Cable partial discharge mode identification method and system based on double attention mechanism CNN, medium and processor

Abstract

The invention relates to the technical field of partial discharge detection, and discloses a cable partial discharge mode identification method and system based on a double attention mechanism CNN, a medium and a processor, and the method comprises the steps: obtaining a cable partial discharge signal; carrying out denoising processing on the partial discharge signal through a denoising method based on the combination of spectrum clustering and EWT to obtain a partial discharge reconstruction signal; inputting a partial discharge reconstruction signal into a CNN introducing a space attention mechanism and a channel attention mechanism to carry out learning training, and obtaining a trained partial discharge mode recognition model; and inputting a cable partial discharge signal acquired in real time into the trained partial discharge mode identification model to obtain a cable partial discharge mode identification result. According to the method, a space and channel double-attention mechanism is introduced in feature optimization and fusion, the feature map of the partial discharge signal is optimized, and the recognition accuracy of the CNN on the partial discharge signal when the CNN processes the complex partial discharge signal is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of partial discharge detection technology, and in particular to a cable partial discharge pattern recognition method, system, medium and processor based on a dual attention mechanism CNN. Background Art

[0002] As modern society continues to increase its requirements for power supply reliability, the safe and stable operation of the power system is of vital importance. As a key device for power transmission, the operating status of cables directly affects the reliability of the power grid. Partial discharge is one of the early manifestations of cable insulation aging and failure. Accurately identifying cable partial discharge patterns can detect potential faults in advance and prevent large-scale power outages.

[0003] Traditional partial discharge detection methods often rely on manual feature extraction, which has problems such as feature redundancy, high dimensionality, and low recognition accuracy. With the development of deep learning technology, especially convolutional neural networks (CNNs) have achieved remarkable results in the field of image recognition, providing new methods for automated feature extraction and classification. However, when processing complex partial discharge signals, simple CNNs may not be able to fully capture important feature information between spaces and channels, resulting in low recognition accuracy of existing CNNs when processing complex partial discharge signals. Summary of the invention

[0004] In view of the problem that the existing CNN in the prior art has low recognition accuracy when processing complex partial discharge signals, the present invention provides a cable partial discharge pattern recognition method, system, medium and processor based on a dual attention mechanism CNN, which can introduce spatial and channel attention mechanisms in feature optimization, optimize the feature map of the partial discharge signal, and effectively improve the recognition accuracy of the model for the partial discharge signal. The specific technical solution is as follows:

[0005] A cable partial discharge pattern recognition method based on dual attention mechanism CNN, comprising:

[0006] Acquire a first partial discharge signal collected by an ultra-high frequency sensor on the cable and a second partial discharge signal collected by an ultrasonic sensor;

[0007] The first partial discharge signal and the second partial discharge signal are denoised by a denoising method based on a combination of spectrum clustering and EWT to obtain a first partial discharge reconstruction signal and a second partial discharge reconstruction signal;

[0008] The CNN with spatial attention mechanism and channel attention mechanism is introduced into the first partial discharge reconstruction signal and the second partial discharge reconstruction signal for learning and training, and a trained partial discharge pattern recognition model is obtained;

[0009] The cable partial discharge signal collected in real time is input into the trained partial discharge pattern recognition model to obtain the cable partial discharge pattern recognition result.

[0010] Preferably, the cable partial discharge pattern recognition method based on the dual attention mechanism CNN also includes:

[0011] The first partial discharge signal and the second partial discharge signal are subjected to analog-to-digital conversion processing to obtain the first partial discharge signal and the second partial discharge signal in digital signal mode.

[0012] Preferably, the denoising process is performed on the first partial discharge signal and the second partial discharge signal by a denoising method based on spectrum clustering combined with improved EWT to obtain the first partial discharge reconstruction signal and the second partial discharge reconstruction signal, comprising:

[0013] Performing Fourier transform on the first partial discharge signal and the second partial discharge signal converted into digital signals to obtain corresponding model spectra, classifying the signal spectra using a K-means algorithm, and performing inverse discrete Fourier transform on the classified signal spectra to obtain partial discharge signals containing only white noise;

[0014] The spectrum of the partial discharge signal containing only white noise is divided by an order statistical filter to obtain the division result of each frequency segment;

[0015] By improving EWT, adaptive filtering is performed on each divided frequency segment to obtain the modal components corresponding to each frequency segment.

[0016] The initial modal components are selected from the modal components of each frequency segment by high-order statistics, and the selected modal components are threshold-judged to obtain the final modal components;

[0017] The first partial discharge signal and the second partial discharge signal are reconstructed according to the final modal components to obtain a first partial discharge reconstructed signal and a second partial discharge reconstructed signal.

[0018] Preferably, the first partial discharge signal and the second partial discharge signal converted into digital signals are subjected to Fourier transformation to obtain corresponding model spectra, the signal spectra are classified by K-means algorithm, and the classified signal spectra are subjected to inverse discrete Fourier transformation to obtain partial discharge signals containing only white noise, which includes:

[0019] Performing discrete Fourier transform on the first partial discharge signal and the partial discharge noise signal in the second partial discharge signal after being converted into digital signals to obtain the corresponding signal spectrum, and calculating the amplitudes corresponding to different frequency points on the signal spectrum to obtain an amplitude set, and using the amplitude set as the first feature quantity of the K-means algorithm;

[0020] According to the amplitudes corresponding to different frequency points, the standard deviation of each amplitude is calculated to obtain the standard deviation set, which is used as the second feature of the K-means algorithm.

[0021] The amplitude set and the standard deviation set are taken as the original samples, and two initial cluster centers are selected from the original samples. The distance from each element in the set to the cluster center is calculated, and each element is assigned to the cluster closest to the cluster center.

[0022] Calculate the mean of each element in the cluster in each dimension and use the mean as the new cluster center;

[0023] Recalculate the distance between each element in the original sample and the new cluster center until the convergence condition is met;

[0024] The signal spectrum that meets the convergence condition is processed by inverse discrete Fourier transform to obtain a partial discharge signal containing only white noise.

[0025] Preferably, the frequency spectrum of the partial discharge signal containing only white noise is divided by an order statistical filter to obtain the division result of each frequency segment, which includes:

[0026] Calculate multiple local maxima of the noisy PD sequence in the PD signal containing only white noise;

[0027] Calculate the Euclidean distance between each two consecutive local maxima in multiple local maxima to obtain a distance set, and obtain a window width calculation model based on the distance aggregation;

[0028] According to the window width calculation model, the local maximum and minimum points are calculated as the boundary segmentation points of the spectrum;

[0029] According to the boundary division points of the spectrum, the division result of each frequency segment is obtained.

[0030] Preferably, the CNN with spatial attention mechanism and channel attention mechanism introduced into the first partial discharge reconstruction signal and the second partial discharge reconstruction signal is used for learning and training to obtain a trained partial discharge pattern recognition model capable of outputting partial discharge patterns, which includes:

[0031] The deep features corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal are extracted by introducing a CNN architecture with a spatial attention mechanism and a channel attention mechanism, and an optimized feature map corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal is obtained;

[0032] Performing feature fusion on the optimized feature graphs corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal to obtain a joint feature graph;

[0033] The joint feature map is flattened into a one-dimensional vector through the fully connected layer of CNN, and feature fusion and dimensionality reduction are performed through multiple fully connected layers. The fused and reduced feature vectors are classified using the Softmax classifier.

[0034] The one-dimensional vector corresponding to the classified joint feature map is cyclically trained under preset training parameters to obtain a trained local discharge pattern recognition model capable of outputting local discharge patterns.

[0035] Preferably, the extracting of the deep features corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal by introducing the CNN architecture of the spatial attention mechanism and the channel attention mechanism to obtain the optimized feature graphs corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal includes:

[0036] Extracting local features corresponding to the first local discharge reconstruction signal and the second local discharge reconstruction signal through the basic convolution layer of CNN through convolution, batch normalization and activation function;

[0037] By introducing the spatial attention mechanism and performing pooling and convolution operations, a weight map is generated;

[0038] By introducing the channel attention mechanism and obtaining weights through global pooling and shared fully connected layers, an enhanced channel feature map is obtained;

[0039] The weight map and the enhanced channel feature map are multiplied with the original feature maps corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal to obtain an optimized feature map.

[0040] A cable partial discharge pattern recognition system based on a dual attention mechanism CNN is applied to the aforementioned cable partial discharge pattern recognition method based on a dual attention mechanism CNN, comprising:

[0041] A partial discharge signal acquisition unit, used to acquire a first partial discharge signal acquired by an ultra-high frequency sensor on the cable, and a second partial discharge signal acquired by an ultrasonic sensor;

[0042] A denoising processing unit, used for denoising the first partial discharge signal and the second partial discharge signal by a denoising method based on a combination of spectrum clustering and EWT, to obtain a first partial discharge reconstruction signal and a second partial discharge reconstruction signal;

[0043] A model training unit is used to introduce a CNN with a spatial attention mechanism and a channel attention mechanism through the first partial discharge reconstruction signal and the second partial discharge reconstruction signal input for learning and training, so as to obtain a trained partial discharge pattern recognition model;

[0044] The partial discharge pattern recognition unit is used to input the cable partial discharge signal collected in real time into the trained partial discharge pattern recognition model to obtain the cable partial discharge pattern recognition result.

[0045] A computer-readable storage medium includes a stored program, wherein when the program is run, the device where the computer-readable storage medium is located is controlled to execute the aforementioned cable partial discharge pattern recognition method based on the dual attention mechanism CNN.

[0046] A processor is used to run a program, wherein the program executes the aforementioned cable partial discharge pattern recognition method based on the dual attention mechanism CNN when running.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] A cable partial discharge pattern recognition method based on a dual attention mechanism CNN of the present invention obtains a first partial discharge signal collected by an ultra-high frequency sensor on a cable and a second partial discharge signal collected by an ultrasonic sensor; denoises the first partial discharge signal and the second partial discharge signal by a denoising method based on a combination of spectrum clustering and EWT to obtain a first partial discharge reconstruction signal and a second partial discharge reconstruction signal; inputs the first partial discharge reconstruction signal and the second partial discharge reconstruction signal into a CNN introduced with a spatial attention mechanism and a channel attention mechanism for learning and training to obtain a trained partial discharge pattern recognition model; inputs the cable partial discharge signal collected in real time into the trained partial discharge pattern recognition model to obtain a cable partial discharge pattern recognition result. The present invention introduces a spatial and channel dual attention mechanism in feature optimization and fusion to optimize the feature map of the partial discharge signal, thereby effectively improving the recognition accuracy of the model for the partial discharge signal and solving the problem that the existing CNN has low recognition accuracy when processing complex partial discharge signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the prior art description. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0050] Figure 1 This is a flow chart of the cable partial discharge pattern recognition method based on the dual attention mechanism CNN of the present invention.

[0051] Figure 2 This is a flow chart of an embodiment of a cable partial discharge pattern recognition method based on a dual attention mechanism CNN of the present invention.

[0052] Figure 3 This is a schematic diagram of the cable partial discharge pattern recognition system based on the dual attention mechanism CNN of the present invention.

[0053] Figure 4 This is a schematic diagram of CNN performance comparison before and after optimization of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0056] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0057] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0058] See the following examples Figures 1 to 4 .

[0059] The embodiment of the present application provides a cable partial discharge pattern recognition method based on a dual attention mechanism CNN, comprising:

[0060] Step S1, obtaining a first partial discharge signal collected by an ultra-high frequency sensor on the cable and a second partial discharge signal collected by an ultrasonic sensor;

[0061] In this embodiment, the cable partial discharge signal is acquired in real time by using an ultra high frequency (UHF) sensor and an ultrasonic sensor.

[0062] UHF sensors can detect UHF electromagnetic wave signals generated by partial discharge of cables. When cables are partially discharged, they will radiate UHF signals, which can be captured by UHF sensors. This signal has the characteristics of strong anti-interference ability because the electromagnetic environment of the UHF band is less affected by external interference.

[0063] Ultrasonic sensors are used to detect ultrasonic signals generated by partial discharge in cables. The energy generated by partial discharge causes vibrations in the surrounding medium, generating ultrasonic waves. Ultrasonic sensors detect mechanical wave signals that reflect the physical effects of partial discharge on the cable insulation.

[0064] After acquiring the first partial discharge signal and the second partial discharge signal, the first partial discharge signal and the second partial discharge signal are subjected to analog-to-digital conversion processing to obtain the first partial discharge signal and the second partial discharge signal in digital signal mode. The collected analog signal is converted into a digital signal. Analog signals are easily interfered by noise during transmission and processing, and are not convenient for computer storage and analysis. Through the analog-to-digital converter, continuous analog signals can be converted into discrete digital signals, which is convenient for subsequent digital signal processing operations.

[0065] Step S2, denoising the first partial discharge signal and the second partial discharge signal by a denoising method based on a combination of spectrum clustering and EWT to obtain a first partial discharge reconstruction signal and a second partial discharge reconstruction signal;

[0066] After distinguishing the noise sub-signal and the partial discharge sub-signal, the component corresponding to the noise sub-signal is removed, and then the remaining partial discharge sub-signal is reconstructed to obtain a first partial discharge reconstructed signal and a second partial discharge reconstructed signal. In this way, the noise interference in the partial discharge signal can be effectively removed and the signal quality can be improved.

[0067] Step S3, introducing a CNN (convolutional neural network) with a spatial attention mechanism and a channel attention mechanism through the first partial discharge reconstruction signal and the second partial discharge reconstruction signal input for learning and training, to obtain a trained partial discharge pattern recognition model;

[0068] The CNN architecture with the spatial attention mechanism and the channel attention mechanism is used to extract the deep features corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal, and obtain the optimized feature graphs corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal; specifically, the following steps are included:

[0069] The local features corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal are extracted through the basic convolution layer of CNN through convolution, batch normalization and activation function; then, a weight map is generated through pooling and convolution operations by introducing a spatial attention mechanism; then, the weights are obtained through global pooling and shared fully connected layers by introducing a channel attention mechanism to obtain enhanced channel feature maps;

[0070] The weight map and the enhanced channel feature map are multiplied with the original feature maps corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal to obtain an optimized feature map.

[0071] Performing feature fusion on the optimized feature graphs corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal to obtain a joint feature graph;

[0072] The joint feature map is flattened into a one-dimensional vector through the fully connected layer of CNN, and feature fusion and dimensionality reduction are performed through multiple fully connected layers. The fused and reduced feature vectors are classified using the Softmax classifier.

[0073] The one-dimensional vector corresponding to the classified joint feature map is cyclically trained under preset training parameters to obtain a trained local discharge pattern recognition model capable of outputting local discharge patterns.

[0074] By multiplying the spatial and channel attention weight maps with the original feature maps, the feature maps are optimized to improve the model performance and generalization ability. Through feature fusion and classification design, the information of different signal sources can be effectively combined to improve the classification accuracy of the model and achieve reliable detection and identification of cable partial discharge.

[0075] Step S4: inputting the cable partial discharge signal collected in real time into the trained partial discharge pattern recognition model to obtain the cable partial discharge pattern recognition result.

[0076] The real-time collected cable partial discharge signal (after the same analog-to-digital conversion as step S1 and the same denoising process as step S2) is input into the trained partial discharge pattern recognition model. The model makes judgments based on the characteristics of the input signal and the learned pattern recognition knowledge, and outputs the cable partial discharge pattern recognition results. This real-time recognition can timely detect abnormal patterns of cable partial discharge, provide an important basis for cable status monitoring and fault warning, and help ensure the safe and stable operation of the power system.

[0077] In this embodiment, the specific setting process of the convolutional neural network architecture in step S2 regarding feature optimization is as follows:

[0078] S201. Design a multi-layer convolutional layer to extract the deep features of the signal. Each convolutional layer includes a convolution operation, batch normalization, and an activation function (such as ReLU, Rectified Linear Unit).

[0079] Convolution operation: Use multiple convolution kernels (Filters) to convolve the input feature map to extract local features;

[0080] Batch normalization: Normalize the output after convolution to reduce the gradient vanishing and exploding problems and improve training speed. Activation function: Introduce nonlinear factors to enable the network to learn complex patterns.

[0081] S202: The spatial attention mechanism highlights the key positions of the partial discharge signal by weighting the spatial dimensions of the feature map. The specific steps are as follows:

[0082] Perform average pooling and maximum pooling on the input feature map F to generate two feature maps:

[0083] A P =AveragePooling(F)

[0084] M P =MaxPooling(F)

[0085] Concatenate the two feature maps in the channel dimension to form a new feature map S:

[0086] S=[A P ;M P ]

[0087] Pass the concatenated feature map S through a convolutional layer to generate a spatial attention weight map M S :

[0088] M S =σ(Conv(S))

[0089] Where σ is the sigmoid function.

[0090] S203: The channel attention mechanism enhances the focus on important channel features by weighting the channel dimension of the feature map. The specific steps are as follows:

[0091] Perform global average pooling (GAP) and global maximum pooling (GMP) on the input feature map F to generate two vectors:

[0092] A C=GPA(F)

[0093] M C =GMP(F)

[0094] Pass through two shared fully connected layers (FC) respectively, and then add them together to get the weight map W obtained by the channel attention mechanism:

[0095] W=σ(FC(A C )+FC(M C ))

[0096] Where σ is the sigmoid function.

[0097] S204: Multiply the weight maps obtained by the spatial attention mechanism and the channel attention mechanism with the original feature map respectively to obtain the optimized feature map F · :

[0098] F · =M S ·F.W.

[0099] The specific setting process of the convolutional neural network architecture for feature fusion in step S2 is as follows:

[0100] S301: Fuse the UHF signal features and the ultrasonic signal features to form a joint feature map. The specific operation is as follows: Perform convolution operations on the UHF signal features and the ultrasonic signal features respectively to extract the depth features. Concatenate the depth feature maps of the two signals in the channel dimension to form a joint feature map.

[0101] S302: Input the extracted joint feature map to the fully connected layer for high-level feature fusion. The specific operations of the fully connected layer are as follows: The joint feature map is converted into a one-dimensional vector through the flattening operation. It is input to several fully connected layers for feature fusion and dimensionality reduction. Finally, the fused feature vector is classified through the Softmax classifier, and the partial discharge pattern recognition result is output.

[0102] A cable partial discharge pattern recognition method based on a dual attention mechanism CNN of the present invention introduces a spatial and channel dual attention mechanism in feature optimization and fusion, optimizes the feature map of the partial discharge signal, effectively improves the recognition accuracy of the model for the partial discharge signal, and solves the problem that the existing CNN has low recognition accuracy when processing complex partial discharge signals.

[0103] Specifically, in a preferred embodiment of the present application, the denoising method based on spectrum clustering combined with improved EWT (empirical wavelet transform) is used to denoise the first partial discharge signal and the second partial discharge signal to obtain the first partial discharge reconstruction signal and the second partial discharge reconstruction signal, including:

[0104] Step S21, Fourier transforming the first partial discharge signal and the second partial discharge signal after being converted into digital signals to obtain corresponding model spectra, classifying the signal spectra by K-means algorithm, and performing inverse discrete Fourier transform on the classified signal spectra to obtain partial discharge signals containing only white noise; specifically comprising:

[0105] The first partial discharge signal and the partial discharge noise signal in the second partial discharge signal converted into digital signals are subjected to discrete Fourier transform to obtain the corresponding signal spectrum, and the amplitudes corresponding to different frequency points on the signal spectrum are calculated to obtain the amplitude set m 1 =[f a1 ,f a2 ,…,f an ], the amplitude set m 1 As the first feature of the K-means algorithm;

[0106] According to the amplitudes corresponding to different frequency points, calculate the standard deviation of each amplitude and get the standard deviation set m 2 =[σ 1 ,σ 2 ,…,σ n ], the standard deviation set is used as the second feature of the K-means algorithm;

[0107] The amplitude set and standard deviation set are taken as the original samples and M = {m 1 ,m 2}, and select two initial cluster centers from the original sample, calculate the distance from each element in the set to the cluster center, and assign each element to the cluster closest to the cluster center;

[0108] Calculate the mean of each element in the cluster in each dimension and use the mean as the new cluster center;

[0109] Recalculate the distance between each element in the original sample and the new cluster center until the convergence condition is met;

[0110] The signal spectrum that meets the convergence condition is processed by inverse discrete Fourier transform to obtain a partial discharge signal containing only white noise.

[0111] Step S22, dividing the spectrum of the partial discharge signal containing only white noise by using an order statistical filter to obtain a division result of each frequency segment; specifically comprising:

[0112] Calculate the multiple local maxima of the noisy PD sequence in the PD signal containing only white noise, which are f(t 1 ),f(t 2 ),…,f(t p ), where p is the number of local maxima;

[0113] Calculate the Euclidean distance D of each two consecutive local maxima among multiple local maxima to obtain the distance set D max =[D 1 ,D 2 ,…,D n ], and obtain the window width calculation model based on distance aggregation; the window width size S OSF It is expressed as:

[0114] maxS OSF min{D max}

[0115] According to the window width calculation model, the local maximum and minimum points are calculated as the boundary segmentation points of the spectrum;

[0116] According to the boundary division points of the spectrum, the division result of each frequency segment is obtained.

[0117] Step S23, adaptively filtering each divided frequency segment by improving EWT to obtain the modal components corresponding to each frequency segment;

[0118] Different from periodic narrowband interference noise, white noise in partial discharge signal has a wider frequency range. At the same time, the spectrum power of white noise is very close to that of partial discharge signal, and it is difficult to separate it from the partial discharge signal by clustering. The signal is processed by empirical wavelet transform. EWT separates different modes of the signal by splitting the Fourier spectrum, and constructs adaptive wavelet filters of different modes to extract the amplitude modulation and frequency modulation components of the signal to reconstruct the signal. The partial discharge denoising performance is further optimized by improving the spectrum segmentation method.

[0119] Step S24, selecting initial modal components from the modal components of each frequency segment by high-order statistics, and performing threshold discrimination on the selected modal components to obtain final modal components;

[0120] When the modal component contains only white noise, its amplitude follows a normal distribution. The normal distribution has specific statistical properties, and its distribution curve presents a typical bell-shaped curve, and the kurtosis (a measure of the steepness of the distribution curve) is normal. This is an important feature for the normal distribution.

[0121] When non-Gaussian partial discharge signals are mixed in the modal components, the partial discharge signals contain impact components, which will cause the pulse amplitude to deviate significantly from the normal distribution. From the statistical characteristics, the kurtosis value will increase significantly at this time. Kurtosis is a fourth-order statistic, which is very sensitive to such impact signals. By observing the changes in the kurtosis value, it is easy to distinguish between modal components containing only white noise and modal components containing partial discharge signals, so as to select useful modal components. This method takes advantage of the significant difference in statistical characteristics between white noise and partial discharge signals, and can effectively screen out components that are valuable for analyzing partial discharge phenomena from complex modal components.

[0122] Step S25 , reconstructing the first partial discharge signal and the second partial discharge signal according to the final modal components to obtain a first partial discharge reconstructed signal and a second partial discharge reconstructed signal.

[0123] In this embodiment, in digital signal processing, second-order statistics are used to process Gaussian signals. Partial discharge signals include non-Gaussian and non-stationary signals, which need to be analyzed and calculated by higher-order statistics. The local amplification signal denoising method based on spectrum clustering and empirical wavelet transform is used to suppress periodic narrowband interference. The spectrum of the partial discharge signal is divided into two categories using the K-means clustering algorithm. The spectrum severely affected by narrowband interference is processed to obtain a partial discharge signal containing only white noise. In order to remove residual white noise, an improved empirical wavelet transform algorithm based on an order statistical filter is proposed. The partial discharge signal is reconstructed by threshold denoising, which has a better denoising effect than the traditional denoising method.

[0124] In addition, in order to verify the practicality and effectiveness of the cable partial discharge pattern recognition method based on the dual attention mechanism CNN of the present invention, the convolutional neural network (CNN) model optimized based on the dual attention mechanism of the present invention is verified. The experimental results show that the optimized model performs well in multiple performance indicators (such as Figure 4 The accuracy of the basic CNN model increased from 85.2% to 94.5%, the precision and recall rates reached 93.2% and 92.1% respectively, and the F1 score also increased from 83.0% to 92.6%. These results verify the effectiveness of the dual attention mechanism in enhancing the model's attention to key features. When processing complex and noisy partial discharge signals, the optimized CNN shows obvious advantages in performance indicators compared with traditional feature extraction methods and basic CNN models, effectively improving the model recognition accuracy.

[0125] The embodiment of the present application also provides a cable partial discharge pattern recognition system based on a dual attention mechanism CNN, which is applied to the aforementioned cable partial discharge pattern recognition method based on a dual attention mechanism CNN, including:

[0126] A partial discharge signal acquisition unit, used to acquire a first partial discharge signal acquired by an ultra-high frequency sensor on the cable, and a second partial discharge signal acquired by an ultrasonic sensor;

[0127] A denoising processing unit, used for denoising the first partial discharge signal and the second partial discharge signal by a denoising method based on a combination of spectrum clustering and EWT, to obtain a first partial discharge reconstruction signal and a second partial discharge reconstruction signal;

[0128] A model training unit is used to introduce a CNN with a spatial attention mechanism and a channel attention mechanism through the first partial discharge reconstruction signal and the second partial discharge reconstruction signal input for learning and training, so as to obtain a trained partial discharge pattern recognition model;

[0129] The partial discharge pattern recognition unit is used to input the cable partial discharge signal collected in real time into the trained partial discharge pattern recognition model to obtain the cable partial discharge pattern recognition result.

[0130] The functional explanation of each unit in the cable partial discharge pattern recognition system based on the dual attention mechanism CNN of this embodiment corresponds to the explanation in the embodiment of the cable partial discharge pattern recognition method based on the dual attention mechanism CNN, and the technical effects correspond, so they will not be repeated.

[0131] An embodiment of the present application also provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the aforementioned cable partial discharge pattern recognition method based on the dual attention mechanism CNN.

[0132] The technical effect of this embodiment is the same as the technical effect of the cable partial discharge pattern recognition method based on the dual attention mechanism CNN in the embodiment, and will not be repeated here.

[0133] The present invention can be used in many general or special computer system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like.

[0134] An embodiment of the present application also provides a processor, which is used to run a program, wherein the program executes the aforementioned cable partial discharge pattern recognition method based on the dual attention mechanism CNN when running.

[0135] The technical effect of this embodiment is the same as the technical effect of the cable partial discharge pattern recognition method based on the dual attention mechanism CNN in the embodiment, and will not be repeated here.

[0136] The processor in this embodiment may be a central processing unit (CPU), a controller, a microcontroller, or other data processing chips.

[0137] Those of ordinary skill in the art will appreciate that the units of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0138] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0139] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0140] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nlyMemory), random access memory (RAM, RandomAccessMemory), mobile hard disk, magnetic disk or optical disk, etc., which can store program code.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A cable partial discharge pattern recognition method based on dual attention mechanism CNN, characterized in that: include: Acquire a first partial discharge signal collected by an ultra-high frequency sensor on the cable and a second partial discharge signal collected by an ultrasonic sensor; The first partial discharge signal and the second partial discharge signal are denoised by a denoising method based on a combination of spectrum clustering and EWT to obtain a first partial discharge reconstruction signal and a second partial discharge reconstruction signal; The CNN with spatial attention mechanism and channel attention mechanism is introduced into the first partial discharge reconstruction signal and the second partial discharge reconstruction signal for learning and training, and a trained partial discharge pattern recognition model is obtained; The cable partial discharge signal collected in real time is input into the trained partial discharge pattern recognition model to obtain the cable partial discharge pattern recognition result.

2. The cable partial discharge pattern recognition method based on dual attention mechanism CNN according to claim 1 is characterized in that: Also includes: The first partial discharge signal and the second partial discharge signal are subjected to analog-to-digital conversion processing to obtain the first partial discharge signal and the second partial discharge signal in digital signal mode.

3. The cable partial discharge pattern recognition method based on dual attention mechanism CNN according to claim 2 is characterized in that: The denoising method based on spectrum clustering combined with improved EWT is used to denoise the first partial discharge signal and the second partial discharge signal to obtain the first partial discharge reconstruction signal and the second partial discharge reconstruction signal, including: Performing Fourier transform on the first partial discharge signal and the second partial discharge signal converted into digital signals to obtain corresponding model spectra, classifying the signal spectra using a K-means algorithm, and performing inverse discrete Fourier transform on the classified signal spectra to obtain partial discharge signals containing only white noise; The spectrum of the partial discharge signal containing only white noise is divided by an order statistical filter to obtain the division result of each frequency segment; By improving EWT, adaptive filtering is performed on each divided frequency segment to obtain the modal components corresponding to each frequency segment; initial modal components are selected from the modal components of each frequency segment by high-order statistics, and threshold discrimination is performed on the selected modal components to obtain the final modal components; The first partial discharge signal and the second partial discharge signal are reconstructed according to the final modal components to obtain a first partial discharge reconstructed signal and a second partial discharge reconstructed signal.

4. The cable partial discharge pattern recognition method based on dual attention mechanism CNN according to claim 3 is characterized in that: The first partial discharge signal and the second partial discharge signal converted into digital signals are subjected to Fourier transformation to obtain corresponding model spectra, the signal spectra are classified by K-means algorithm, and the signal spectra after classification are subjected to inverse discrete Fourier transformation to obtain partial discharge signals containing only white noise, including: Performing discrete Fourier transform on the first partial discharge signal and the partial discharge noise signal in the second partial discharge signal after being converted into digital signals to obtain the corresponding signal spectrum, and calculating the amplitudes corresponding to different frequency points on the signal spectrum to obtain an amplitude set, and using the amplitude set as the first feature quantity of the K-means algorithm; According to the amplitudes corresponding to different frequency points, the standard deviation of each amplitude is calculated to obtain the standard deviation set, which is used as the second feature of the K-means algorithm. The amplitude set and the standard deviation set are taken as the original samples, and two initial cluster centers are selected from the original samples. The distance from each element in the set to the cluster center is calculated, and each element is assigned to the cluster closest to the cluster center. Calculate the mean of each element in the cluster in each dimension and use the mean as the new cluster center; Recalculate the distance between each element in the original sample and the new cluster center until the convergence condition is met; The signal spectrum that meets the convergence condition is processed by inverse discrete Fourier transform to obtain a partial discharge signal containing only white noise.

5. The cable partial discharge pattern recognition method based on dual attention mechanism CNN according to claim 4 is characterized in that: The frequency spectrum of the partial discharge signal containing only white noise is divided by an order statistical filter to obtain the division result of each frequency segment, including: Calculate multiple local maxima of the noisy PD sequence in the PD signal containing only white noise; Calculate the Euclidean distance between each two consecutive local maxima in multiple local maxima to obtain a distance set, and obtain a window width calculation model based on the distance aggregation; According to the window width calculation model, the local maximum and minimum points are calculated as the boundary segmentation points of the spectrum; According to the boundary division points of the spectrum, the division result of each frequency segment is obtained.

6. The cable partial discharge pattern recognition method based on dual attention mechanism CNN according to claim 1 is characterized in that: The CNN with spatial attention mechanism and channel attention mechanism introduced by the first partial discharge reconstruction signal and the second partial discharge reconstruction signal is trained to obtain a partial discharge pattern recognition model with output partial discharge pattern, which includes: The deep features corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal are extracted by introducing a CNN architecture with a spatial attention mechanism and a channel attention mechanism, and an optimized feature map corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal is obtained; Performing feature fusion on the optimized feature graphs corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal to obtain a joint feature graph; The joint feature map is flattened into a one-dimensional vector through the fully connected layer of CNN, and feature fusion and dimensionality reduction are performed through multiple fully connected layers. The fused and reduced feature vectors are classified using the Softmax classifier. The one-dimensional vector corresponding to the classified joint feature map is cyclically trained under preset training parameters to obtain a trained local discharge pattern recognition model capable of outputting local discharge patterns.

7. The cable partial discharge pattern recognition method based on dual attention mechanism CNN according to claim 1 is characterized in that: The CNN architecture that introduces the spatial attention mechanism and the channel attention mechanism is used to extract the deep features corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal, and the optimized feature graphs corresponding to the first partial discharge reconstruction signal and the second partial discharge reconstruction signal are obtained, including: Extracting local features corresponding to the first local discharge reconstruction signal and the second local discharge reconstruction signal through the basic convolution layer of CNN through convolution, batch normalization and activation function; By introducing the spatial attention mechanism and performing pooling and convolution operations, a weight map is generated; By introducing the channel attention mechanism, weights are obtained through global pooling and shared fully connected layers to obtain an enhanced channel feature map; the weight map and the enhanced channel feature map are multiplied with the original feature map corresponding to the first local discharge reconstruction signal and the second local discharge reconstruction signal to obtain an optimized feature map.

8. A cable partial discharge pattern recognition system based on dual attention mechanism CNN, characterized in that: The cable partial discharge pattern recognition method based on the dual attention mechanism CNN as described in any one of claims 1 to 7 comprises: A partial discharge signal acquisition unit, used to acquire a first partial discharge signal acquired by an ultra-high frequency sensor on the cable, and a second partial discharge signal acquired by an ultrasonic sensor; A denoising processing unit, used for denoising the first partial discharge signal and the second partial discharge signal by a denoising method based on a combination of spectrum clustering and EWT, to obtain a first partial discharge reconstruction signal and a second partial discharge reconstruction signal; A model training unit is used to introduce a CNN with a spatial attention mechanism and a channel attention mechanism through the first partial discharge reconstruction signal and the second partial discharge reconstruction signal input for learning and training, so as to obtain a trained partial discharge pattern recognition model; The partial discharge pattern recognition unit is used to input the cable partial discharge signal collected in real time into the trained partial discharge pattern recognition model to obtain the cable partial discharge pattern recognition result.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the cable partial discharge pattern recognition method based on the dual attention mechanism CNN as described in any one of claims 1 to 7.

10. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the cable partial discharge pattern recognition method based on the dual attention mechanism CNN described in any one of claims 1 to 7.

Images