Method and apparatus for detecting partial discharge mode in switchgear

Abstract

This invention provides a method and apparatus for detecting partial discharge (PD) modes in switchgear. The method includes: acquiring real samples, which include real PD signals from the switchgear and corresponding PD modes; inputting the real samples into a sample generation model, and after dimensionality reduction and inverse reconstruction by the generator of the sample generation model, generating PD samples with a distribution close to that of the real samples; the generator is used to enhance the constraints on the characteristics of each dimension of the real PD signal during dimensionality reduction and inverse reconstruction; training a PD prediction model based on the PD samples, and inputting the real-time PD signal from the switchgear into the PD prediction model to obtain the PD mode of the switchgear. This invention can improve the accuracy of PD mode detection in switchgear.

Description

Technical Field

[0001] This invention relates to the field of power transmission and transformation technology, and in particular to a method and apparatus for detecting partial discharge modes in switchgear. Background Technology

[0002] Switchgear plays a crucial role in power systems, and its proper operation is fundamental to ensuring the safe and reliable operation of the power system. With the rapid development of my country's ultra-high-voltage power grid, partial discharge has become a key factor causing reduced or even deteriorated transformer insulation performance. Different types of partial discharge can cause varying degrees of damage to electrical equipment. Therefore, accurate identification of partial discharge types can provide reliable reference information for assessing the health status of power equipment and can also serve as an auxiliary means of partial discharge fault location, helping to improve the accuracy of partial discharge fault location.

[0003] However, in partial-frequency pattern recognition, the limited number of samples and the uneven distribution of types lead to problems such as missing samples and incomplete information during the classifier training process. This affects the accuracy of the classifier in recognizing partial-frequency patterns. Summary of the Invention

[0004] This invention provides a method and apparatus for detecting partial discharge modes in switchgear, which can improve the accuracy of partial discharge mode detection in switchgear.

[0005] In a first aspect, the present invention provides a method for detecting partial discharge modes in a switchgear, comprising: acquiring real samples, the real samples including real partial discharge signals of the switchgear and partial discharge modes corresponding to the real partial discharge signals; inputting the real samples into a sample generation model, and generating partial discharge samples with a distribution close to that of the real samples after dimensionality reduction processing and inverse reconstruction by the generator of the sample generation model; the generator is used to enhance the constraints on the features of each dimension of the real partial discharge signal during the dimensionality reduction processing and inverse reconstruction process; training a partial discharge prediction model based on the partial discharge samples, and inputting the real-time partial discharge signal of the switchgear into the partial discharge prediction model to obtain the partial discharge mode of the switchgear.

[0006] This invention provides a method for detecting partial discharge (PD) modes in switchgear. On one hand, a sample generation model generates PD samples, increasing their diversity and enriching the training samples for the PD prediction model. This reduces the problems of missing samples and incomplete information during the PD prediction process, improving the accuracy of the PD prediction model in detecting PD modes. On the other hand, the generator in the sample generation model enhances the constraints on the features of the real PD signal in each dimension during dimensionality reduction and reverse reconstruction. This increases the similarity between the features of the generated PD samples and the features of the real PD signal, meaning the distribution of the PD samples is closer to that of the real samples, improving the authenticity of the PD samples and further enhancing the accuracy of PD mode detection in switchgear.

[0007] In one possible implementation, the generator includes an OGC module and a SAG module. Real samples are input into the sample generation model, and after dimensionality reduction and inverse reconstruction by the generator of the sample generation model, partial discharge (PD) samples that closely approximate the distribution of the real samples are generated. This includes: inputting real samples into the sample generation model; using the OGC module to perform dimensionality reduction on the real PD signals in the real samples to obtain dimensionality-reduced signals; using the SAG module to perform inverse reconstruction on the dimensionality-reduced signals to obtain simulated PD signals; and generating PD samples based on the simulated PD signals and PD modes.

[0008] In one possible implementation, real samples are input into the sample generation model, and the real partial discharge signals in the real samples are dimensionality-reduced through the OGC module to obtain a dimensionality-reduced signal. This includes: for each dimensionality reduction process, the OGC module adaptively determines the weight adjustment value of the learning matrix in this dimensionality reduction process; the weight adjustment value is used to adjust the weights corresponding to each feature value in the dimensionality reduction process; based on the partial discharge signal obtained in the previous dimensionality reduction, the learning matrix, and the weight adjustment value, the dimensionality-reduced signal obtained in this dimensionality reduction process is determined.

[0009] In one possible implementation, the dimensionality-reduced signal is reverse-reconstructed using the SAG module to obtain a simulated partial discharge signal. This includes: for each reverse reconstruction process, the network parameter weights are determined using the SAG module based on a self-attention mechanism; the network parameter weights are used to adjust the weights corresponding to each feature value during the reverse reconstruction process; and the partial discharge signal obtained in the current reverse reconstruction process is determined based on the partial discharge signal obtained in the previous reverse reconstruction process and the network parameter weights.

[0010] In one possible implementation, before generating localized samples that closely approximate the distribution of real samples after the real samples are input into the sample generation model and after dimensionality reduction and reverse reconstruction by the generator of the sample generation model, the following steps are also included: optimizing the model parameters of the pre-set sample generation model based on the real samples to obtain the sample generation model.

[0011] In one possible implementation, the sample generation model is obtained by optimizing the pre-set model parameters based on real samples, including: Step 1, fixing the generator parameters, optimizing the discriminator parameters based on real samples with the goal of reducing the discriminator's loss function and the result of the discriminator determining the result to be true; Step 2, fixing the generator parameters, optimizing the discriminator parameters based on real samples and adding random noise to the generator with the goal of reducing the discriminator's loss function and the result of the discriminator determining the result to be false; Step 3, fixing the discriminator parameters, optimizing the generator parameters based on real samples and adding random noise to the generator, obtaining gradient information with the result of the discriminator determining the result to be true; optimizing the generator parameters based on the gradient information with the goal of reducing the generator's loss function; Step 4, if the generator's loss function and the discriminator's loss function reach a Nash equilibrium point, the optimization process stops, and the sample generation model is obtained; if the generator's loss function and the discriminator's loss function do not reach a Nash equilibrium point, steps 1 to 4 are repeated until the iteration process is exited.

[0012] In one possible implementation, the dimensionality reduction signal during the dimensionality reduction process is generated based on the following formula;

[0013]

[0014] in, The weighting function used to adjust the weights. X is the weighting function used to adjust the weights. n The feature vector W in the nth iteration process b (n+1) For learning the matrix parameter weights for each dimension, X (n+1) These are the eigenvalues ​​during the (n+1)th iteration.

[0015] The partial discharge signal during the reverse reconstruction process is generated based on the following formula;

[0016]

[0017] Among them, W (n+1) W is the weight vector in the (n+1)th iteration. n Let X be the weight vector in the nth iteration, representing the network parameter weights in this dimension. n Let σ be the feature vector at the nth iteration, and σ be the activation function.

[0018] Secondly, embodiments of the present invention provide a detection device for partial discharge modes in a switchgear, comprising: a communication module for acquiring real samples, the real samples including real partial discharge signals of the switchgear and partial discharge modes corresponding to the real partial discharge signals; a processing module for inputting the real samples into a sample generation model, and generating partial discharge samples with distributions close to the real samples after dimensionality reduction processing and inverse reconstruction by the generator of the sample generation model; the generator is used to enhance the constraints on the features of each dimension of the real partial discharge signal during the dimensionality reduction processing and inverse reconstruction process; a partial discharge prediction model is trained based on the partial discharge samples, and the real-time partial discharge signals of the switchgear are input into the partial discharge prediction model to obtain the partial discharge modes of the switchgear.

[0019] In one possible implementation, the generator includes an OGC module and a SAG module; a processing module is specifically used to input real samples into the sample generation model, and through the OGC module, to perform dimensionality reduction processing on the real partial discharge signals in the real samples to obtain dimensionality-reduced signals; through the SAG module, to perform inverse reconstruction on the dimensionality-reduced signals to obtain simulated partial discharge signals; and to generate partial discharge samples based on the simulated partial discharge signals and partial discharge modes.

[0020] In one possible implementation, the processing module is specifically used to, for each dimensionality reduction process, adaptively determine the weight adjustment value of the learning matrix in that dimensionality reduction process through the OGC module; the weight adjustment value is used to adjust the weights corresponding to each feature value in the dimensionality reduction process; based on the partial discharge signal obtained in the previous dimensionality reduction, the learning matrix, and the weight adjustment value, determine the dimensionality reduction signal obtained in this dimensionality reduction process.

[0021] In one possible implementation, the processing module is specifically used to determine the network parameter weights for each reverse reconstruction process through the SAG module based on the self-attention mechanism; the network parameter weights are used to adjust the weights corresponding to each feature value during the reverse reconstruction process; and the partial discharge signal obtained in the current reverse reconstruction process is determined based on the partial discharge signal obtained in the previous reverse reconstruction process and the network parameter weights.

[0022] In one possible implementation, the processing module is also used to optimize the model parameters of a pre-set sample generation model based on real samples to obtain a sample generation model.

[0023] In one possible implementation, the processing module specifically performs the following steps: Step 1, fixing the generator parameters, optimizing the discriminator parameters based on real samples with the goal of reducing the discriminator's loss function, and taking a true result as the result; Step 2, fixing the generator parameters, adding random noise to the generator based on real samples, optimizing the discriminator parameters with the goal of reducing the discriminator's loss function, and taking a false result as the result; Step 3, fixing the discriminator parameters, adding random noise to the generator based on real samples, obtaining gradient information with a true result as the result; optimizing the generator parameters based on the gradient information with the goal of reducing the generator's loss function; Step 4, if the generator's loss function and the discriminator's loss function reach a Nash equilibrium point, stopping the optimization process and obtaining the sample generation model; if the generator's loss function and the discriminator's loss function do not reach a Nash equilibrium point, repeating steps 1 to 4 until exiting the iteration process.

[0024] In one possible implementation, the dimensionality reduction signal during the dimensionality reduction process is generated based on the following formula;

[0025]

[0026] in, The weighting function used to adjust the weights. X is the weighting function used to adjust the weights. n The feature vector W in the nth iteration process b (n+1) For learning the matrix parameter weights for each dimension, X (n+1) These are the eigenvalues ​​during the (n+1)th iteration.

[0027] The partial discharge signal during the reverse reconstruction process is generated based on the following formula;

[0028]

[0029] Among them, W (n+1) W is the weight vector in the (n+1)th iteration. n Let X be the weight vector in the nth iteration, representing the network parameter weights in this dimension. n Let σ be the feature vector at the nth iteration, and σ be the activation function.

[0030] Thirdly, embodiments of the present invention provide an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, the processor being configured to call and run the computer program stored in the memory to perform the steps of the method as described in the first aspect and any possible implementation thereof.

[0031] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps of the method as described in the first aspect and any possible implementation thereof.

[0032] The technical effects of any of the implementation methods in the second to fourth aspects mentioned above can be found in the technical effects of the corresponding implementation method in the first aspect, and will not be repeated here. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart illustrating a method for detecting partial discharge modes in a switchgear according to an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram of a real partial discharge signal under various partial discharge modes provided in an embodiment of the present invention;

[0036] Figure 3 This is a two-dimensional distribution diagram of a real sample and a partial discharge sample provided in an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram of the structure of a partial discharge mode detection device in a switch cabinet provided in an embodiment of the present invention;

[0038] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0039] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0040] In the description of this invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" and "more than one" refer to two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0041] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner to facilitate understanding.

[0042] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include other steps or modules not listed, or may optionally include other steps or modules inherent to such process, method, product, or device.

[0043] To make the objectives, technical solutions, and advantages of the present invention clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0044] As described in the background section, current partial-sound pattern recognition suffers from a limited number of samples and an unbalanced type distribution, leading to issues such as missing samples and incomplete information during classifier training. This negatively impacts the accuracy of the classifier in recognizing partial-sound patterns.

[0045] To solve the above technical problems, such as Figure 1 As shown, this embodiment of the invention provides a method for detecting partial discharge modes in a switchgear. The main body executing this detection method is a detection device for partial discharge modes in a switchgear. The detection method includes steps S101-S103.

[0046] S101. Obtain real samples.

[0047] In this embodiment of the application, the real sample includes the real partial discharge signal of the switch cabinet and the partial discharge mode corresponding to the real partial discharge signal.

[0048] In some embodiments, partial discharge modes may include tip discharge, suspension discharge, bubble discharge, and surface discharge.

[0049] S102. Input the real sample into the sample generation model. After dimensionality reduction and reverse reconstruction by the generator of the sample generation model, a localized sample with a distribution close to that of the real sample is generated.

[0050] In this embodiment, the generator is used to enhance the constraints on the characteristics of each dimension of the real partial discharge signal during dimensionality reduction and reverse reconstruction.

[0051] In some embodiments, the sample generation model may include a generator and a discriminator. Generative adversarial networks (GANs) continuously play against new partial discharge (PD) signals based on the real PD signal distribution through the generator and discriminator in the sample generation model, resulting in a richer variety of PD samples. In some embodiments, the generator includes an OGC module and a SAG module.

[0052] It should be noted that traditional generators use stacked convolutional neural networks to generate new data from the original data by first reducing the dimensionality and then increasing it (i.e., dimensionality reduction followed by reconstruction). The entire generation process only involves sampling and dimensionality reduction via convolution, without distinguishing the feature weights of each dimension of the original data, thus lacking strong constraints. The resulting new, locally generated data often lacks realism. This invention addresses the problem of poor generation quality caused by simply stacking convolutional kernels by using an OGC module combined with a SAG module to construct the generator.

[0053] In some embodiments, the OGC module uses a two-branch BAN module to upsample the original data, reducing its dimensionality and mapping it to a new dimension. During training, the two-branch BAN structure can accelerate the training process by adjusting the network weights. Compared to the traditional generator that only updates the generator parameters by reversing the discriminator, the two-branch BAN structure of the OGC module can adaptively update and adjust the representative features inherited from the original data, thereby updating the learning matrix parameter weights of each dimension during the dimensionality reduction process.

[0054] The feature vectors obtained by OGC dimensionality reduction have a stronger correspondence with the original data, which enhances the generator's constraint on the dimensionality reduction process of the original partial discharge data and provides a strong guarantee for the authenticity of SAG-reconstructed partial discharge data.

[0055] In some embodiments, the SAG module consists of four SAGN networks with self-attention mechanisms. Unlike traditional convolutional methods, the self-attention mechanism captures weighted responses from all feature locations, while the weights are estimated with low computational cost. This effectively aggregates global feature information in the feature vector and updates the SAGN network parameters in this weight update manner. That is, during the reverse reconstruction process, the network parameter weights are updated through the self-attention mechanism, making the generation direction closer to the original partial discharge data, which also enhances the constraint on the generation process.

[0056] By constraining the approximation of reality at each stage in the two processes of generator dimensionality reduction and reverse reconstruction, the authenticity and richness of the generated partial discharge data are effectively enhanced.

[0057] As one possible implementation, the detection device can generate a partial discharge sample based on steps S1021-S1024.

[0058] S1021. Input the real sample into the sample generation model, and use the OGC module to perform dimensionality reduction processing on the real partial discharge signal in the real sample to obtain the dimensionality-reduced signal.

[0059] As one possible implementation, for each dimensionality reduction process, the detection device can use the OGC module to adaptively determine the weight adjustment value of the learning matrix in this dimensionality reduction process; the weight adjustment value is used to adjust the weights corresponding to each feature value in the dimensionality reduction process; based on the partial discharge signal obtained in the previous dimensionality reduction, the learning matrix, and the weight adjustment value, the dimensionality reduction signal obtained in this dimensionality reduction process is determined.

[0060] For example, the dimensionality reduction signal in the dimensionality reduction process is generated based on the following formula.

[0061]

[0062] in, The weighting function used to adjust the weights. X is the weighting function used to adjust the weights. n The feature vector W in the nth iteration process b (n+1) For learning the matrix parameter weights for each dimension, X (n+1) These are the eigenvalues ​​during the (n+1)th iteration. This represents the XOR operation.

[0063] For example, the partial discharge signal during the reverse reconstruction process is generated based on the following formula.

[0064]

[0065] Among them, W (n+1) W is the weight vector in the (n+1)th iteration. n Let X be the weight vector in the nth iteration, representing the network parameter weights in this dimension. n Let be the feature vector at the nth iteration, σ be the activation function, and ⊙ represent the XOR operation.

[0066] S1022. The reduced-dimensional signal is reconstructed in reverse using the SAG module to obtain the analog partial discharge signal.

[0067] As one possible implementation, for each reverse reconstruction process, the detection device can use the SAG module to determine the network parameter weights in the current reverse reconstruction process based on the self-attention mechanism; the network parameter weights are used to adjust the weights corresponding to each feature value in the reverse reconstruction process; based on the partial discharge signal obtained in the previous reverse reconstruction process and the network parameter weights, the partial discharge signal obtained in the current reverse reconstruction process is determined.

[0068] S1023. Generate partial discharge samples based on analog partial discharge signals and partial discharge modes.

[0069] S103. Based on partial discharge samples, a partial discharge prediction model is trained.

[0070] S104. Input the real-time partial discharge signal of the switchgear into the partial discharge prediction model to obtain the partial discharge mode of the switchgear.

[0071] This invention provides a method for detecting partial discharge (PD) modes in switchgear. On one hand, a sample generation model generates PD samples, increasing their diversity and enriching the training samples for the PD prediction model. This reduces the problems of missing samples and incomplete information during the PD prediction process, improving the accuracy of the PD prediction model in detecting PD modes. On the other hand, the generator in the sample generation model enhances the constraints on the features of the real PD signal in each dimension during dimensionality reduction and reverse reconstruction. This increases the similarity between the features of the generated PD samples and the features of the real PD signal, meaning the distribution of the PD samples is closer to that of the real samples, improving the authenticity of the PD samples and further enhancing the accuracy of PD mode detection in switchgear.

[0072] It should be noted that with the rapid development of advanced sensing and computer technologies, artificial intelligence technology has been widely applied in the field of power equipment fault diagnosis. Machine learning algorithms have been fully integrated with partial discharge (PD) fault diagnosis, and fault diagnosis systems based on artificial neural networks (ANN), support vector machines (SVM), and random forests (RF) have achieved significant results in transformer operation and maintenance management. Traditional machine learning algorithms are essentially shallow learning, making it difficult to mine the high-dimensional features of PD signals, resulting in low accuracy in PD signal pattern recognition. Deep learning, due to its superior data feature mining capabilities, has gradually become the fastest-growing and most promising method in the field of PD fault diagnosis. Deep learning-based PD signal pattern recognition models have strong requirements for the quantity and type balance of PD signal samples. Too few samples can lead to overfitting or even non-convergence during training, while an imbalanced sample structure can cause the pattern recognition results to favor the majority class, resulting in poor recognition performance for the minority class. In actual production, since partial discharge is an occasional fault, the number of partial discharge samples is scarce and the distribution of types is extremely uneven, which may lead to missed detection of partial discharge faults and low accuracy of pattern recognition. This limits the recognition effect and generalization ability of deep learning-based partial discharge signal pattern recognition models. Therefore, the work of enhancing partial discharge signal samples is particularly important for improving the accuracy of partial discharge signal pattern recognition.

[0073] To overcome the challenges of limited sample size and uneven class distribution in partial discharge signal pattern recognition, data augmentation methods are commonly used to improve the training effect and generalization ability of classifiers. Traditional data augmentation methods include undersampling, oversampling, and image transformation. In machine vision, image transformation methods are widely used, generating new samples by performing a series of transformations such as rotation, translation, and scaling on the image. Undersampling loses samples from many classes, leading to information loss and limiting its application. Oversampling utilizes existing minority samples for data augmentation to achieve the goal of training set balance. Researchers have proposed synthetic minority class oversampling techniques, which generate new, non-repeating samples by interpolating between neighboring samples of the minority class, thus avoiding overfitting. Most oversampling methods typically rely on artificial samples composed of a subset of the minority class, ignoring the overall distribution characteristics, thus offering limited improvement in model classification performance. Currently, deep learning algorithms are rapidly developing in the fields of natural language and image recognition. Generative adversarial networks (GANs) can obtain the latent distribution patterns between data and generate new artificial samples through unsupervised learning without any prior assumptions. While traditional GAN ​​models can generate artificial datasets, they are very sensitive to noise. Because they use JS divergence as a measure of the distance between the generated data and the original data, if the distribution of the initial input noise is not chosen properly, the probability distribution of the generated data will not coincide with the probability distribution of the real dataset. In this case, the generator's loss function will always be constant, which means that the gradient vanishing situation has been encountered, and the entire training cannot be carried out.

[0074] Optionally, the method for detecting partial discharge mode in switchgear provided in this embodiment of the invention further includes step S201 before step S102.

[0075] S201. Based on real samples, optimize the model parameters of the pre-set sample generation model to obtain the sample generation model.

[0076] As one possible implementation, the detection device can obtain a sample generation model based on steps one through four.

[0077] Step 1: Fix the parameters of the generator, optimize the parameters of the discriminator based on real samples with the goal of reducing the loss function of the discriminator and the result of the discriminator judging it as true.

[0078] For example, the discriminator loss function can be expressed as the following formula.

[0079]

[0080] Where V1 represents the discriminator loss function, E x～Pr(x)Let D(x) represent the expectation of the probability distribution that the real partial discharge signal follows, and let D(x) represent the output of the discriminator after the real partial discharge signal is input. This represents the expectation of the probability distribution that the analog partial discharge signal generated by the generator follows. This represents the output of the discriminator after the analog partial discharge signal generated by the generator is input.

[0081] Step 2: Fix the parameters of the generator, based on real samples, and add random noise to the generator to reduce the loss function of the discriminator. Optimize the parameters of the discriminator by taking the result of the discriminator judging it as false.

[0082] Step 3: Fix the parameters of the discriminator, add random noise to the generator based on real samples, and obtain gradient information by taking the discriminator's judgment as true. Based on the gradient information, optimize the generator parameters with the goal of reducing the generator's loss function.

[0083] For example, the generator loss function can be expressed as the following formula.

[0084] V2 = E z～P(z) log(D(G(z)));

[0085] Where V2 represents the generator loss function, E z～P(z) G(z) represents the expected probability distribution that the partial discharge signal follows after adding random noise, G(z) represents the partial discharge signal generated by the generator after adding random noise, and D() represents the output of the discriminator.

[0086] It should be noted that, in order to address the problem of gradient vanishing in the JS divergence objective function of traditional models, the sample generation model in this embodiment of the invention introduces Wasserstein distance. Even if the two probability distributions do not overlap, gradient information can still be preserved for the generator during training. The distance formula for gradient information is shown below.

[0087]

[0088] Among them, P r P represents the probability distribution of the real partial discharge signal. f Π(P) represents the probability distribution of the simulated partial discharge signal in the partial discharge sample. r ,P f ) represents the set of joint distributions of the probability distributions of the real partial discharge signal and the analog partial discharge signal, where γ follows the sequence Π(P r ,P f The joint probability distribution follows Π(P) for any . r ,P f The set of γ values ​​from the joint probability distribution, from which a pair (x, y) is sampled, E(x,y)～γ [||xy||] represents the expected value of the partial discharge signal with respect to distance under the joint probability distribution γ set, in Π(P r ,P f The infimum of the expected value of all sets of γ satisfying the joint probability distribution in the set W(P) is the Wasserstein distance of that joint probability distribution. r ,P f ) represents the gradient information, inf() is the infimum function, and ||xy|| is the norm.

[0089] It should be noted that the Wasserstein distance is transformed into the Kantorovich-Rubinstein dual form, and the dual form of the Wasserstein distance is shown below.

[0090]

[0091] The dual form of the Wasserstein distance is to fit a function f(x) such that at the Lipschitz constant ||f|| of f(x), ... L Under the condition that it does not exceed K, The maximum value that can be obtained, divided by K, is the Wasserstein distance between the two probability distributions.

[0092] It should be noted that, in this embodiment of the invention, a gradient penalty function is added while introducing the Wasserstein distance. The gradient penalty is applied independently to each sample to limit the problem of the gradient descent of the sample data too fast during training, thereby achieving the Lipschitz constraint. The loss function of the sample generation model SG-GAN is shown below.

[0093]

[0094] Among them, L G Let L be the generator loss function. D Let P be the discriminator loss function. r Let z be the prior distribution of random noise. This represents the expected value of the probability distribution that the partial discharge signal follows after adding random noise. λ is the gradient penalty term, and λ is the regularization coefficient.

[0095] Step 4: If the loss functions of the generator and the discriminator reach the Nash equilibrium point, the optimization process stops and the sample generation model is obtained; if the loss functions of the generator and the discriminator do not reach the Nash equilibrium point, steps 1 to 4 are repeated until the iteration process is exited.

[0096] For example, such as Figure 2As shown, this embodiment of the invention provides a schematic diagram of a real partial discharge signal under various partial discharge modes. Using this embodiment, the real partial discharge signal is dimensionality-reduced to generate partial discharge samples. For example... Figure 3 As shown in the figure, this embodiment of the invention provides a two-dimensional distribution diagram of real samples and partial emission samples. The similarity index between real samples and partial emission samples is shown in Table 1.

[0097] Table 1

[0098] Tip discharge Suspension discharge Bubble discharge Surface discharge mean SG-GAN 0.9863 0.9766 0.9398 0.8164 0.9298

[0099] It should be noted that the embodiments of the present invention can evaluate the pattern recognition accuracy of the partial discharge prediction model based on two types of classification accuracy indicators.

[0100] For example, embodiments of the present invention can calculate the first type of index based on the following formula.

[0101]

[0102] Where, λ accuracy The numbers represent the pattern recognition accuracy, where TP represents the number of positive classes identified as positive, TN represents the number of negative classes identified as negative, FN represents the number of positive classes identified as negative, and FP represents the number of negative classes identified as positive.

[0103] For example, embodiments of the present invention can calculate the second type of index based on the following formula.

[0104]

[0105] Where, λ F1 The F1 score is represented by PR, which is the product of precision and recall. P represents precision, R represents recall, TP represents the number of positive classes identified as positive, FP represents the number of negative classes identified as positive, and FN represents the number of negative classes identified as negative.

[0106] This invention applies data augmentation to two pattern recognition methods, SVM and SAE. The pattern recognition results before and after data augmentation are shown in Table 2.

[0107] Table 2

[0108]

[0109] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0110] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0111] Figure 4 A schematic diagram of a partial discharge mode detection device in a switchgear according to an embodiment of the present invention is shown. The detection device 300 includes a communication module 301 and a processing module 302.

[0112] The communication module 301 is used to acquire real samples, which include the real partial discharge signal of the switch cabinet and the partial discharge mode corresponding to the real partial discharge signal.

[0113] The processing module 302 is used to input real samples into the sample generation model. After dimensionality reduction and inverse reconstruction by the generator of the sample generation model, partial discharge samples with a distribution close to that of real samples are generated. The generator is used to enhance the constraints on the characteristics of each dimension of the real partial discharge signal during the dimensionality reduction and inverse reconstruction process. Based on the partial discharge samples, a partial discharge prediction model is trained, and the real-time partial discharge signal of the switchgear is input into the partial discharge prediction model to obtain the partial discharge mode of the switchgear.

[0114] In one possible implementation, the generator includes an OGC module and a SAG module; the processing module 302 is specifically used to input real samples into the sample generation model, perform dimensionality reduction processing on the real partial discharge signals in the real samples through the OGC module to obtain dimensionality-reduced signals; perform reverse reconstruction on the dimensionality-reduced signals through the SAG module to obtain simulated partial discharge signals; and generate partial discharge samples based on the simulated partial discharge signals and partial discharge modes.

[0115] In one possible implementation, the processing module 302 is specifically used to, for each dimensionality reduction process, adaptively determine the weight adjustment value of the learning matrix in that dimensionality reduction process through the OGC module; the weight adjustment value is used to adjust the weights corresponding to each feature value in the dimensionality reduction process; and based on the partial discharge signal obtained in the previous dimensionality reduction, the learning matrix, and the weight adjustment value, determine the dimensionality reduction signal obtained in this dimensionality reduction process.

[0116] In one possible implementation, the processing module 302 is specifically used to determine the network parameter weights in each reverse reconstruction process through the SAG module based on the self-attention mechanism; the network parameter weights are used to adjust the weights corresponding to each feature value in the reverse reconstruction process; and the partial discharge signal obtained in the current reverse reconstruction process is determined based on the partial discharge signal obtained in the previous reverse reconstruction process and the network parameter weights.

[0117] In one possible implementation, the processing module 302 is further configured to optimize the model parameters of a pre-set sample generation model based on real samples to obtain a sample generation model.

[0118] In one possible implementation, the processing module 302 is specifically used to perform the following steps: Step 1, fix the parameters of the generator, optimize the parameters of the discriminator based on real samples with the goal of reducing the loss function of the discriminator, and take the result of the discriminator determining it as true; Step 2, fix the parameters of the generator, add random noise to the generator based on real samples, optimize the parameters of the discriminator with the goal of reducing the loss function of the discriminator, and take the result of the discriminator determining it as false; Step 3, fix the parameters of the discriminator, add random noise to the generator based on real samples, and take the result of the discriminator determining it as true to obtain gradient information; optimize the parameters of the generator based on the gradient information with the goal of reducing the loss function of the generator; Step 4, if the loss function of the generator and the loss function of the discriminator reach the Nash equilibrium point, stop the optimization process and obtain the sample generation model; if the loss function of the generator and the loss function of the discriminator do not reach the Nash equilibrium point, repeat steps 1 to 4 until exiting the iteration process.

[0119] In one possible implementation, the dimensionality reduction signal during the dimensionality reduction process is generated based on the following formula;

[0120]

[0121] in, The weighting function used to adjust the weights. X is the weighting function used to adjust the weights. n The feature vector W in the nth iteration process b (n+1) For learning the matrix parameter weights for each dimension, X (n+1) These are the eigenvalues ​​during the (n+1)th iteration.

[0122] The partial discharge signal during the reverse reconstruction process is generated based on the following formula;

[0123]

[0124] Among them, W (n+1) W is the weight vector in the (n+1)th iteration. n Let X be the weight vector in the nth iteration, representing the network parameter weights in this dimension. n Let σ be the feature vector at the nth iteration, and σ be the activation function.

[0125] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. For example... Figure 5As shown, the electronic device 400 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, it implements the steps in the above-described method embodiments, for example... Figure 1 Steps 101 to 103 are shown. Alternatively, when the processor 401 executes the computer program 403, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 4 The functions of the communication module 301 and the processing module 302 shown are illustrated.

[0126] For example, the computer program 403 can be divided into one or more modules / units, which are stored in the memory 402 and executed by the processor 401 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program 403 in the electronic device 400. For example, the computer program 403 can be divided into... Figure 4 The communication module 301 and the processing module 302 are shown.

[0127] The processor 401 may be a Central Processing Unit (CPU), or 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. A general-purpose processor may be a microprocessor or any conventional processor.

[0128] The memory 402 can be an internal storage unit of the electronic device 400, such as a hard disk or memory of the electronic device 400. The memory 402 can also be an external storage device of the electronic device 400, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 400. Furthermore, the memory 402 can include both internal and external storage units of the electronic device 400. The memory 402 is used to store the computer program and other programs and data required by the terminal. The memory 402 can also be used to temporarily store data that has been output or will be output.

[0129] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0130] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0131] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0132] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0133] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0134] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0135] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0136] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for detecting partial discharge mode in a switchgear, characterized in that, include: Obtain real samples, which include real partial discharge signals of the switchgear and partial discharge modes corresponding to the real partial discharge signals; The real samples are input into the sample generation model. After dimensionality reduction and inverse reconstruction by the generator of the sample generation model, partial discharge samples that are close to the distribution of the real samples are generated. The generator is used to enhance the constraints on the characteristics of each dimension of the real partial discharge signal during the dimensionality reduction and inverse reconstruction process. The generator includes an OGC module and a SAG module. Based on the partial discharge samples, a partial discharge prediction model is trained, and the real-time partial discharge signal of the switchgear is input into the partial discharge prediction model to obtain the partial discharge mode of the switchgear. The step of inputting the real sample into the sample generation model, and generating a partial discharge sample that closely approximates the distribution of the real sample after dimensionality reduction and inverse reconstruction by the generator of the sample generation model, includes: inputting the real sample into the sample generation model; performing dimensionality reduction processing on the real partial discharge signal in the real sample through the OGC module to obtain a dimensionality-reduced signal; performing inverse reconstruction on the dimensionality-reduced signal through the SAG module to obtain a simulated partial discharge signal; and generating the partial discharge sample based on the simulated partial discharge signal and the partial discharge mode. Before generating localized samples that closely approximate the distribution of real samples after dimensionality reduction and inverse reconstruction by the generator of the sample generation model, the process further includes: Step 1, fixing the parameters of the generator, optimizing the parameters of the discriminator based on real samples with the goal of reducing the loss function of the discriminator and the result of the discriminator determining the result to be true; Step 2, fixing the parameters of the generator, optimizing the parameters of the discriminator based on the real samples and adding random noise to the generator with the goal of reducing the loss function of the discriminator and the result of the discriminator determining the result to be false; Step 3, fixing the parameters of the discriminator, optimizing the parameters of the generator based on the real samples and adding random noise to the generator, obtaining gradient information with the result of the discriminator determining the result to be true; optimizing the parameters of the generator based on the gradient information with the goal of reducing the loss function of the generator; Step 4, if the loss function of the generator and the loss function of the discriminator reach the Nash equilibrium point, the optimization process is stopped, and the sample generation model is obtained; if the loss function of the generator and the loss function of the discriminator do not reach the Nash equilibrium point, steps 1 to 4 are repeated until the iteration process is exited.

2. The method for detecting partial discharge mode in a switchgear according to claim 1, characterized in that, The step of inputting the real sample into the sample generation model and performing dimensionality reduction processing on the real partial discharge signal in the real sample through the OGC module to obtain the dimensionality-reduced signal includes: For each dimensionality reduction process, the OGC module adaptively determines the weight adjustment value of the learning matrix for that dimensionality reduction process; the weight adjustment value is used to adjust the weights corresponding to each feature value during the dimensionality reduction process. Based on the partial discharge signal obtained in the previous dimensionality reduction, the learning matrix, and the weight adjustment value, the dimensionality-reduced signal obtained in this dimensionality reduction process is determined.

3. The method for detecting partial discharge mode in a switchgear according to claim 1, characterized in that, The step of reconstructing the dimensionality-reduced signal using the SAG module to obtain the analog partial discharge signal includes: For each reverse reconstruction process, the SAG module determines the network parameter weights in that reverse reconstruction process based on the self-attention mechanism; the network parameter weights are used to adjust the weights corresponding to each feature value in the reverse reconstruction process. Based on the partial discharge signal obtained in the previous reverse reconstruction process and the network parameter weights, the partial discharge signal obtained in the current reverse reconstruction process is determined.

4. The method for detecting partial discharge mode in a switchgear according to any one of claims 1-3, characterized in that, The dimensionality reduction signal in the dimensionality reduction process is generated based on the following formula; ； in, The weighting function used to adjust the weights. The weighting function used to adjust the weights. The feature vector in the nth iteration. To learn the weights of the matrix parameters for each dimension, These are the eigenvalues ​​during the (n+1)th iteration. The partial discharge signal in the reverse reconstruction process is generated based on the following formula; ； in, This is the weight vector in the (n+1)th iteration. Let be the weight vector in the nth iteration, representing the network parameter weights in this dimension. Let be the feature vector at the nth iteration. This is the activation function.

5. A detection device for partial discharge mode in a switch cabinet, characterized in that, include: The communication module is used to acquire real samples, which include real partial discharge signals of the switchgear and partial discharge modes corresponding to the real partial discharge signals; The processing module is used to input the real samples into the sample generation model, and after dimensionality reduction and inverse reconstruction by the generator of the sample generation model, generate partial discharge samples that are close to the distribution of the real samples; the generator is used to enhance the constraints on the features of each dimension of the real partial discharge signal during the dimensionality reduction and inverse reconstruction process; the generator includes an OGC module and a SAG module; based on the partial discharge samples, a partial discharge prediction model is trained, and the real-time partial discharge signal of the switchgear is input into the partial discharge prediction model to obtain the partial discharge mode of the switchgear; The processing module is specifically used to input the real sample into the sample generation model, perform dimensionality reduction processing on the real partial discharge signal in the real sample through the OGC module to obtain a dimensionality-reduced signal, perform reverse reconstruction on the dimensionality-reduced signal through the SAG module to obtain a simulated partial discharge signal, and generate the partial discharge sample based on the simulated partial discharge signal and the partial discharge mode. The processing module is further configured to perform the following steps: Step 1, fix the parameters of the generator, optimize the parameters of the discriminator based on real samples with the goal of reducing the loss function of the discriminator, and take the discriminator's judgment as true as the result; Step 2, fix the parameters of the generator, add random noise to the generator based on the real samples, optimize the parameters of the discriminator with the goal of reducing the loss function of the discriminator, and take the discriminator's judgment as false as the result; Step 3, fix the parameters of the discriminator, add random noise to the generator based on the real samples, and take the discriminator's judgment as true as the result, to obtain gradient information; Based on the gradient information, the parameters of the generator are optimized with the goal of reducing the loss function of the generator; in step four, if the loss function of the generator and the loss function of the discriminator reach the Nash equilibrium point, the optimization process is stopped and the sample generation model is obtained; if the loss function of the generator and the loss function of the discriminator do not reach the Nash equilibrium point, steps one to four are repeated until the iteration process is exited.

6. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor being configured to invoke and run the computer program stored in the memory to perform the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 4 above.

Images