Partial discharge signal classification method and device of cable, terminal equipment and storage medium

Abstract

The invention discloses a partial discharge signal classification method and device of a cable, terminal equipment and a storage medium, after a partial discharge signal is acquired, time domain features and frequency domain features in the partial discharge signal are extracted based on a signal classification model, and the partial discharge signal is classified based on local feature information corresponding to the two types of features and a dependency relationship between the features. And identifying the category corresponding to the target partial discharge signal. According to the method, the characteristics of the partial discharge signals can be reflected more comprehensively by extracting the time domain and frequency domain characteristics at the same time, and in the process of identifying the types, the local characteristic information and the dependency relationship of the characteristics are combined, so that the signal classification model can identify the partial discharge signals of different types more accurately, and the accuracy of the partial discharge signals is improved. Compared with the prior art, the method reduces the time and cost of manual analysis, can accurately classify the types of the partial discharge signals, achieves the timely and accurate finding of the insulation defects and faults in equipment, and guarantees the safe and stable operation of a power system.

Description

Technical Field

[0001] The present invention relates to the technical field of discharge signal processing, and particularly relates to a method, device, terminal device and storage medium for classifying partial discharge signals of a cable. Background Art

[0002] Partial discharge refers to the phenomenon of arc discharge or corona discharge occurring in a local area of an electrical device. When there are defects or damages in the insulation system of a power device, the electric field strength in the local area will exceed the tolerance of the insulating material, thus triggering partial discharge. Common types of partial discharge include air-gap discharge, corona discharge, and surface discharge, etc. Different types of discharge phenomena have different characteristics and manifestations. For example, air-gap discharge usually occurs at the defects in the insulating medium, while corona discharge usually occurs on the surfaces of high-voltage devices, wires, cables and other insulators. By monitoring the partial discharge signal and identifying its corresponding discharge category, insulation defects and faults in the device can be detected in time, and corresponding repair measures can be taken to avoid equipment damage and power outage accidents.

[0003] Traditional classification techniques have many deficiencies in classifying partial discharge signals of cables. For example, relying on specific empirical rules or expert knowledge limits their scope of application and accuracy; or only focusing on some single features of partial discharge signals (such as one of the time-domain features or frequency-domain features), some discharge signals that are obvious in the frequency domain may be missed, thus unable to accurately classify the category of partial discharge signals, which will lead to the inability to timely and accurately discover insulation defects and faults in the device. Therefore, incorrect classification may lead to multiple attempts of different repair methods, increasing the repair cost and time, not only wasting resources, but also possibly affecting the normal operation of the power system. Summary of the Invention

[0004] Embodiments of the present invention provide a method, device, terminal device and storage medium for classifying partial discharge signals of a cable. By simultaneously extracting time-domain and frequency-domain features and considering the dependence relationship between the features, more comprehensive feature extraction and classification are realized, so that the category of partial discharge signals can be accurately classified, effectively solving the problem in the prior art that due to the inability to accurately classify the category of partial discharge signals, insulation defects and faults in the device cannot be timely and accurately discovered, which not only wastes resources, but also may affect the normal operation of the power system.

[0005] An embodiment of the present invention provides a method for classifying partial discharge signals of a cable, including:

[0006] Obtain a target partial discharge signal corresponding to the current time period of the cable;

[0007] Input the target partial discharge signal into a signal classification model, so that the signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal; output the category corresponding to the target partial discharge signal according to the local feature information of each signal feature in the feature matrix and the dependency relationship of each signal feature in the feature matrix; wherein, the signal features include: time domain features and frequency domain features;

[0008] Wherein, the generation of the signal classification model includes:

[0009] Use a number of partial discharge signal samples and the actual category corresponding to each partial discharge signal sample as input, and the predicted category of each partial discharge signal sample as output, and perform iterative training on the signal classification model to be trained until the model converges, generating a trained signal classification model.

[0010] Preferably, obtaining the target partial discharge signal corresponding to the current time period of the cable includes:

[0011] Obtain the initial partial discharge signal of the cable in the current time period;

[0012] Decompose the initial partial discharge signal according to the empirical mode decomposition method to generate a number of initial IMF components;

[0013] Filter each initial IMF component to generate a corresponding denoised IMF component;

[0014] Accumulate each initial IMF component and each denoised IMF component to generate a reconstructed partial discharge signal;

[0015] Use the reconstructed partial discharge signal as the target partial discharge signal.

[0016] Preferably, the time domain features include: amplitude feature, waveform feature, phase feature and time domain integral feature; the frequency domain features include: frequency feature and frequency domain integral feature;

[0017] The signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal, including:

[0018] The signal classification model generates corresponding amplitude features, waveform features and phase features respectively according to the signal intensity of the target partial discharge signal in the current time period, the pattern of the target partial discharge signal changing with time, and the phase angle of the target partial discharge signal in each cycle;

[0019] Generate a time domain integral feature according to the integral operation result of the target partial discharge signal in the time domain;

[0020] Generate frequency features according to the distribution characteristics of the target partial discharge signal in the frequency domain;

[0021] Generate frequency domain integral features according to the result of the integral operation of the target partial discharge signal in the frequency domain;

[0022] Generate a feature matrix with multi-dimensional features according to the amplitude features, waveform features, phase features, time domain integral features, frequency features and frequency domain integral features.

[0023] Preferably, the output of the category corresponding to the target partial discharge signal according to the local feature information of each signal feature in the feature matrix and the dependence relationship of each signal feature in the feature matrix includes:

[0024] Perform a convolution operation on the feature matrix to generate a first feature vector for characterizing the local feature information of the feature matrix;

[0025] Generate a second feature vector for characterizing the global feature of the feature matrix according to the feature matrix and the local feature information between each signal feature in the first feature vector;

[0026] Generate a third feature vector for characterizing the long-distance dependence relationship of the feature matrix according to the first feature vector and the second feature vector;

[0027] Generate the category corresponding to the target partial discharge signal according to the third feature vector.

[0028] Preferably, the signal classification model further includes: a multi-scale feature extraction network layer;

[0029] The performing a convolution operation on the feature matrix to generate a first feature vector for characterizing the local feature information of the feature matrix includes:

[0030] Perform a sparse convolution operation on the local feature information of the feature matrix through convolution kernels of different sizes in the multi-scale feature extraction network layer to generate a feature map corresponding to each convolution kernel; wherein, each feature map corresponds to a different scale;

[0031] After performing an average pooling operation on the feature map corresponding to each convolution kernel, generate a first feature vector for characterizing the local feature information of different scales fused.

[0032] Preferably, the signal classification model further includes: a global feature extraction network layer;

[0033] The generating a second feature vector for characterizing the global feature of the feature matrix according to the feature matrix and the local feature information between each signal feature in the first feature vector includes:

[0034] Convert the feature matrix into a corresponding query matrix, key matrix, and value matrix through a global feature extraction network layer;

[0035] Perform a dot product operation on the query matrix and the key matrix to obtain an attention matrix;

[0036] Perform weighted summation on the value matrix based on the attention matrix to generate a number of correlation features; wherein, each correlation feature is used to characterize the correlation between two different signal features;

[0037] Concatenate each correlation feature with the first feature vector to generate a second feature vector for characterizing the global feature of the feature matrix.

[0038] Preferably, the signal classification model further includes: a number of attention layers;

[0039] The generating, according to the first feature vector and the second feature vector, a third feature vector for characterizing the long-range dependence relationship of the feature matrix includes:

[0040] Through each attention layer, concatenate the first feature vector and the second feature vector to generate a concatenated feature vector; based on the self-attention mechanism, capture the long-range dependence relationship in the concatenated feature vector;

[0041] Generate a third feature vector according to the long-range dependence relationships output by each attention layer.

[0042] Based on the above method embodiments, the present invention correspondingly provides apparatus embodiments.

[0043] An embodiment of the present invention provides a partial discharge signal classification apparatus for a cable, including: a signal acquisition module and a signal classification module;

[0044] The signal acquisition module is used to acquire a target partial discharge signal corresponding to the cable in the current time period;

[0045] The signal classification module is used to input the target partial discharge signal into the signal classification model, so that the signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal; and output the category corresponding to the target partial discharge signal according to the local feature information of each signal feature in the feature matrix and the dependence relationship between each signal feature in the feature matrix; wherein, the signal features include: time domain features and frequency domain features;

[0046] Wherein, the generation of the signal classification model includes:

[0047] Using a number of partial discharge signal samples and the actual category corresponding to each partial discharge signal sample as input, and the predicted category of each partial discharge signal sample as output, iteratively train the signal classification model to be trained until the model converges, generating a trained signal classification model.

[0048] Based on the above method embodiments, the present invention correspondingly provides terminal device embodiments.

[0049] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a method for classifying partial discharge signals of a cable as described in the above invention embodiments.

[0050] Based on the above method embodiments, the present invention correspondingly provides storage medium embodiments.

[0051] Another embodiment of the present invention provides a storage medium. The computer-readable storage medium includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute a method for classifying partial discharge signals of a cable as described in the above invention embodiments.

[0052] By implementing the present invention, the following beneficial effects are achieved:

[0053] An embodiment of the present invention provides a method, apparatus, terminal device, and storage medium for classifying partial discharge signals of a cable. After obtaining a target partial discharge signal, the present invention can extract time-domain features and frequency-domain features from the signal based on a signal classification model, and identify the category corresponding to the target partial discharge signal based on the local feature information of the feature matrix corresponding to these two types of features and the dependence relationship of each signal feature in the feature matrix. Since the time-domain features provide direct information on how the signal changes over time, and the frequency-domain features reveal the components and distribution of the signal at different frequencies, the present invention can more comprehensively reflect the characteristics of the partial discharge signal by simultaneously extracting time-domain and frequency-domain features. Moreover, during the process of identifying the category, the local feature information and dependence relationship of these features are also combined, enabling the signal classification model to more accurately identify different categories of partial discharge signals. Compared with the prior art, the automated classification process of the present invention greatly reduces the time and cost of manual classification, thereby not only improving the efficiency of partial discharge signal classification but also enhancing the efficiency of cable fault diagnosis. Additionally, by simultaneously extracting time-domain and frequency-domain features and considering the dependence relationship between the features, the present invention achieves more comprehensive feature extraction and classification, enabling accurate classification of the categories of partial discharge signals, timely and accurately detecting insulation defects and faults in equipment, avoiding unnecessary resource waste, ensuring the safe and stable operation of the power system, and reducing the occurrence of power outages. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 FIG. is a schematic flowchart of a method for classifying partial discharge signals of a cable provided by an embodiment of the present invention.

[0055] Figure 2 FIG. is a schematic flowchart of another embodiment of the present invention for classifying partial discharge signals.

[0056] Figure 3 FIG. is a schematic structural diagram of an apparatus for classifying partial discharge signals of a cable provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0058] As Figure 1 shown, FIG. is a schematic flowchart of a method for classifying partial discharge signals of a cable provided by an embodiment of the present invention. The method for classifying partial discharge signals of the cable includes:

[0059] Step S1: Obtain the target partial discharge signal corresponding to the cable in the current time period;

[0060] Step S2: Input the target partial discharge signal into the signal classification model, so that the signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal; Output the category corresponding to the target partial discharge signal according to the local feature information of each signal feature in the feature matrix and the dependency relationship of each signal feature in the feature matrix; wherein, the signal features include: time domain features and frequency domain features.

[0061] For step S1, in a preferred embodiment, before obtaining the target partial discharge signal, the present invention can also reconstruct and filter the initial partial discharge signal corresponding to the cable in the current time period, which can effectively remove noise, improve the signal-to-noise ratio of the target partial discharge signal, make the signal clearer and more accurate, and is beneficial to subsequent feature extraction and classification.

[0062] Specifically, the obtaining of the target partial discharge signal corresponding to the cable in the current time period includes:

[0063] Obtain the initial partial discharge signal of the cable in the current time period;

[0064] Decompose the initial partial discharge signal according to the empirical mode decomposition method to generate a number of initial IMF components;

[0065] Filter each initial IMF component to generate a corresponding noise-removed IMF component;

[0066] Accumulate each initial IMF component and each noise-removed IMF component to generate a reconstructed partial discharge signal;

[0067] Use the reconstructed partial discharge signal as the target partial discharge signal.

[0068] It can be understood that in signal classification, the accuracy of features is crucial for the classification result. Through the preprocessing steps (such as the above reconstruction and filtering), it can ensure that the time domain features and frequency domain features extracted by the subsequent model are more real and reliable, thereby improving the recognition accuracy of the signal classification model.

[0069] High-quality signals can reduce the computational amount and complexity in the recognition process of the classification model, because the reduction of noise means that the signal components that the model needs to process are more single and clear. Therefore, the present invention can improve the efficiency of the classification process through the empirical mode decomposition method and filtering, so that the model can give an accurate classification result faster.

[0070] For step S2, in a preferred embodiment, the time-domain features include: amplitude features, waveform features, phase features, and time-domain integral features; the frequency-domain features include: frequency features and frequency-domain integral features.

[0071] After inputting the target partial discharge signal into the signal classification model, when the signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal, it specifically includes:

[0072] The signal classification model first generates corresponding amplitude features, waveform features, and phase features respectively according to the signal intensity of the target partial discharge signal in the current time period, the pattern of change of the target partial discharge signal over time, and the phase angle of the target partial discharge signal in each cycle.

[0073] Generate time-domain integral features according to the integral operation result of the target partial discharge signal in the time domain.

[0074] Next, the signal classification model generates frequency features according to the distribution characteristics of the target partial discharge signal in the frequency domain.

[0075] Generate frequency-domain integral features according to the result of the integral operation of the target partial discharge signal in the frequency domain.

[0076] Finally, according to the amplitude features, waveform features, phase features, time-domain integral features, frequency features, and frequency-domain integral features, construct a feature matrix with multi-dimensional features.

[0077] Specifically, when processing the signal in the time domain, the model can generate features reflecting the amplitude change of the signal by calculating or statistically analyzing parameters such as the peak value and average value of the signal, and these features constitute the amplitude feature part in the feature matrix. By extracting waveform parameters of the signal (such as waveform symmetry, waveform distortion degree, etc.) or performing waveform matching operations, parameters reflecting the waveform features of the signal can be generated, and these parameters constitute the waveform feature part in the feature matrix. The model also analyzes the phase angle of the target partial discharge signal in each cycle, that is, the phase information of the signal. By calculating parameters such as the phase difference and phase stability of the signal, features reflecting the phase change of the signal can be generated, and these features constitute the phase feature part in the feature matrix. The model performs an integral operation on the target partial discharge signal in the time domain to obtain the time-domain integral value of the signal, and this integral value reflects the overall energy or intensity of the signal in the time domain, which constitutes the time-domain integral feature part in the feature matrix.

[0078] Furthermore, when performing frequency-domain processing on the signal, the model analyzes the distribution characteristics of the target partial discharge signal in the frequency domain, that is, the signal spectrum. By calculating parameters such as the spectrum density and main frequency of the signal, features reflecting the frequency distribution of the signal can be generated, and these features constitute the frequency feature part in the feature matrix. The model performs an integration operation on the target partial discharge signal in the frequency domain to obtain the frequency-domain integral value of the signal. The frequency-domain integral value reflects the overall energy or intensity of the signal in the frequency domain, thus constituting the frequency-domain integral feature part in the feature matrix.

[0079] Finally, the model integrates all the above features (amplitude features, waveform features, phase features, time-domain integral features, frequency features, and frequency-domain integral features) together to construct a feature matrix with multi-dimensional features. Schematically, this feature matrix comprehensively reflects the time-domain and frequency-domain characteristics of the target partial discharge signal, and thus provides a rich information basis for subsequent signal classification.

[0080] Therefore, the signal classification model of the present invention can more comprehensively reflect the characteristics of the target partial discharge signal by simultaneously considering the time-domain and frequency-domain characteristics of the signal and extracting feature parameters in multiple dimensions, so that the signal classification model can more accurately identify different types of partial discharge signals, improving the accuracy of classification. Moreover, the construction of the multi-dimensional feature matrix enables the signal classification model to better adapt to signal changes under different conditions, improving the generalization ability of the model.

[0081] As Figure 2 shown, during the training process of the signal classification model of the present invention, the training set and the test set are prepared through the following process:

[0082] First, the original partial discharge (PD) signals of multiple PD events are obtained, and the adaptive ability of empirical mode decomposition (EMD) is used to decompose the original PD signal. Then, the filtered intrinsic mode function (IMF) components after decomposition are filtered, and finally, the denoised IMF and the non-denoised IMF signals are accumulated to reconstruct the desired signal, that is, the reconstructed partial discharge (PD) signal can be obtained.

[0083] Next, the FroFA (Frozen Feature Augmentation) algorithm is used to enhance the features of the reconstructed partial discharge (PD) signal, obtaining more than 30 kinds of feature information including statistical features, time-domain features, frequency-domain features, and wavelet features. Subsequently, it is input into the signal classification model. The DenseNet 201 network in the signal classification model can extract typical features from each feature information in the reconstructed partial discharge (PD) signal, obtaining a multi-dimensional feature vector matrix. Schematically, the multi-dimensional feature vector matrix includes but is not limited to the maximum amplitude of the PD signal, the duration of the PD signal, the curve area of the PD signal, etc. Based on the PD typical feature dataset, the data is divided into a training set and a test set in a ratio of 8:2.

[0084] Schematically, the denoised IMF components and the non-denoised IMF components are accumulated to reconstruct the desired PD signal. The purpose of this step is to minimize the interference of noise while retaining the useful information in the original signal.

[0085] DenseNet 201 is a deep convolutional neural network with powerful feature extraction capabilities. Through the DenseNet 201 network, a multi-dimensional feature vector matrix can be obtained. These feature vector matrices contain various typical features of the PD signal, such as the maximum amplitude, duration, curve area, etc. Schematically, the curve area of the PD signal refers to the area enclosed by the partial discharge signal in the time-amplitude plane, which can reflect the total energy or intensity of the discharge. The size of the curve area is related to the strength and duration of the discharge activity and is an important parameter for evaluating the insulation performance of power equipment. Moreover, the curve area can be obtained by integrating or summing the amplitudes of the discharge signal, so the curve area is the time-domain integral feature or the frequency-domain integral feature.

[0086] In a preferred embodiment, during the training process of the model, the predicted value and the actual value can be input into the Huber Loss together, and the CBO (Colliding Bodies Optimization) optimization algorithm is used to adjust the parameters and train the model together with the test set to obtain the finally trained signal classification model.

[0087] In a preferred embodiment, as Figure 2 shown, the present invention can also, during the actual application of the model, input the signal to be recognized into the trained signal classification model after denoising. After model operation, the category output of the partial discharge is obtained. Then, the obtained partial discharge category and the corresponding partial discharge signal can be recorded into the database together to update and expand the quantity and variety of data. On this basis, the signal classification model can be repeatedly trained to continuously update the parameters of the model.

[0088] In a preferred embodiment, after the extraction of time-domain features and frequency-domain features and the generation of the feature matrix, the category corresponding to the target partial discharge signal is output according to the local feature information of each signal feature in the feature matrix and the dependency relationship of each signal feature in the feature matrix, specifically including:

[0089] Perform a convolution operation on the feature matrix to generate a first feature vector for characterizing the local feature information of the feature matrix;

[0090] According to the feature matrix and the local feature information between each signal feature in the first feature vector, generate a second feature vector for characterizing the global feature of the feature matrix;

[0091] According to the first feature vector and the second feature vector, generate a third feature vector for characterizing the long-distance dependency relationship of the feature matrix;

[0092] Generate the category corresponding to the target partial discharge signal according to the third feature vector.

[0093] It can be understood that the convolution operation moves a sliding window (convolution kernel) on the feature matrix to calculate the weighted sum of the elements in the window and the convolution kernel, thereby generating a new feature representation. Then, through multiple convolutional layers, deeper local feature information can be gradually extracted to form the first feature vector.

[0094] In a preferred embodiment, the signal classification model further includes: a multi-scale feature extraction network layer;

[0095] Then, the performing a convolution operation on the feature matrix to generate a first feature vector for characterizing the local feature information of the feature matrix specifically includes:

[0096] Perform a sparse convolution operation on the local feature information of the feature matrix through convolution kernels of different sizes in the multi-scale feature extraction network layer to generate a feature map corresponding to each convolution kernel; wherein, each feature map corresponds to a different scale;

[0097] After performing an average pooling operation on the feature map corresponding to each convolution kernel, generate a first feature vector for characterizing the local feature information that fuses different scales.

[0098] Schematically, the input feature matrix is a multi-dimensional matrix rather than an image. Therefore, as Figure 2As shown, the network structure of the signal classification model of the present invention is improved based on the original SMT (Scale-Aware Modulation Transformer) network. It does not use traditional convolution operations but uses sparse convolution for operations. It can process multi-dimensional matrices with more than three dimensions, thus solving the problem that the partial discharge multi-scale information cannot be reflected when extracting features, and can greatly reduce the operation time. Based on the multi-dimensional feature matrix, the improved SMT network of the present invention can use sparse convolution and can obtain the predicted category after block operations such as SAM Block, MIX Block, and MSA Block.

[0099] It can be understood that the SAM Block module includes two parts, MHMC and SAA. MHMC introduces convolution kernels of different sizes to perform sparse convolution on the multi-dimensional feature matrix, which can capture multi-scale features; SAA (Multi-Scale Awareness Aggregation) is to enhance the information interaction of multiple scales in MHMC, so as to obtain the first feature vector that fuses the local feature information of different scales. Schematically, in traditional convolution operations, the convolution kernel traverses the entire input data space, including a large number of zero values or invalid regions, which leads to unnecessary computational waste. While sparse convolution only focuses on valid data points (i.e., non-zero value points), thus significantly reducing the computational amount.

[0100] The MIX Block module can stack a SAM Block and a multi-head self-attention (MSA) Block to model the transition from capturing local to global dependencies. The final MSA Block module can effectively capture long-range dependencies and finally output to the fully connected layer to obtain the predicted category.

[0101] It can be understood that the multi-scale feature extraction network layer includes the above-mentioned SAM Block module. The multi-scale feature extraction network layer can capture the feature information of the input data at different scales by introducing convolution kernels of different sizes, thus greatly enhancing the feature expression ability of the model. Since convolution kernels of different scales can capture various types of features, this makes the model more robust in the face of small changes in the input data. In other words, even if the input data undergoes a certain deformation or noise interference, the model can still accurately extract the key feature information.

[0102] Moreover, in the multi-scale feature extraction network layer, feature maps of different scales are fused through operations such as average pooling to generate the first feature vector that fuses the local feature information of different scales, so that the model can learn a more comprehensive and rich feature representation, thereby improving the performance of tasks such as classification and detection.

[0103] Further, in a preferred embodiment, the signal classification model further includes: a global feature extraction network layer;

[0104] Then, generating a second feature vector for characterizing the global features of the feature matrix according to the feature matrix and the local feature information between the signal features in the first feature vector includes:

[0105] Converting the feature matrix into a corresponding query matrix, key matrix, and value matrix through the global feature extraction network layer;

[0106] Performing a dot product operation on the query matrix and the key matrix to obtain an attention matrix;

[0107] Performing a weighted sum on the value matrix based on the attention matrix to generate a number of correlation features; wherein each of the correlation features is used to characterize the correlation between two different signal features;

[0108] Concatenating each correlation feature with the first feature vector to generate a second feature vector for characterizing the global features of the feature matrix.

[0109] Schematically, the global feature extraction network layer may include stacking one SAM Block and one multi-head self-attention (MSA) Block of the above MIX Block module; that is, the feature matrix can be converted into a corresponding query matrix, key matrix, and value matrix through the multi-head self-attention (MSA) Block, and a dot product operation is performed on the query matrix and the key matrix, so as to perform a weighted sum on the value matrix and output a number of correlation features.

[0110] Then, introducing the global feature extraction network layer is to further capture the global correlation between the signal features in the feature matrix and generate a second feature vector for characterizing these global features. The design of this global feature extraction network layer makes full use of the attention mechanism, especially the multi-head self-attention (MSA) mechanism, to capture the complex relationships between features. That is, through the multi-head self-attention mechanism, the global feature extraction network layer can capture the global correlation between different signal features in the feature matrix, such as the similarity, dependence, or mutual influence between features.

[0111] By combining local features (the first feature vector) and global correlation features (generated by the MSA Block), the global feature extraction network layer can generate a more rich and comprehensive feature representation, enabling the model to better understand the internal structure and rules of the input data.

[0112] Further, the signal classification model further includes: a number of attention layers; it can be understood that the number of attention layers can be the multi-head attention layers in the MSA Block module.

[0113] Then, generating a third feature vector for characterizing the long-range dependence relationship of the feature matrix according to the first feature vector and the second feature vector includes:

[0114] Through each attention layer, splicing the first feature vector and the second feature vector to generate a spliced feature vector; based on the self-attention mechanism, capturing the long-range dependence relationship in the spliced feature vector;

[0115] Generating a third feature vector according to the long-range dependence relationships output by each attention layer.

[0116] It can be understood that the long-range dependence relationship refers to the possible correlation or mutual influence between elements that are far apart in the feature matrix. Each attention layer splices the first feature vector (local feature information) and the second feature vector (global feature information) and inputs them into the attention layer. Based on the self-attention mechanism, the correlation between elements that are far apart in the spliced feature vector can be captured. Usually, it involves calculating the query matrix, the key matrix, and the value matrix, and obtaining the attention matrix through dot product operations; according to the attention matrix, performing weighted summation on the value matrix to obtain a new representation of each element. This new representation not only contains the information of the element itself but also integrates the information of other relevant elements, thereby capturing the long-range dependence relationship. Finally, the model can make full use of these two types of feature information, fuse the long-range dependence relationships output by each layer, and generate a third feature vector that integrates local and global features. Therefore, the signal classification model can make full use of the local and global feature information of the input data and the long-range dependence relationship between the features, thereby improving the accuracy and robustness of the classification task.

[0117] Partial discharges can be classified into surface discharges, tip discharges, floating discharges, internal discharges, corona discharges, etc. according to the discharge type. Therefore, after obtaining the category of the partial discharge signal, corresponding repair steps or measures can be taken to eliminate the discharge phenomenon and restore the normal operation of the equipment. By eliminating the partial discharge phenomenon, the occurrence of equipment failures can be reduced, and the reliability and stability of the equipment can be improved.

[0118] In summary, accurately identifying the category of the partial discharge signal and taking corresponding repair steps or measures is of great significance for improving equipment reliability, extending equipment life, and ensuring personnel safety.

[0119] As Figure 3 shown, based on the embodiments of the above various methods for classifying partial discharge signals of cables, the present invention correspondingly provides embodiments of the device item;

[0120] An embodiment of the present invention provides a device for classifying partial discharge signals of a cable, including: a signal acquisition module and a signal classification module;

[0121] The signal acquisition module is used to acquire the target partial discharge signal corresponding to the cable in the current time period;

[0122] The signal classification module is used to input the target partial discharge signal into a signal classification model, so that the signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal; and outputs the category corresponding to the target partial discharge signal according to the local feature information of each signal feature in the feature matrix and the dependence relationship of each signal feature in the feature matrix; wherein, the signal features include: time domain features and frequency domain features;

[0123] Among them, the generation of the signal classification model includes:

[0124] Taking a number of partial discharge signal samples and the actual category corresponding to each partial discharge signal sample as inputs, and taking the predicted category of each partial discharge signal sample as an output, iteratively training the signal classification model to be trained until the model converges, and generating a trained signal classification model.

[0125] It should be noted that the device embodiments described above are only illustrative. The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules. They may be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement without creative efforts.

[0126] Those skilled in the art can clearly understand that for the convenience and brevity, the specific working process of the device described above can refer to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0127] Based on the above embodiments of the various methods for classifying partial discharge signals of a cable, the present invention correspondingly provides embodiments of a terminal device.

[0128] An embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, a method for classifying partial discharge signals of a cable according to any method item embodiment of the present invention is implemented.

[0129] The terminal device may be a computing terminal device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory.

[0130] The so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc. The processor is the control center of the terminal device, and connects various parts of the entire terminal device through various interfaces and lines.

[0131] The memory may be used to store the computer program. The processor realizes various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system, application programs required for at least one function, etc.; the data storage area may store data created according to the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash device, or other volatile solid-state storage devices.

[0132] Based on the above embodiments of the method for classifying partial discharge signals of various cables, the present invention correspondingly provides an embodiment of a storage medium.

[0133] An embodiment of the present invention provides a storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute a partial discharge signal classification method for a cable described in any method embodiment of the present invention.

[0134] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, the steps of the above various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0135] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A method for classifying partial discharge signals of a cable, characterized in that: include: Obtaining a target partial discharge signal corresponding to the cable in the current time period; Input the target partial discharge signal into a signal classification model so that the signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal; output the category corresponding to the target partial discharge signal according to the local feature information of each signal feature in the feature matrix and the dependency relationship of each signal feature in the feature matrix; wherein the signal features include: time domain features and frequency domain features; The generation of the signal classification model includes: Taking a number of partial discharge signal samples and the actual category corresponding to each partial discharge signal sample as input and the predicted category of each partial discharge signal sample as output, the signal classification model to be trained is iteratively trained until the model converges to generate a trained signal classification model.

2. A method for classifying partial discharge signals of a cable as claimed in claim 1, characterized in that: The step of obtaining a target partial discharge signal corresponding to the cable in the current time period includes: Obtaining the initial partial discharge signal of the cable in the current time period; According to the empirical mode decomposition method, the initial partial discharge signal is decomposed to generate several initial IMF components; Filtering each initial IMF component to generate a corresponding IMF component from which noise has been removed; Accumulating each initial IMF component and each IMF component from which noise has been removed to generate a reconstructed partial discharge signal; The reconstructed partial discharge signal is used as the target partial discharge signal.

3. A method for classifying partial discharge signals of a cable as claimed in claim 2, characterized in that: The time domain features include: amplitude features, waveform features, phase features and time domain integral features; the frequency domain features include: frequency features and frequency domain integral features; The signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal, including: The signal classification model generates corresponding amplitude features, waveform features and phase features according to the signal strength of the target partial discharge signal in the current time period, the mode of change of the target partial discharge signal over time and the phase angle of the target partial discharge signal in each cycle; Generate a time domain integral feature according to the integral operation result of the target partial discharge signal in the time domain; Generate frequency features according to the distribution characteristics of the target partial discharge signal in the frequency domain; Generate frequency domain integral features according to the result of the integral operation of the target partial discharge signal in the frequency domain; A feature matrix with multi-dimensional features is generated according to the amplitude features, waveform features, phase features, time domain integral features, frequency features and frequency domain integral features.

4. A method for classifying partial discharge signals of a cable as claimed in claim 3, characterized in that: Outputting the category corresponding to the target partial discharge signal according to the local feature information of each signal feature in the feature matrix and the dependency relationship of each signal feature in the feature matrix includes: Performing a convolution operation on the feature matrix to generate a first feature vector for characterizing local feature information of the feature matrix; Generate a second eigenvector for characterizing the global features of the feature matrix according to the feature matrix and the local feature information between the signal features in the first eigenvector; Generate a third eigenvector for characterizing the long-distance dependency relationship of the feature matrix according to the first eigenvector and the second eigenvector; A category corresponding to the target partial discharge signal is generated according to the third eigenvector.

5. A method for classifying partial discharge signals of a cable as claimed in claim 4, characterized in that: The signal classification model further includes: a multi-scale feature extraction network layer; The step of performing a convolution operation on the feature matrix to generate a first feature vector for characterizing local feature information of the feature matrix includes: Through convolution kernels of different sizes in the multi-scale feature extraction network layer, a sparse convolution operation is performed on the local feature information of the feature matrix to generate a feature map corresponding to each convolution kernel; wherein each feature map corresponds to a different scale; After performing an average pooling operation on the feature map corresponding to each convolution kernel, a first feature vector is generated to represent the local feature information fused at different scales.

6. A method for classifying partial discharge signals of a cable as claimed in claim 5, characterized in that: The signal classification model further includes: a global feature extraction network layer; The step of generating a second eigenvector for characterizing the global features of the feature matrix according to the feature matrix and the local feature information between the signal features in the first eigenvector comprises: The feature matrix is ​​converted into corresponding query matrix, key matrix and value matrix through a global feature extraction network layer; Perform dot product operation on the query matrix and the key matrix to obtain the attention matrix; Performing weighted summation on the value matrix based on the attention matrix to generate a plurality of correlation features; wherein each of the correlation features is used to characterize the correlation between two different signal features; Each correlation feature is concatenated with the first feature vector to generate a second feature vector for characterizing the global features of the feature matrix.

7. A method for classifying partial discharge signals of a cable as claimed in claim 6, characterized in that: The signal classification model further includes: a plurality of attention layers; The step of generating a third eigenvector for characterizing the long-distance dependency relationship of the feature matrix according to the first eigenvector and the second eigenvector includes: Through each attention layer, the first feature vector and the second feature vector are concatenated to generate a concatenated feature vector; based on the self-attention mechanism, the long-distance dependency relationship in the concatenated feature vector is captured; The third eigenvector is generated based on the long-distance dependencies output by each attention layer.

8. A partial discharge signal classification device for a cable, characterized in that: include: Signal acquisition module and signal classification module; The signal acquisition module is used to acquire the target partial discharge signal corresponding to the cable in the current time period; The signal classification module is used to input the target partial discharge signal into the signal classification model so that the signal classification model generates a corresponding feature matrix according to the signal features in the target partial discharge signal; output the category corresponding to the target partial discharge signal according to the local feature information of each signal feature in the feature matrix and the dependency relationship of each signal feature in the feature matrix; wherein the signal features include: time domain features and frequency domain features; The generation of the signal classification model includes: Taking a number of partial discharge signal samples and the actual category corresponding to each partial discharge signal sample as input and the predicted category of each partial discharge signal sample as output, the signal classification model to be trained is iteratively trained until the model converges to generate a trained signal classification model.

9. A terminal device, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, a method for classifying partial discharge signals of a cable as claimed in any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is executed, the device where the storage medium is located is controlled to execute the method for classifying partial discharge signals of a cable according to any one of claims 1 to 7.

Images