Partial discharge phase spectrogram classification method and system

Abstract

The invention discloses a partial discharge phase spectrogram classification method and system, and the method comprises the steps: recording the discharge energy and a power frequency phase corresponding to the discharge when power equipment generates partial discharge, obtaining a PRPD spectrogram of the partial discharge of the power equipment, carrying out the convolution and pooling feature extraction operation of the PRPD spectrogram, and carrying out the classification of the partial discharge phase spectrogram. Obtaining a convolution feature map containing the morphological features of the partial discharge clusters; performing kernel transformation on the convolution feature map with the partial discharge cluster morphological features through 1 * 1 convolution, and then obtaining an attention feature map by using self-attention calculation; based on the obtained self-attention feature map, discharge types and corresponding probabilities are obtained through full-connection calculation, a domain adversarial loss function is set, and based on the loss function and the calculated probability of each discharge type, a phase spectrogram classification neural network is trained; and utilizing the trained phase spectrogram classification neural network to realize partial discharge phase spectrogram classification. The focusing of key information can be effectively maintained, and information loss is avoided.

Description

Background Art

[0002] Partial discharge is the breakdown discharge of a local area of an insulating medium under the action of an electric field, but no through-channel is formed. It mainly exists in air gaps or defects inside insulating materials, discharges along the surface of insulators in a dirty or humid environment, air ionization near the tip of a conductor, etc. Partial discharge detection is an important means to evaluate the insulation state of power equipment, which can detect potential problems such as bubbles and cracks before the complete breakdown of the equipment, and avoid power outages caused by sudden insulation failures. By classifying the types of partial discharges in power equipment, such as detecting winding deformation of transformers, oil-paper insulation defects, joint deterioration or main insulation damage of cables, particle contamination or conductor surface defects of gas-insulated switchgear, etc., the reliability of the power system can be significantly improved, the equipment life can be extended, and the development of modern smart grids can be promoted.

[0003] The pattern recognition of early partial discharges relied more on expert experience, and the fault classification was carried out by manually analyzing the phase-discharge quantity-discharge times statistical characteristics of the PRPD spectrogram. However, such methods are limited by subjective judgment and low efficiency, and it is difficult to process a large amount of monitoring data. With the development of computer technology, automatic classification methods combining statistical features (such as fractal dimension, gray image features) and machine learning algorithms (such as support vector machines, BP neural networks) have gradually emerged in the field of partial discharge classification. For example, feature vectors are extracted using the two-dimensional or three-dimensional distribution of the PRPD spectrogram and input into the classifier to achieve discharge type discrimination. Such methods are suitable for small-sample classification, such as distinguishing corona discharge, air-gap discharge and surface discharge of power equipment. Their advantages lie in strong classification ability in high-dimensional space, clear classification principle and flexible classification method, but there are still certain limitations. Traditional machine learning methods rely too much on the quality of feature extraction and are sensitive to noise.

[0004] In recent years, deep learning methods have been used for end-to-end classification of PRPD spectrograms. For example, convolutional neural networks automatically extract the spatial aggregation morphological features of partial discharge pulses through convolutional layers, and obtain cluster or strip patterns in the discharge phase distribution. Such methods achieve automatic learning of spatial features through the form of multi-layer convolution and full connection, avoiding the deviation of artificial feature extraction. Machine learning is significantly superior to traditional methods in partial discharge pattern recognition. Especially, deep learning can process complex signals end-to-end, which is beneficial to the construction of an intelligent diagnosis system adapted to a changing environment.

[0005] Although current research has made progress in the field of partial discharge pattern recognition, there are still certain limitations. Existing convolutional methods analyze local and global features rather disjointedly. Although convolutional networks are good at extracting local spatial features, they have poor ability to analyze the cross-phase global patterns of PRPD spectrograms and are difficult to establish long-range dependencies across phases in PRPD spectrograms, such as the correlation of discharge activities in different phase intervals, resulting in limited global feature expression ability. Most current studies use experimental data under the same working conditions for training and validation, leading to a relatively high accuracy of machine learning but weak generalization ability, poor universality of the model, and insufficient adaptability to new working conditions. These defects of machine learning models for partial discharge pattern recognition restrict the accuracy and robustness of partial discharge diagnosis under complex working conditions. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for classifying partial discharge phase spectrograms to solve the technical problems of insufficient feature modeling and limited generalization ability existing in the existing machine learning models for partial discharge spectrogram classification.

[0007] The present invention adopts the following technical solutions:

[0008] A method for classifying partial discharge phase spectrograms includes the following steps:

[0009] S1. When partial discharge occurs in a power device, record the discharge energy and the power frequency phase corresponding to the discharge, obtain the PRPD spectrogram of the partial discharge of the power device, and perform feature extraction operations of convolution and pooling on the PRPD spectrogram to obtain a convolution feature map containing the morphological features of partial discharge clusters.

[0010] S2. Perform kernel transformation on the convolution feature map with the morphological features of partial discharge clusters through 1x1 convolution, and then use self-attention calculation to obtain an attention feature map.

[0011] S3. Based on the obtained self-attention feature map, obtain the discharge type and the corresponding probability through full connection calculation, set a domain adversarial loss function, train the phase spectrogram classification neural network based on the loss function and the probability of each discharge type calculated, and use the trained phase spectrogram classification neural network to realize the classification of partial discharge phase spectrograms.

[0012] Preferably, step S1 is specifically as follows:

[0013] S101. The convolution process is a two-dimensional discrete operation, and local weighted summation is performed on the input feature map through a sliding window.

[0014] S102. Capture the local spatial pattern of the discharge cluster in the PRPD spectrogram through a learnable convolution kernel W, and extract local features related to the discharge type.

[0015] In S103, the maximum value is taken within the pooling window Ω as the output during the downsampling process, which realizes feature dimensionality reduction while ensuring translational invariance.

[0016] Preferably, the dimensions of the output feature map are as follows:

[0017]

[0018] Where, H out is the dimension of the output feature map, H in is the dimension of the input feature map, p is the padding quantity on both sides in the convolution operation, k is the convolution kernel size, and s is the convolution stride.

[0019] Preferably, step S2 is specifically as follows:

[0020] S201. After multiple convolutions and downsamplings, a convolution graph for feature extraction is obtained, and the convolution features are linearly transformed through three different learnable 1x1 convolution kernels to obtain Query, Key, and Value;

[0021] S202. Calculate the phase - to - phase correlation through the dot product of Q and K, and use V to weight - aggregate the features, thereby modeling the cross - phase long - range dependence of the PRPD spectrogram and realizing the self - attention mechanism;

[0022] S203. Use multiple groups of parallel attention heads in the self - attention convolutional network to learn diverse association patterns of discharge features from different sub - spaces, and splice the results of all heads to enhance the feature expression ability;

[0023] S204. After obtaining the self - attention feature map, calculate the distribution difference of the two - domain features in the reproducing kernel Hilbert space;

[0024] S205. Construct a binary cross - entropy loss to train the domain discriminator to distinguish the spectrogram data of the source domain and the target domain. At the same time, the feature extractor adversarially deceives the discriminator, and uses the adversarial training method to prompt the network to extract the discharge essential features independent of the device, thereby improving the migration and generalization ability of the model.

[0025] Preferably, learning diverse association patterns of discharge features from different sub - spaces and splicing the results of all heads are specifically as follows:

[0026] Multihead(Q,K,V)=Concat(head1,head2,...,head i )W O

[0027] Among them, Multihead(Q, K, V) is the output self-attention feature, Concat() is the matrix concatenation operation, head1, head2,..., head i are i attention heads, and W O is the weight matrix.

[0028] Preferably, the distribution difference between the two-domain features is calculated in the reproducing kernel Hilbert space as follows:

[0029]

[0030] Among them, L MMD is the loss value of the distribution difference between the two-domain features, N s is the number of samples in the source domain, is the attention feature map obtained by calculating the source domain spectrogram through the network, N t is the number of samples in the target domain, is the attention feature map obtained by calculating the target domain spectrogram through the network.

[0031] Preferably, the binary domain adversarial loss L adv is:

[0032]

[0033] Among them, D is the domain classifier, f is the feature extractor, is the classification mathematical expectation value of the source domain samples, is the classification mathematical expectation value of the target domain samples, is the i th source domain sample, is the i-th target domain sample.

[0034] Preferably, step S3 is specifically:

[0035] S301. Use the classifier to classify the feature maps of the source domain and the target domain after convolutional feature extraction and self-attention dot product processing to obtain the classification loss;

[0036] S302. Construct a joint loss function to balance the classification accuracy and the domain adaptation ability, and this process can be controlled and optimized by manually adjusting the hyperparameters α, β, γ.

[0037] Preferably, the binary cross-entropy loss function L C is:

[0038]

[0039] Among them, L C is the binary cross-entropy classification loss value, M is the number of partial discharge defect types, y iis the real discharge label, p i is the predicted discharge probability;

[0040] The combined loss function L total is:

[0041] L total = αL C + βL adv + γL MMD

[0042] where α, β, γ are hyperparameters, and L C is the binary cross-entropy classification loss value, L adv is the binary domain adversarial discriminator loss value, and L MMD is the loss value of the difference in the two-domain feature distributions.

[0043] Second, an embodiment of the present invention provides a partial discharge phase spectrogram classification system, including:

[0044] A feature module, when a power device generates partial discharge, records the discharge energy and the power frequency phase corresponding to the discharge, obtains the PRPD spectrogram of the partial discharge of the power device, and performs feature extraction operations of convolution and pooling on the PRPD spectrogram to obtain a convolution feature map containing the morphological features of the partial discharge cluster;

[0045] A transformation module, performs kernel transformation on the convolution feature map with the morphological features of the partial discharge cluster through 1x1 convolution, and then uses self-attention calculation to obtain an attention feature map;

[0046] A classification module, based on the obtained self-attention feature map, obtains the discharge type and the corresponding probability through full connection calculation, sets a domain adversarial loss function, trains the phase spectrogram classification neural network based on the loss function and the probability of each discharge type calculated, and uses the trained phase spectrogram classification neural network to implement partial discharge phase spectrogram classification.

[0047] Third, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above partial discharge phase spectrogram classification method are implemented.

[0048] Fourth, an embodiment of the present invention provides a computer-readable storage medium, including a computer program. When the computer program is executed by a processor, the steps of the above partial discharge phase spectrogram classification method are implemented.

[0049] Fifth, a chip includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above partial discharge phase spectrogram classification method are implemented.

[0050] In a sixth aspect, an embodiment of the present invention provides an electronic device, including a computer program, which implements the steps of the above partial discharge phase spectrogram classification method when executed by the electronic device.

[0051] Compared with the prior art, the present invention has at least the following beneficial effects:

[0052] A partial discharge phase spectrogram classification method uses a convolutional neural network and a self-attention mechanism to jointly extract partial discharge spectrogram features, and combines domain adversarial training to optimize network parameters. Its core purpose is to construct an intelligent diagnosis model with both local feature sensitivity and global pattern correlation, and at the same time break through the domain shift bottleneck of traditional methods in cross-device and cross-condition scenarios. The convolution operation accurately captures the microscopic spatial morphology of the discharge clusters in the spectrogram through multi-scale deformable convolution kernels, strengthening the ability to express spatial features sensitive to discharge types; the self-attention mechanism analyzes the global statistical characteristics that are difficult to perceive by traditional CNNs through cross-phase long-range dependence modeling. Further introducing domain adversarial training, the gradient reversal strategy is used to force the network to learn device-independent domain-invariant features, eliminating the feature distribution shift caused by sensor differences, device structure differences or environmental interference, so that the model can still maintain high-precision diagnostic capabilities in new devices or sudden working conditions that have not been seen before.

[0053] Furthermore, the convolution layer is set to use deformable convolution kernels to adaptively adjust the receptive field shape and accurately capture the microscopic morphology of irregular discharge clusters; the pooling operation enhances the feature translation invariance through spatial downsampling and suppresses the interference caused by sensor position deviation. Since the discharge triggering probability of insulation defects within the power frequency voltage cycle is strongly correlated with the voltage phase, the convolutional network forms feature extraction strongly associated with the fault type by statistically counting the discharge times and energies in different phase windows.

[0054] Furthermore, the dimensions of the output map are unified to ensure that when the feature map is input into the self-attention model subsequently, the fluctuation of the feature map size caused by the difference in the input spectrogram size can be avoided. Thus, it is ensured that the projection matrix of the self-attention module can statically allocate memory, reducing the computational overhead caused by dynamic adjustment, and also making the feature vectors of different scale discharge patterns in the same metric space.

[0055] Furthermore, using the self-attention module to perform attention extraction on the convolutional feature map can break through the limitation of the local receptive field, establish cross-phase global correlation, and directly model the correlation of discharge activities in any two phase intervals by calculating the dot product attention at all positions of the feature map. At the same time, the morphological features of the partial discharge clusters extracted by the convolutional network and the phase distribution law captured by the self-attention are complementary. Through global relationship modeling, the self-attention can more effectively strip device-related noise and focus on the essential physical features of the discharge.

[0056] Furthermore, multi-head attention is set to learn diverse association patterns of discharge features from different subspaces. The results of all heads are concatenated to ensure collaborative perception in multiple physical dimensions, capturing the distribution law of high-energy discharge pulses from the perspectives of amplitude, phase, etc., and analyzing features such as the dependence relationship between discharge activities and the phase of power frequency voltage. Different attention heads can achieve differential suppression of noise through independent weight allocation, enhancing noise robustness.

[0057] Furthermore, the distribution differences between the features of the source domain and the target domain obtained after feature extraction by the calculation model are calculated. This calculation result provides a differentiable distribution difference metric, directly reflecting the degree of feature offset caused by device differences, and can effectively distinguish the systematic differences existing in the PRPD spectrograms of different power equipment, such as the frequency response characteristic differences caused by the models of built-in sensors and external sensors, and the electromagnetic wave propagation loss differences of partial discharges in power equipment. By minimizing the distribution differences between the two domains, the network is trained to extract domain-invariant features that are independent of the device.

[0058] Furthermore, a domain adversarial loss function is introduced, which can reduce the influence from data distribution differences through adversarial training. Domain adversarial training can make the features learned by the network during training have stronger domain invariance. The network can not only learn more stable feature representations that are not easily affected by operating conditions, but also adapt to different discharge types and noise environments, improving the accuracy and stability of classification. Furthermore, the network is trained using a joint loss function. The classification loss focuses on improving classification accuracy, and the domain difference loss aims to reduce the distribution differences between different domains. Through adversarial training, the domain invariance of the model is further enhanced, avoiding the bias of a single loss function. By introducing hyperparameters to control the ratio between the losses, the weights of each loss term can be flexibly adjusted according to the specific task and dataset characteristics. This enables the model to automatically adjust the focus of the loss function at different training stages and under different data distribution conditions, thereby further improving the training stability and network performance.

[0059] It can be understood that the beneficial effects of the second to sixth aspects above can refer to the relevant descriptions in the first aspect above, and will not be elaborated here.

[0060] In summary, by combining a convolutional network and a self-attention module, the present invention enables the network to automatically extract multi-level features, especially long-range dependencies and global context information, significantly improving the classification accuracy of partial discharge spectrograms of power equipment. Introducing the self-attention mechanism allows the model to more accurately capture important features when facing complex data, improving the accuracy and stability of classification. Using the domain adversarial training loss effectively reduces the distribution difference between the source domain and the target domain, enhancing the domain invariance of the model. Through domain adversarial training, the network can automatically adjust to adapt to changes in different data sources, improving its performance in a cross-domain environment. Due to the domain adaptability and classification ability of the network, the present invention can play a role in a variety of practical application scenarios, especially for complex and variable electrical equipment fault detection, signal classification and other tasks, and can achieve more stable and reliable classification results.

[0061] The technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings required for use in the embodiments of the present application will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present application, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0063] Figure 1 It is a flowchart of the present invention;

[0064] Figure 2 is a correct rate matrix diagram of different models for identifying four typical partial discharge spectrograms. Among them, (a) is the self-attention domain adversarial network, (b) is the deep residual network, (c) is the convolutional neural network, and (d) is the multi-layer perceptron;

[0065] Figure 3 It is a box plot of the correct rate of the model for identifying PRPD spectrograms;

[0066] Figure 4 It is a schematic diagram of a computer device provided by an embodiment of the present invention;

[0067] Figure 5 It is a block diagram of an electronic device provided by an embodiment of the present invention.

[0068] Among them, 60. computer device; 61. processor; 62. memory; 63. computer program; 600. electronic device; 610. processing unit; 620. storage unit; 6201. random access storage unit; 6202. cache storage unit; 6203. read-only storage unit; 6204. program / utilities; 6205. program module; 630. bus; 640. display unit; 650. input / output interface; 660. network adapter; 700. external device. Detailed implementation manners

[0069] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0070] In the description of the present invention, it should be understood that the terms "include" and "comprise" indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0071] It should also be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0072] It should be further understood that the term " / " used in the specification of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the related listed items, and includes these combinations. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present invention generally represents an "or" relationship between the preceding and following related objects.

[0073] It should be understood that although the terms first, second, third, etc. may be used in the embodiments of the present invention to describe preset ranges, etc., these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from each other. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0074] Depending on the context, as used herein, the word "if" can be interpreted as "when" or "while" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detecting (stated condition or event)" can be interpreted as "when determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)".

[0075] Schematic diagrams of various structures according to the disclosed embodiments of the present invention are shown in the drawings. These figures are not drawn to scale, where for the purpose of clear expression, some details are enlarged and some details may be omitted. The shapes of various regions and layers shown in the figures, as well as their relative sizes and positional relationships, are merely exemplary and may deviate in practice due to manufacturing tolerances or technical limitations. Those skilled in the art can additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0076] The present invention provides a method for classifying partial discharge phase spectrograms, constructs a joint feature extraction method of a convolutional network and a self-attention network, and improves the classification accuracy of different types of partial discharge defects through domain adversarial training, which is beneficial to the improvement of intelligent diagnosis performance and the digital transformation of the operation and maintenance mode.

[0077] Embodiment 1

[0078] Please refer to Figure 1 , a method for classifying partial discharge phase spectrograms of the present invention includes the following steps:

[0079] S1. Collect the PRPD spectrogram of partial discharge of the power equipment, establish a cascaded convolutional-attention architecture, and perform feature extraction on the PRPD spectrogram through a convolutional network to obtain a convolutional feature map with the morphological characteristics of partial discharge clusters;

[0080] S101. The convolution process is a two-dimensional discrete operation, and local weighted summation is performed on the input feature map through a sliding window;

[0081] S102. Capture the local spatial pattern of the discharge cluster in the PRPD spectrogram through a learnable convolution kernel W, and extract local features related to the discharge type;

[0082] The convolution formula is as follows:

[0083]

[0084] Where s is the stride, p is the convolution blank padding, k is the convolution kernel size, and c is the convolution channel.

[0085] After convolution, the dimension of the output feature map is as follows:

[0086]

[0087] S103. The downsampling process takes the maximum value within the pooling window Ω as the output to achieve feature dimensionality reduction while ensuring translational invariance.

[0088] For the PRPD spectrogram, the pooling operation can retain significant discharge pulses while suppressing random noise, and at the same time expand the receptive field of the subsequent layer.

[0089]

[0090] S2. The convolution feature map with the morphological features of the partial discharge cluster is subjected to kernel transformation through 1x1 convolution, and then the self-attention dot product is used to obtain the self-attention feature map. The partial discharge spectrogram recognition model is jointly trained through the maximum mean discrepancy calculation, cross-domain discriminator and classifier to improve the classification performance and generalization performance of the partial discharge spectrogram recognition model;

[0091] S201. After multiple convolutions and downsamplings, a convolution map for feature extraction is obtained. The convolution features are linearly transformed through three different learnable 1x1 convolution kernels to obtain Query, Key and Value;

[0092]

[0093] Among them, W Q , W K , W V are projection matrices.

[0094] S202. Calculate the phase correlation through the dot product of Q and K, and use V to weight and aggregate the features, so as to model the cross-phase long-range dependence of the PRPD spectrogram and realize the self-attention mechanism;

[0095]

[0096] Among them, d k is the attention dimension.

[0097] S203. Use multiple groups of parallel attention heads in the self-attention convolutional network to learn diverse association patterns of discharge features from different subspaces, and splice the results of all heads to enhance the feature expression ability;

[0098] Multihead(Q,K,V) = Concat(head1,head2,...,head i )W O (6)

[0099]

[0100] After obtaining the self-attention feature map, calculate the distribution difference of the two-domain features in the reproducing kernel Hilbert space;

[0101] The Gaussian kernel function maps the original features to a high-dimensional space, enabling the quantification of distribution differences with different statistical characteristics. For example, calculate the difference in PRPD spectrograms of equipment with different voltage levels, as follows:

[0102]

[0103] where φ(·) is the mapping function in the reproducing kernel Hilbert space.

[0104] S205. Construct a binary cross-entropy loss to train the domain discriminator to distinguish the spectrogram data of the source domain and the target domain. At the same time, the feature extractor adversarially deceives the discriminator, and an adversarial training method is used to prompt the network to extract the discharge essential features independent of the device, thereby improving the migration and generalization ability of the model.

[0105]

[0106] where D is the domain classifier and f is the feature extractor.

[0107] S3. Use the trained partial discharge spectrogram recognition model to identify the discharge type and the corresponding probability.

[0108] S301. Use the classifier to classify the feature maps of the source domain and the target domain after convolutional feature extraction and self-attention dot product processing to obtain the classification loss;

[0109] The classification loss is as follows:

[0110]

[0111] where y i is the true label and p i is the predicted probability.

[0112] S302. Construct a joint loss function to balance the classification accuracy and the domain adaptation ability. This process can be controlled and optimized by manually adjusting the hyperparameters α, β, γ.

[0113] The joint loss function L total is:

[0114] L total = αL C + βL adv + γL MMD (11)

[0115] where α, β, γ are hyperparameters, L C is, L advTrain the domain discriminator with binary cross - entropy loss, \(L\). MMD is the loss value of the difference in feature distributions between the two domains.

[0116] Those skilled in the art can understand that various aspects of the present invention can be implemented as a system, a method, or a program product. Therefore, various aspects of the present invention can be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software aspects, which can be collectively referred to as "circuit", "module", or "platform" here.

[0117] Embodiment 2

[0118] The present invention provides a partial discharge phase spectrum classification system, which can be used to implement the above - mentioned partial discharge phase spectrum classification method. Specifically, the partial discharge phase spectrum classification system includes a feature module, a transformation module, and a classification module.

[0119] Among them, the feature module, when a power device generates partial discharge, records the discharge energy and the power frequency phase corresponding to the discharge, obtains the PRPD spectrum of the partial discharge of the power device, and performs feature extraction operations of convolution and pooling on the PRPD spectrum to obtain a convolution feature map containing the morphological features of the partial discharge cluster.

[0120] The transformation module performs kernel transformation on the convolution feature map with the morphological features of the partial discharge cluster through 1x1 convolution, and then obtains an attention feature map using self - attention calculation.

[0121] The classification module, based on the obtained self - attention feature map, obtains the discharge type and the corresponding probability through fully - connected calculation, sets a domain - adversarial loss function, trains the phase spectrum classification neural network based on the loss function and the probability of each discharge type calculated, and uses the trained phase spectrum classification neural network to implement partial discharge phase spectrum classification.

[0122] Embodiment 3

[0123] The present invention provides a terminal device, which includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Graphics Processing Unit (GPU), Tensor Processing Unit (TPU), Digital Signal Processor (DSP), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function. The processor described in the embodiments of the present invention can be used for the operation of the partial discharge phase spectrogram classification method, including:

[0124] When a power device generates partial discharge, record the discharge energy and the power frequency phase corresponding to the discharge, obtain the PRPD spectrogram of the partial discharge of the power device, perform feature extraction operations of convolution and pooling on the PRPD spectrogram to obtain a convolution feature map containing the morphological features of the partial discharge cluster; perform kernel transformation on the convolution feature map with the morphological features of the partial discharge cluster through 1x1 convolution, and then use self-attention calculation to obtain an attention feature map; based on the obtained self-attention feature map, obtain the discharge type and the corresponding probability through fully connected calculation, set a domain adversarial loss function, and based on the loss function and the probability of each discharge type calculated, train the phase spectrogram classification neural network, and use the trained phase spectrogram classification neural network to implement partial discharge phase spectrogram classification.

[0125] Please refer to Figure 4 , the terminal device is a computer device. The computer device 60 in this embodiment includes: a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When the computer program 63 is executed by the processor 61, it implements the partial discharge phase spectrogram classification method in the embodiment. To avoid repetition, it will not be elaborated here one by one. Alternatively, when the computer program 63 is executed by the processor 61, it implements the functions of each model / unit in the partial discharge phase spectrogram classification system in the embodiment. To avoid repetition, it will not be elaborated here one by one.

[0126] The computer device 60 can be a computing device such as a desktop computer, a notebook, a handheld computer, and a cloud server. The computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art can understand that Figure 4 These are merely examples of the computer device 60 and do not constitute a limitation on the computer device 60. It may include more or fewer components than shown in the figure, or combine certain components, or have different components. For example, the computer device may also include input / output devices, network access devices, a bus, etc.

[0127] The so-called processor 61 may be a central processing unit (CPU), or may also be other general-purpose processors, a graphics processing unit (GPU), a tensor processing unit (TPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), 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.

[0128] The memory 62 may be an internal storage unit of the computer device 60, such as the hard disk or memory of the computer device 60. The memory 62 may also be an external storage device of the computer device 60, such as a plug-in hard disk equipped on the computer device 60, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0129] Furthermore, the memory 62 may also include both an internal storage unit and an external storage device of the computer device 60. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 may also be used to temporarily store data that has been output or will be output.

[0130] Please refer to Figure 5 , the terminal device is the electronic device 600, and the electronic device 600 is presented in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0131] Among them, the storage unit stores program code, and the program code can be executed by the processing unit 610, so that the processing unit 610 executes the steps according to various exemplary embodiments of the present invention described in the method part of the present specification above. For example, the processing unit 610 can execute the steps as shown in Figure 1 shown.

[0132] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 6201 and / or a cache storage unit 6202, and may further include a read-only storage unit (ROM) 6203.

[0133] The storage unit 620 may further include a program / utilities 6204 having a set (at least one) of program modules 6205. Such program modules 6205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0134] The bus 630 may represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any one of the multiple bus structures.

[0135] The electronic device 600 may also communicate with one or more external devices 700 (such as a keyboard, a pointing device, a Bluetooth device, etc.), and may also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device that enables the electronic device 600 to communicate with one or more other computing devices (such as a router, a modem). Such communication may be performed through the input / output interface 650. And, the electronic device 600 may also communicate with one or more networks (such as a local area network, a wide area network, and / or a public network, such as the Internet) through the network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 through the bus 630. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in combination with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

[0136] Embodiment 4

[0137] The present invention also provides a storage medium, specifically a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and, of course, the expandable storage medium supported by the terminal device. It can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. The computer-readable storage medium provides a storage space that stores the operating system of the terminal. And, in this storage space, there are also stored one or more instructions suitable for being loaded and executed by the processor, and these instructions can be one or more computer programs (including program codes). It should be noted that more specific examples of the computer-readable storage medium here include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0138] The computer-readable storage medium also includes data signals propagated in a baseband or as part of a carrier wave, in which the readable program code is carried. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than the readable storage medium, and this readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical cable, radio frequency, etc., or any suitable combination of the above.

[0139] The program code for performing the operations of the present invention can be written in any combination of one or more programming languages. The programming languages include object-oriented programming languages - such as Java, C++, etc., and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, executed as an independent software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user's computing device through any type of network, including a local area network or a wide area network, or can be connected to an external computing device (for example, by using an Internet service provider to connect through the Internet).

[0140] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the partial discharge phase spectrogram classification method in the above embodiments; the one or more instructions in the computer-readable storage medium are loaded and executed by the processor as follows:

[0141] When a power device generates partial discharge, record the discharge energy and the power frequency phase corresponding to the discharge, obtain the PRPD spectrogram of the partial discharge of the power device, perform feature extraction operations of convolution and pooling on the PRPD spectrogram to obtain a convolution feature map containing the morphological features of the partial discharge cluster; perform kernel transformation on the convolution feature map with the morphological features of the partial discharge cluster through 1x1 convolution, and then use self-attention calculation to obtain an attention feature map; based on the obtained self-attention feature map, obtain the discharge type and the corresponding probability through full connection calculation, set the domain adversarial loss function, train the phase spectrogram classification neural network based on the loss function and the calculated probability of each discharge type, and use the trained phase spectrogram classification neural network to realize the classification of partial discharge phase spectrograms.

[0142] The databases involved in the embodiments provided in the present application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., and is not limited thereto. The processors involved in the embodiments provided in the present application may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., and is not limited thereto.

[0143] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the present invention described and shown in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents the selected embodiments of the present invention. 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.

[0144] Embodiment 5

[0145] The usage process is as follows:

[0146] 1) Use a broadband UHF sensor array (300 MHz - 3 GHz) to synchronously collect partial discharge signals of a power device under various operating conditions;

[0147] 2) Convert the three-dimensional data of phase-discharge quantity-discharge frequency of the original waveform signal into a two-dimensional PRPD spectrogram;

[0148] 3) Annotate the PRPD spectrogram by combining the actual working condition information of the power equipment, and construct a fault label system for subsequent model training;

[0149] 4) Establish a cascaded convolutional-attention architecture, use convolutional operations to extract the morphological features of partial discharge clusters in the PRPD spectrogram, and construct cross-phase dependencies through position encoding and multi-head attention mechanisms;

[0150] 4) Construct a domain adversarial transfer strategy, use the data samples in the laboratory as the source domain, and the spectrograms collected under actual working conditions as the target domain, and use the domain adversarial training method proposed in the present invention to train the self-attention convolutional network;

[0151] 5) Combine the classification loss value, the domain adversarial loss value, and the accuracy rate of the test set, and manually adjust the hyperparameters α, β, γ of formula (11) to select the best network model after training;

[0152] 6) Install a PRPD acquisition device and a host computer on the power equipment under operating conditions, collect the phase and discharge quantity data of partial discharge, convert them into spectrograms through the host computer and record them;

[0153] 7) Use the trained phase spectrogram classification network to automatically calculate the possible discharge types and the corresponding probabilities.

[0154] To verify the partial discharge phase spectrogram classification network based on domain adversarial and self-attention mechanisms proposed in the present invention, deep residual networks, convolutional neural networks, and multi-layer perceptron models commonly used in existing spectrogram classifications were constructed. During training, the four models used the same partial discharge spectrogram data. The data set was divided into a source domain and a target domain according to the working conditions, and the training set and the test set were divided within the two domains, with a ratio of 7:3.

[0155] Table 1 Classification accuracy of partial discharge spectrograms of different networks

[0156]

[0157] Table 1 shows the accuracy rates of the test sets in the target domain after training of the five networks. The performance of the self-attention domain adversarial network is slightly higher than that of other networks, and the accuracy rate reaches 95.354%, which is 3.191% higher than that of the deep residual network, 1.918% higher than that of the convolutional neural network, and 8.106% higher than that of the multi-layer perceptron, respectively. This is because the network model proposed in the present invention realizes the collaborative perception from local to global, captures the microscopic morphology of discharge clusters through convolution, and at the same time uses the self-attention mechanism to capture the macroscopic phase law, and the design of domain adversarial also ensures that the model has sufficient generalization performance.

[0158] In contrast, although the deep residual network can alleviate the problem of gradient disappearance, the low signal-to-noise ratio characteristic of the PRPD spectrogram causes the deep network to be vulnerable to high-frequency noise interference, resulting in the diffusion of invalid activations in the feature map and the gradient being contaminated by noise during backpropagation. The traditional convolutional neural network is restricted by the local receptive field and is difficult to capture the periodic law of the discharge phase distribution. As a traditional machine learning model, the multi-layer perceptron's fully connected structure cannot effectively model the spatial topological characteristics of the partial discharge spectrogram, resulting in a low accuracy rate.

[0159] Please refer to Figure 2, which shows the recognition accuracy rate of the model for the target domain of four typical partial discharge spectrograms. Labels 0 to 3 represent floating potential discharge, free moving particle discharge, tip corona discharge, and insulator surface discharge respectively. The recognition rate of the self-attention model for the four discharge types all exceeds 91%, while the accuracy rate of other models is slightly lower in the more difficult-to-recognize surface discharge. Due to reasons such as random discharge channels and surface contamination in surface discharge, its spectrogram is often difficult to recognize. The model proposed in the present invention maintains focus on the essential characteristics of surface discharge through adversarial training and self-attention mechanism, ensuring the accuracy rate of pattern recognition.

[0160] Please refer to Figure 3 , which is the box plot of the accuracy rate of the model for recognizing the target domain PRPD spectrogram and can effectively characterize the stability and generalization ability of the model. It can be seen that due to the lack of phase space modeling ability, the stability of the multi-layer perceptron model is slightly worse than that of other models, while the self-attention network proposed in the present invention has the highest accuracy rate in the target domain, indicating that its generalization ability is stronger after domain adversarial training and can meet the data migration problem under different on-site working conditions. To sum up, for a method and system for classifying partial discharge phase spectrograms in the present invention, the network has achieved a high accuracy rate on the target domain test set, which is better than other commonly used network models; compared with the deep residual network, convolutional neural network, and multi-layer perceptron, the model of this patent shows higher accuracy. This indicates that when dealing with the classification task of partial discharge phase spectrograms, the model can effectively extract features and perform accurate classification, with strong performance advantages. The design of domain adversarial training ensures the data migration ability of the model under different working conditions. In the target domain, the accuracy rate of the model is not only high but also has strong stability. The self-attention network after domain adversarial training shows higher stability compared with other models and can cope with the data distribution differences caused by changes between the source domain and the target domain, thus having better generalization ability in practical applications. The network can achieve collaborative perception from local to global; it captures the microscopic morphology of partial discharge clusters through the convolutional layer, and at the same time uses the self-attention mechanism to capture the macroscopic phase law. This way of multi-scale feature extraction enables the model to more comprehensively understand and recognize the discharge spectrogram; especially when facing complex discharge types, the network can effectively maintain focus on key information and avoid information loss.

[0161] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for easy distinction and do not limit the protection scope of this application. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiment and will not be elaborated herein.

[0162] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0163] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present invention can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in the form of hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0164] In the embodiments provided by the present invention, it should be understood that the disclosed device / terminal and method can be implemented in other ways. For example, the device / terminal embodiments described above are only illustrative. For example, the division of the module or unit is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical or other form.

[0165] The unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0166] In addition, the functional units in various embodiments of the present invention may be integrated into one processing unit, may exist separately as individual physical units, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of software functional units.

[0167] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it may also be completed by instructing relevant hardware through a computer program. The computer program may be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments may be implemented. Among them, the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium may 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.

[0168] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses, and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate a device for implementing the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0169] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the processesFigure 1 one process or multiple processes and / or blocks Figure 1 the functions specified in one block or multiple blocks.

[0170] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0171] The above content is only to illustrate the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention falls within the protection scope of the claims of the present invention.

Claims

1. A method for classifying partial discharge phase spectrograms, characterized in that, It includes the following steps: S1. When partial discharge occurs in the power equipment, record the discharge energy and the power frequency phase corresponding to the discharge, obtain the PRPD spectrogram of the partial discharge of the power equipment, and perform feature extraction operations of convolution and pooling on the PRPD spectrogram to obtain a convolution feature map containing the morphological features of the partial discharge cluster; S2. Perform kernel transformation on the convolution feature map with the morphological features of the partial discharge cluster through 1x1 convolution, and then obtain the attention feature map using self-attention calculation; S3. Based on the obtained self-attention feature map, obtain the discharge type and the corresponding probability through fully connected calculation, set the domain adversarial loss function, train the phase spectrogram classification neural network based on the loss function and the probability of each calculated discharge type, and use the trained phase spectrogram classification neural network to realize the classification of the partial discharge phase spectrogram.

2. The partial discharge phase spectrum diagram classification method according to claim 1, characterized in that Step S1 is specifically as follows: S101. The convolution process is a two-dimensional discrete operation, and local weighted summation is performed on the input feature map through a sliding window; S102. Capture the local spatial pattern of the discharge cluster in the PRPD spectrogram through the learnable convolution kernel W, and extract the local features related to the discharge type; S103. In the downsampling process, take the maximum value within the pooling window Ω as the output to achieve feature dimension reduction while ensuring translational invariance.

3. The partial discharge phase spectrum diagram classification method according to claim 2, characterized in that, The dimension of the output feature map is as follows: Among them, H out is the dimension of the output feature map, and H in is the dimension of the input feature map. p is the padding number on both sides in the convolution operation, k is the convolution kernel size, and s is the convolution stride.

4. The partial discharge phase spectrum classification method according to claim 1, characterized in that Step S2 is specifically as follows: S201. After multiple convolutions and downsamplings, obtain the convolution map for feature extraction, and perform linear transformation on the convolution features through three different learnable 1x1 convolution kernels to obtain the query Query, key Key, and value Value; S202. Calculate the phase correlation through the dot product of Q and K, and use V to perform weighted aggregation on the features, thereby modeling the cross-phase long-range dependence of the PRPD spectrogram and realizing the self-attention mechanism; S203. Use multiple groups of parallel attention heads in the self-attention convolution network to learn diverse association patterns of discharge features from different subspaces, splice the results of all heads, and enhance the feature expression ability; S204. After obtaining the self-attention feature map, calculate the distribution difference of the two-domain features in the reproducing kernel Hilbert space; S205. Construct a binary cross-entropy loss to train the domain discriminator to distinguish the spectrogram data of the source domain and the target domain. At the same time, the feature extractor adversarially deceives the discriminator, and uses the adversarial training method to prompt the network to extract the discharge essential features independent of the device, thereby improving the migration and generalization ability of the model.

5. The partial discharge phase spectrogram classification method according to claim 4, wherein Learn diverse association patterns of discharge features from different subspaces, splice the results of all heads, specifically as follows: Multihead(Q,K,V)=Concat(head1,head2,...,head i )W O Among them, Multihead(Q, K, V) is the output self-attention feature, Concat() is the matrix concatenation operation, head1, head2,..., head i are the i attention heads, and W O is the weight matrix.

6. The partial discharge phase spectrogram classification method according to claim 4, characterized in that Calculate the distribution difference of the two-domain features in the reproducing kernel Hilbert space as follows: Among them, L MMD is the loss value of the difference in the two-domain feature distributions, N s is the number of samples in the source domain, is the attention feature map obtained by network calculation of the source domain spectrogram, N t is the number of samples in the target domain, is the attention feature map obtained by network calculation of the target domain spectrogram.

7. The partial discharge phase spectrogram classification method according to claim 4, characterized in that Binary field adversarial loss L adv is as follows: Among them, D is the domain classifier, and f is the feature extractor. is the classification mathematical expectation of source domain samples, is the classification mathematical expectation of target domain samples, is the i th source domain sample, is the i-th target domain sample.

8. The method for classifying partial discharge phase spectrograms according to claim 1, characterized in that, Step S3 is specifically as follows: S301. Use the classifier to classify the feature maps of the source domain and the target domain after convolution feature extraction and self-attention dot product processing to obtain the classification loss; S302. Construct a joint loss function to balance classification accuracy and domain adaptation ability. This process can be achieved by manually adjusting hyperparameters α and β ,γ to control the optimization direction.

9. The partial discharge phase spectrogram classification method according to claim 8, wherein Binary cross-entropy loss function L C is as follows: Among them, L C is the binary cross-entropy classification loss value, M is the number of partial discharge defect types, y i is the true discharge label, and p i is the predicted discharge probability; Combined loss function L total is as follows: L total = αL C + βL adv + γL MMD where α, β, γ are hyperparameters, and L C is the binary cross-entropy classification loss value, and L adv is the binary domain adversarial discriminator loss value, and L MMD is the loss value of the difference in the feature distributions of the two domains.

10. A partial discharge phase spectrum classification system, characterized in that, It includes: A feature module. When partial discharge occurs in the power equipment, record the discharge energy and the power frequency phase corresponding to the discharge, obtain the PRPD spectrogram of the partial discharge of the power equipment, and perform feature extraction operations of convolution and pooling on the PRPD spectrogram to obtain a convolution feature map containing the morphological features of the partial discharge cluster; A transformation module that performs kernel transformation on the convolutional feature map with the morphological characteristics of partial discharge clusters through 1x1 convolution, and then obtains an attention feature map using self-attention calculation; A classification module that, based on the obtained self-attention feature map, obtains the discharge type and the corresponding probability through fully connected calculation, sets a domain adversarial loss function, trains a phase spectrum classification neural network based on the loss function and the probability of each discharge type obtained by calculation, and uses the trained phase spectrum classification neural network to achieve partial discharge phase spectrum classification.

Images