discharge with noise are derived as follows: Fig.13 , Fig.14 , Fig.15 , Fig.16 As shown in the figure, it can be seen that noise signals such as wind and rain are mainly concentrated in the low frequency band of 0-5kHz. Fig.16 As shown in Figure 1, in the case of mixed noise, the characteristics of the discharge signal are masked by a large amount of noise at low frequencies. In this case, noise reduction processing is particularly important.

[0063] S3, performing variational mode decomposition VMD on the noisy partial discharge signal; obtaining the modal component IMF closest to the center frequency of the optimal modal decomposition layer number, and using this modal component IMF as a threshold to filter the IMF component dominated by the partial discharge signal, and the remaining modal component IMFs are regarded as noise-dominated IMF components;

[0064] When K=7, as shown in Table 2, there are three IMF components, IMF5, IMF6, and IMF7, whose center frequencies are basically the same as the center frequencies of the modal components of the discharge signal, and their frequency bands overlap. They can be used as IMF components dominated by the discharge signal, and the remaining components are filtered out as IMF components dominated by noise.

[0065] Table 2 Center frequency of each IMF component

[0066]

[0067] The IMF and residual of the modal component dominated by the selected partial discharge signal are reconstructed, and the reconstructed signal is used as the denoised partial discharge signal. The reconstructed signal and various partial discharge signals containing different noises are combined into a partial discharge signal data set.

[0068] The time domain diagram and spectrum diagram of the reconstructed signal are as follows Fig.19 , Fig. 20 As shown in Table 3, the noise evaluation index shows that the noise reduction effect is obvious and the discharge signal characteristics are more obvious.

[0069] Table 3 Noise evaluation index before and after noise reduction

[0070]

[0071] S4, reading partial discharge signal data sets of various noise types, creating labels for the partial discharge signal data sets and converting the labels into one-hot encoding; dividing the partial discharge signal data sets into a training set and a test set and performing normalization processing;

[0072] The local discharge signal data of various noise types are read. 240,000 data points are intercepted for the discharge signal data of different noise types, and the intercepted data of different noise types are divided into 600 samples according to the standard of 400 data points for a single sample. Labels are created for the local discharge signal data set and converted into unique hot encoding. Each type of sample is divided into training set and test set in a ratio of 7:3, and normalized.

[0073] S5. Convolutional neural network CNN model construction:

[0074] The convolutional neural network CNN model includes an input layer, a convolutional layer, a normalization layer, a maximum pooling layer, a fully connected layer, and an output layer;

[0075] The model structure is as follows Fig.21As shown. The input layer dimension is 20×20×1. The convolution kernel size of the first convolution layer is 16×16, and the convolution kernel sizes of the second, third, and fourth convolution layers are 8×8, 8×8, and 4×4, respectively. The number of convolution kernels uses variables nr1, nr2, nr3, and nr4. The given values ​​must be optimized using the Grey Wolf Optimization Algorithm GWO algorithm, and the initial values ​​are randomly generated. Each convolution layer is followed by a normalization layer and a pooling layer. The pooling layer window size is unified to 2×2, with a stride of 2, which reduces the size of the feature map, reduces computational complexity, and extracts the main features. The flattening layer flattens the multi-dimensional feature map into a one-dimensional vector in preparation for the fully connected layer. The number of neurons in the fully connected layer is 64. The final output layer contains 7 neurons, corresponding to 7 classification tasks.

[0076] Table 4 GWO-CNN model parameter settings

[0077]

[0078] Note: nr1, nr2, nr3, and nr4 are hyperparameters, representing the number of convolution kernels. The optimal parameter combination is obtained after GWO optimization.

[0079] Convolutional neural network (CNN) is a type of deep learning model, mainly used in image processing, computer vision and natural language processing. CNN performs well in processing grid-structured data. Its structure can effectively capture local features and recognize complex patterns through multiple levels of abstraction. The typical CNN structure consists of input layer, convolution layer, pooling layer, fully connected layer and output layer.

[0080] The function of the input layer is to receive raw data from the neural network.

[0081] The convolution layer uses multiple convolution kernels, and these filters slide on the input data to generate feature maps. And convolution operations are performed on the input signal to extract important features. Its expression is as follows:

[0082]

[0083] Where x i is the current input feature matrix, x i+1 is the calculated feature matrix, ω i is the convolution kernel weight parameter, b i is the bias parameter, is the convolution operation.

[0084] The pooling layer downsamples the feature information, reducing the spatial dimension of the feature while retaining important information. There are two common pooling operations in convolutional neural networks, namely maximum pooling and average pooling. The present invention uses a maximum pooling layer to perform a maximum pooling operation on the feature information, taking the maximum value within the pooling window, reducing the data dimension while reducing the parameters; and selecting the largest statistic to obtain the local features of the area.

[0085] The fully connected layer is responsible for integrating the features and performing final classification or regression.

[0086] The output layer is located after the fully connected layer and generates the final result according to the task requirements.

[0087] S6. Use the Grey Wolf Optimization Algorithm (GWO) to optimize the number of convolution kernels in the convolutional layer of the convolutional neural network (CNN) model.

[0088] Set the number of iterations of the GWO optimization algorithm to 5, the search range of gray wolves to 4, and the number of gray wolves to 5. The upper bounds of the four hyperparameters are set to (20, 300, 300, 300), and the lower bounds are set to (1, 1, 1, 1). The algorithm steps are as follows:

[0089] First, set the initial number range of convolution kernels and define the fitness function. Randomly generate a group of gray wolves, representing different combinations of the number of convolution kernels. Then use the number of convolution kernels of each gray wolf to build a CNN model, and evaluate its fitness through training. According to the rules of the GWO algorithm, use the information of the best gray wolf to update the positions of other gray wolves to find a better solution. Repeat the evaluation and update steps until the set number of iterations is reached or the fitness is no longer significantly improved. Select the combination of the number of convolution kernels with the highest fitness, apply it to the CNN model, and update the parameters to improve the performance and accuracy of the CNN model.

[0090] The GWO algorithm simulates the leadership hierarchy and hunting mechanism of gray wolves in nature. Gray wolves are divided into four types: α, β, δ, and ω to simulate the hierarchy. Figure 2 shown.

[0091] First layer: α-layer wolf pack. Wolf pack leader - the optimal solution of the optimization algorithm.

[0092] The second layer: the β-layer wolf pack. The wolf pack's sub-leader assists the α-layer wolf pack - the sub-optimal solution of the optimization algorithm.

[0093] The third layer: δ wolf pack. Obey the decisions of α and β - α and β with poor fitness will be reduced to δ.

[0094] The fourth layer: the ω-layer wolf pack, which is responsible for executing the decisions of the superiors - updating the positions around α, β and δ.

[0095] In addition, three main stages of searching for prey, surrounding prey and attacking prey were simulated.

[0096] Surrounding prey: The gray wolf gradually reduces the distance between itself and its prey by keeping a certain distance from it. It dynamically updates its position based on the positions of α, β, and δ.

[0097] Tracking prey: When prey is spotted, the wolves adjust their position based on the prey's location. This process includes both random and deterministic updates to the prey's location.

[0098] Hunting strategy: Gray wolves will constantly adjust their movement strategies while chasing prey, and mathematical models are used to simulate the process of gray wolves encircling and hunting prey.

[0099] The overall process of the algorithm is as follows Figure 4 As shown, a certain number of gray wolf individuals are randomly generated and their fitness is evaluated according to the objective function. The gray wolves are sorted according to their fitness and α, β, and δ wolves are determined. The position of the individual is updated by calculating the distance between the current individual and α, β, and δ, using the following formula:

[0100] D=|C·X p (T)-X(t)|;

[0101] X(t+1)=X(t)+A·D;

[0102] Where t represents the current iteration number, A and C are coefficients, Xp is the prey position, X(t) is the position of the t-generation gray wolf individual, and D is the distance between the target position and the current individual position. The calculation method of A and C is as follows:

[0103] A = 2a·r1-a;

[0104] C = 2·r2;

[0105] Where r1 and r2 are random values ​​in [0,1]. To simulate approaching prey, A is a random value in the interval [-a,a], where a decreases from 2 to 0 during the iteration.

[0106] Repeat the sorting and updating operations. Figure 3 As shown, until the maximum number of iterations is reached. The iterative convergence curve is shown as Fig. 22 shown.

[0107] The parameter settings and output parameters of the GWO optimization algorithm are shown in Table 5;

[0108] Table 5 Optimization algorithm parameters

[0109] Iterations 5 Gray Wolf's search range 4 The number of gray wolves seeking value 5 Optimal fitness 0.173810 Optimal parameters (number of convolution kernels) 18；300；239；243 Hyperparameter upper bound 20，300，300，300 Hyperparameter lower bounds 1，1，1，1

[0110] S7, importing the preprocessed training set and test set into the convolutional neural network (CNN) model after parameter optimization and performing training and pattern recognition on the convolutional neural network (CNN) model;

[0111] S8. Use principal component analysis (PCA) and matplotlib to visualize the results of convolutional neural network (CNN) model pattern recognition.

[0112] The output is the accuracy curve, loss curve, confusion matrix, and scatter plots before and after pattern recognition when the number of iterations is 50. Fig.23 , Fig.24 , Fig.25 , Fig.26 , Fig. 27 shown.

[0113] The training accuracy of the GWO-CNN model is 99.39%; the test set accuracy is 99.15%; combined with the obtained visualization graph, it can be seen that the optimized model has achieved a very high accuracy and the results are very ideal.

[0114] In order to more clearly see the high accuracy of the GWO-CNN model, the same preprocessed data is imported into different models, the obtained data are compared together, and the accuracy curve is drawn, such as Fig.28 As shown in the figure, the selected comparison models are CNN, Alex Net, and LSTM.

[0115] From the comparison in the figure, we can see that a high accuracy rate has been achieved after about 15 iterations, and the accuracy rate is basically stable after 20 iterations, and the accuracy rate is the highest among the selected models. The CNN model optimized by GWO has been significantly improved compared with the original CNN model. This shows the superiority of the GWO-CNN model, and the powerful pattern recognition ability of the VMD-GWO-CNN model after the addition of VMD for various noisy partial discharge signals.

[0116] Table 6 Noise type classification and sample number

[0117] Noise type Number of samples wind 600 rain 600 wind and rain 600 Partial Discharge 600 Wind, partial discharge 600 Rain, partial discharge 600 Wind, rain, partial discharge 600

[0118] Therefore, the present invention adopts the above-mentioned overhead high-voltage transmission line discharge signal recognition method based on VMD-GWO-CNN. The variational mode decomposition VMD can effectively decompose complex signals into multiple modes with different frequency characteristics, and the denoising of the discharge signal is completed through VMD variational mode decomposition; the hyperparameters of CNN are optimized using the gray wolf optimization algorithm GWO; and principal component analysis PCA is used to visualize the results of pattern recognition.

[0119] 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 they can still modify or replace the technical solution of the present invention with 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 discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN, characterized in that: The following steps are involved: S1, collecting partial discharge signals without noise and with noise; S2, performing variational mode decomposition (VMD) on the noise-free partial discharge signal; S3, performing variational mode decomposition (VMD) on the noisy partial discharge signal; reconstructing the IMF and residual of the modal component dominated by the screened partial discharge signal; S4, reading partial discharge signal data sets of various noise types, creating labels for the partial discharge signal data sets and converting the labels into one-hot encoding; dividing the partial discharge signal data sets into a training set and a test set and performing normalization processing; S5. Convolutional neural network CNN model construction: S6. Use the Grey Wolf Optimization Algorithm (GWO) to optimize the number of convolution kernels in the convolutional layer of the convolutional neural network (CNN) model. S7, importing the preprocessed training set and test set into the convolutional neural network (CNN) model after parameter optimization and performing training and pattern recognition on the convolutional neural network (CNN) model; S8. Use principal component analysis (PCA) and matplotlib to visualize the results of convolutional neural network (CNN) model pattern recognition.

2. The method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to claim 1 is characterized in that: In step S1, the sound sensor and the signal acquisition card are placed on a discharge test bench to collect 20kv partial discharge signals at a sampling frequency of 48000Hz, and the partial discharge signals containing noise are obtained by playing the wind, rain and other noises recorded in advance.

3. The method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to claim 1 is characterized in that: In step S2, the decomposition of the modal component IMFs at different modal decomposition levels is calculated, and the center frequency of each modal component IMF is obtained; the optimal modal decomposition level is selected and the center frequency of the corresponding modal component IMF is recorded.

4. The method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to claim 1, characterized in that: In step S3, the modal component IMF closest to the center frequency of the optimal modal decomposition layer is obtained, and this modal component IMF is used as a threshold to filter the IMF components dominated by the partial discharge signal, and the remaining modal component IMFs are regarded as noise-dominated IMF components.

5. The method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to claim 1 is characterized in that: In step S3, the obtained reconstructed signal is used as the local discharge signal after noise reduction, and the reconstructed signal and various local discharge signals containing different noises are combined into a local discharge signal data set.

6. The method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to claim 1, characterized in that: In step S4, local discharge signal data of various noise types are read, 240,000 data points are intercepted for the discharge signal data of different noise types, and the intercepted data of different noise types are divided into 600 samples according to the standard of 400 data points for a single sample; labels are made for the local discharge signal data set and converted into unique hot encoding, and each type of sample is divided into a training set and a test set in a ratio of 7:3, and normalized.

7. The method for identifying discharge signals of overhead high-voltage transmission lines based on VMD-GWO-CNN according to claim 1, characterized in that: In step S5, the convolutional neural network (CNN) model includes an input layer, a convolution layer, a normalization layer, a maximum pooling layer, a fully connected layer, and an output layer.

Images