Partial discharge identification method of power equipment and electronic equipment

Abstract

The invention discloses a partial discharge identification method of power equipment and electronic equipment. The method comprises the following steps: acquiring a target partial discharge audio signal of power equipment; performing multi-scale feature extraction on the target partial discharge audio signal to obtain multi-scale features of the target partial discharge audio signal, the multi-scale features representing features of the target partial discharge audio signal obtained under multiple scales, and the multiple scales being in one-to-one correspondence with the multiple convolution kernel sizes; performing feature fusion on the multi-scale features to obtain fusion features of the target partial discharge audio signal; and determining a target partial discharge type of the power equipment based on the fusion features. According to the invention, the technical problem of low accuracy of partial discharge identification of power equipment caused by incomplete consideration of factors in a complex and changeable power field environment is solved.

Description

Technical Field

[0001] This invention relates to the field of smart grids, and more specifically, to a method and electronic device for identifying partial discharge in power equipment. Background Technology

[0002] Partial discharge is typically caused by defects or non-uniformity in insulation materials. Partial discharge not only accelerates insulation aging but is also a significant early sign of electrical equipment failure. Therefore, accurate and real-time monitoring and identification of partial discharge is crucial for ensuring the stable operation of power systems and equipment safety. Audio signals, as a non-contact monitoring method, have become an important technology in partial discharge monitoring due to their strong resistance to electromagnetic interference, wide detection range, and convenient acquisition. The audio signals generated by partial discharge are typically in the frequency range of 20Hz-20kHz and contain rich information about the sound pressure wave of the partial discharge pulse, reflecting the type and intensity of discharge within electrical equipment. However, related technologies have significant shortcomings in the identification of partial discharge audio signals, mainly in the following aspects:

[0003] Various types of partial discharges can occur in power equipment, each with its own specific audio signal characteristics. Their dominant frequency distribution, time-domain pulse shape, and energy distribution differ. Therefore, a precise identification method capable of distinguishing these different types is needed. In real-world power environments, partial discharge audio signals are often accompanied by various background noises, such as equipment vibration and environmental noise. These noises reduce the signal-to-noise ratio, blurring the characteristics of the partial discharge signal and increasing the difficulty of identification. Partial discharge audio signals contain important information in both the time and frequency domains, but related technologies often process time-domain or multi-frequency domain features separately, failing to fully utilize the correlation between the time and frequency domains, leading to a decrease in identification accuracy. Real-time monitoring in power fields requires models to not only have high identification accuracy but also fast inference speed to adapt to the computing power limitations of edge computing devices. For partial discharge signal identification, high-quality labeled datasets are crucial for training the model. However, obtaining such labeled datasets in power equipment fields is very difficult. On the one hand, samples of certain fault types are extremely scarce; on the other hand, manual labeling requires personnel with specialized knowledge of power equipment insulation, which is not only inefficient but also costly, limiting the scale of model training and performance improvement. Although the relevant partial discharge audio signal recognition technology can provide a relatively high recognition accuracy within a certain range, when faced with complex and ever-changing power field environments, the recognition accuracy of partial discharge types is insufficient due to incomplete coverage of multi-scale features and failure to fully utilize the complementarity of time-domain and frequency-domain features.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method and electronic device for identifying partial discharge in power equipment, which at least solves the technical problem of low accuracy in identifying partial discharge in power equipment due to incomplete consideration of factors when facing complex and ever-changing power field environments.

[0006] According to one aspect of the present invention, a method for identifying partial discharge of power equipment is provided, comprising: acquiring a target partial discharge audio signal of the power equipment, wherein the target partial discharge audio signal is obtained based on the original partial discharge audio signal of the power equipment; performing multi-scale feature extraction on the target partial discharge audio signal to obtain multi-scale features of the target partial discharge audio signal, wherein the multi-scale features represent the features of the target partial discharge audio signal obtained at multiple scales, and the multiple scales correspond one-to-one with multiple convolutional kernel sizes; performing feature fusion on the multi-scale features to obtain fused features of the target partial discharge audio signal; and determining the type of target partial discharge of the power equipment based on the fused features.

[0007] According to another aspect of the present invention, a partial discharge identification device for power equipment is also provided, comprising: a target partial discharge audio signal acquisition module, configured to acquire a target partial discharge audio signal of the power equipment, wherein the target partial discharge audio signal is obtained based on the original partial discharge audio signal of the power equipment; a multi-scale feature determination module, configured to extract multi-scale features from the target partial discharge audio signal to obtain multi-scale features of the target partial discharge audio signal, wherein the multi-scale features represent the features of the target partial discharge audio signal obtained at multiple scales, and the multiple scales correspond one-to-one with multiple convolutional kernel sizes; a fusion feature determination module, configured to fuse the multi-scale features to obtain fused features of the target partial discharge audio signal; and a target partial discharge type determination module, configured to determine the target partial discharge type of the power equipment based on the fused features.

[0008] According to another aspect of the present invention, a non-volatile storage medium is also provided, the non-volatile storage medium storing a plurality of instructions adapted for loading by a processor and executing any one of the partial discharge identification methods for power equipment described herein.

[0009] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the partial discharge identification method for any of the power devices described above.

[0010] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the partial discharge identification method for any of the power devices described in the present invention.

[0011] In this embodiment of the invention, a target partial discharge audio signal of a power device is acquired, wherein the target partial discharge audio signal is obtained based on the original partial discharge audio signal of the power device; multi-scale feature extraction is performed on the target partial discharge audio signal to obtain multi-scale features of the target partial discharge audio signal, wherein the multi-scale features represent the features of the target partial discharge audio signal obtained at multiple scales, and the multiple scales correspond one-to-one with multiple convolution kernel sizes; feature fusion is performed on the multi-scale features to obtain fused features of the target partial discharge audio signal; based on the fused features, the target partial discharge type of the power device is determined. This achieves the goal of accurately determining the partial discharge type of the power device by performing multi-scale feature extraction and feature fusion on the acquired target partial discharge audio signal, thereby improving the technical effect of partial discharge identification accuracy. This solves the technical problem of low accuracy in identifying partial discharge of power devices due to incomplete consideration of factors when facing complex and ever-changing power field environments. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0013] Figure 1 This is a flowchart of a partial discharge identification method for power equipment according to an embodiment of the present invention;

[0014] Figure 2 This is a schematic diagram of the hardware system of an optional power device according to an embodiment of the present invention;

[0015] Figure 3 This is a flowchart of an optional partial discharge identification method for power equipment according to an embodiment of the present invention;

[0016] Figure 4 This is a schematic diagram of a partial discharge identification device for power equipment according to an embodiment of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] Partial discharge (PD) refers to a brief current path that occurs in a part of the insulation system of high-voltage electrical equipment when the electric field strength locally reaches or exceeds the breakdown field strength of the insulation material, but is not sufficient to cause complete breakdown of the entire insulation system.

[0020] Pooling is a commonly used downsampling technique in Convolutional Neural Networks (CNNs). The main purpose of pooling is to reduce the size of the feature map, thereby reducing the complexity of subsequent calculations, while retaining the most important feature information and enhancing the model's invariance and resistance to overfitting.

[0021] Convolution is an important mathematical operation widely used in signal processing, image processing, pattern recognition, and machine learning, playing a central role as a fundamental building block in convolutional neural networks. The basic idea of ​​convolution is to slide a matrix called the kernel across the input data (such as an image or signal), performing element-wise multiplication of local regions of the input data with the kernel and then summing the results to produce a new output. This process can extract various features from the input data.

[0022] According to an embodiment of the present invention, a method for identifying partial discharge in power equipment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0023] Figure 1 This is a flowchart of a partial discharge identification method for power equipment according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0024] Step S102: Obtain the target partial discharge audio signal of the power equipment, wherein the target partial discharge audio signal is obtained based on the original partial discharge audio signal of the power equipment.

[0025] Optionally, to improve the quality of the original partial discharge audio signal in the power equipment, it is necessary to perform a preprocessing process on the original partial discharge audio signal to make the obtained target partial discharge audio signal clearer and improve the quality of the partial discharge audio signal, which is beneficial for subsequent analysis and identification of partial discharge types.

[0026] In one optional embodiment, when the target partial discharge audio signal includes a first partial discharge audio signal, a second partial discharge audio signal, and a third partial discharge audio signal, acquiring the target partial discharge audio signal of the power equipment includes: performing wavelet denoising processing on the original partial discharge audio signal of the power equipment to obtain the first partial discharge audio signal; performing frequency domain feature transformation on the first partial discharge audio signal using a short-time Fourier transform method to obtain the second partial discharge audio signal; and performing time-frequency domain feature transformation on the second partial discharge audio signal using a Mel-frequency conversion method to obtain the third partial discharge audio signal.

[0027] Optionally, wavelet denoising is an effective method for suppressing noise in the original partial discharge audio signal because it can analyze the signal at multiple scales and selectively suppress noise without damaging important features of the signal. Specifically, firstly, a three-level wavelet decomposition of the original partial discharge audio signal can be performed using a wavelet basis, as follows: ;in, This represents the original partial discharge audio signal. This represents the low-frequency approximation coefficients, used to reflect the characteristics of the original time-domain partial discharge audio signal in the low-frequency global portion. This represents the high-frequency detail coefficients of the first layer. This represents the high-frequency detail coefficients of the second layer. This represents the high-frequency detail coefficients of the third layer, where, These reflect the characteristics of the original time-domain partial discharge audio signal in different high-frequency ranges. Next, during wavelet denoising, thresholding is typically applied to the high-frequency detail coefficients. Using the Stein unbiased risk estimation method, the threshold for any layer of high-frequency detail coefficients is determined as follows: ,in, The threshold representing the high-frequency detail coefficients of any layer. This represents the noise standard deviation of the high-frequency detail coefficients at any layer. MAD represents the median absolute deviation, j represents the index of the high-frequency detail coefficients at any layer, and N represents the number of sampling points. This represents the high-frequency detail coefficients at any level. For example, assuming N=480000, 2ln480000≈2×13.08≈5.12, if... If MAD=0.1, then =0.1 / 0.6745×5.12≈0.759, this threshold can effectively filter out The noise components in the data. Furthermore, the high-frequency detail coefficients after wavelet decomposition at any level are processed as follows: Soft thresholding is used for processing. Among them, Dj Let represent any high-frequency detail coefficient after soft-thresholding, and sign(·) denote the sign function. This process preserves the effective signal in the original time-domain partial discharge audio signal while suppressing interference signals. Finally, based on the high-frequency detail coefficients after wavelet decomposition at any level after soft-thresholding, the first partial discharge audio signal, i.e., the denoised original partial discharge audio signal, is obtained as follows: ,in, This represents the audio signal of the first partial discharge, reflecting its time-domain characteristics. This represents the first layer of high-frequency detail coefficients after soft-thresholding. This represents the second layer of high-frequency detail coefficients after soft-thresholding. This represents the high-frequency detail coefficients of the third layer after soft thresholding.

[0028] Since partial discharge signals have important characteristics in both the time and frequency domains, and short-time Fourier transform (SFT) can simultaneously capture the time-varying and spectral characteristics of partial discharge signals, a SFT can be performed on the first partial discharge audio signal. The Hanning window function can then be used to segment the first partial discharge audio signal into frames, resulting in multiple frame frequency domain signals. Each frame frequency domain signal represents a signal segment of a predetermined sampling duration obtained from the first partial discharge audio signal. Any frame frequency domain signal can be obtained as follows: ,in, Let n represent the index of any frame of the frequency domain signal, where n = 0, 1, 2. , , Indicates the number of signals in the frame frequency domain. , R represents the preset sampling rate, and R represents the frame shift, which is the number of sampling points between two adjacent frames of frequency domain signal. This represents the window length of the Hanning window, which is the number of sampling points in any frame of the frequency domain signal. This represents the Hanning window function. , This represents the index of the sampling point in the Hanning window function. Representing frequency points, f=0, 1, 2 , The predetermined sampling duration is obtained by the ratio of the Hanning window length to the preset sampling rate. Multiple frame frequency domain signals are obtained by obtaining any one frame frequency domain signal, where the second partial discharge audio signal includes multiple frame frequency domain signals. For example, assuming a preset sampling rate of 480000Hz, a Hanning window length of 512, and a frame shift of 256, the number of frame frequency domain signals can be calculated to be 1876, containing 257 frequency points. The frequency domain feature X corresponding to the second partial discharge audio signal can then be obtained. The dimension representing the frequency domain characteristics corresponding to the second partial discharge audio signal.

[0029] To further enhance the time-frequency domain characteristics of the partial discharge audio signal, especially suitable for important frequency ranges in the human auditory system (sensitive to low frequencies, insensitive to high frequencies), the second partial discharge audio signal is converted into a third partial discharge audio signal using Mel-frequency spectrum analysis. Mel-frequency spectrum analysis, designed based on the frequency perception characteristics of the human ear, is a frequency- and time-dependent characteristic representation that better reflects the changes in the energy distribution of the partial discharge audio signal with frequency and time. The third partial discharge audio signal is obtained as follows: First, Mel-frequency scaling is used to convert the frequency... Convert to Mel frequency In, among them, The Mel scale is a scale designed based on the human ear's perceptual characteristics of different frequencies of sound. Next, multiple Mel filters are constructed, where the frequency response of any filter satisfies the following function: ,in, For any Mel filter, k represents the index of any Mel filter. This represents the center frequency of any Mel filter, calculated based on a uniform distribution of the Mel scale, covering a frequency range from 0 to 24 kHz. This represents the center frequency of the preceding Mel filter for any given Mel filter. This represents the center frequency of the next Mel filter after any Mel filter. Finally, the frequency domain signal of any frame is... With any Mel filter Perform convolution processing to obtain any Mel spectrum ,in, Multiple Mel spectra are obtained by using the same method as obtaining any Mel spectrum, where the third partial discharge audio signal includes multiple Mel spectra. For example, assuming 40 Mel filters are constructed, based on the above calculation, the number of frame frequency domain signals is 1876, and the time-frequency domain features M corresponding to the third partial discharge audio signal can be obtained. The dimension representing the time-frequency domain characteristics corresponding to the third partial discharge audio signal.

[0030] Step S104: Multi-scale feature extraction is performed on the target partial discharge audio signal to obtain the multi-scale features of the target partial discharge audio signal. The multi-scale features represent the features of the target partial discharge audio signal obtained at multiple scales, and the multiple scales correspond one-to-one with the multiple convolution kernel sizes.

[0031] Optionally, convolutional neural networks can identify features at different scales using convolutional kernels of different sizes. In partial discharge audio signal recognition, convolutional kernels of different sizes slide across different positions of the signal, performing feature detection on segments of the partial discharge audio signal through dot product operations. Through convolution operations at different scales, each branch generates its own feature map. These feature maps are then concatenated to form a multi-scale feature map, i.e., multi-scale features. This signal processing strategy can effectively capture the multi-level, multi-scale features of partial discharge signals, making them more accurate, reliable, and practical in power equipment condition monitoring.

[0032] In one optional embodiment, feature extraction is performed on the target partial discharge audio signal to obtain multi-scale features of the target partial discharge audio signal, including: performing convolution processing on the target partial discharge audio signal to obtain an output feature map of the target partial discharge audio signal; performing pooling processing on the output feature map to obtain a pooled output feature map, wherein the pooling processing is used to adjust the spatial size of the output feature map; performing feature concatenation processing on the pooled output feature map to obtain a concatenated output feature map; and obtaining multi-scale features based on the concatenated output feature map.

[0033] Optionally, deep learning is used to process key points in the partial discharge audio signal, aiming to capture multi-scale features in the signal by using convolutional kernels of different sizes, thereby improving the accuracy of partial discharge type identification. Table 1 shows the core parameters in an optional process of determining the output feature map after pooling according to an embodiment of the present invention. First, when the target partial discharge audio signal includes multiple signals, such as a first partial discharge audio signal, a second partial discharge audio signal, and a third partial discharge audio signal, the output feature map of each target partial discharge audio signal is obtained in the same way. Three convolutional branches (corresponding to the first, second, and third partial discharge audio signals) are designed, and each branch uses a convolutional kernel of a different size. For example, the convolutional kernel of the small-scale branch is 1×3, the convolutional kernel of the medium-scale branch is 1×7, and the convolutional kernel of the large-scale branch is 1×11. These kernels slide across the target partial discharge audio signal to perform a dot product operation to capture the local patterns of the target partial discharge audio signal. Each convolutional kernel corresponds to a specific scale level. The small-scale kernel focuses on high-frequency details, the medium-scale kernel captures mid-frequency patterns, and the large-scale kernel focuses on low-frequency global trends. Through two rounds of convolution, the number of channels increases after each round, for example, from 64 in the first round to 128 in the second round. Each branch can generate a deeper feature representation, resulting in the output feature map of the target partial discharge audio signal. Next, after each round of convolution, max pooling is performed using a 1×2 pooling window with a stride of 2. This helps reduce the size of the output feature map and the computational cost, while preserving important feature information by selecting the maximum value in a local region. Pooling compresses the size of the output feature map, resulting in a pooled output feature map. For example, the size of the output feature map changes from 128×1×120000 to 128×1×30000 for the small-scale branch, from 128×257×1876 to 128×257×469 for the medium-scale branch, and from 128×40×1876 to 128×40×469 for the large-scale branch. The size is uniformly 128×H×W (128 channels, where H represents the height of the pooled output feature map and W represents the width of the pooled output feature map).

[0034] Table 1

[0035]

[0036] The pooled output feature map of any round of convolution corresponding to the partial discharge audio signal of each branch target can be obtained in the following way: ,in, Let represent the pooled output feature map of any convolution round corresponding to the partial discharge audio signal of each branch target, where i represents the index of any input channel of any convolution round, and j represents the index of the one-dimensional position of the output feature map of any convolution round. This represents a local region of the output feature map for any round of convolution. Using the method of obtaining the pooled output feature map after any round of convolution, the pooled output feature maps corresponding to the two rounds of convolution for the target partial discharge audio signal of each branch are obtained. After each branch completes two rounds of convolution and pooling processing, the two pooled output feature maps corresponding to the target partial discharge audio signal of each branch are concatenated along the channel dimension to integrate multi-level information from small scale to large scale. The concatenated output feature map can be obtained as follows: ,in, This represents the concatenated output feature map. This represents the feature map after two rounds of pooling for the small-scale branch. This represents the feature map after two rounds of pooling in the mesoscale branch. This represents the feature map after two rounds of pooling for the large-scale branch. For example, the dimensionality of the concatenated feature map can be obtained through the above process. (384 channels, H represents the height of the stitched feature map, and W represents the width of the stitched feature map).

[0037] Finally, the concatenated feature maps can be batch normalized to obtain multi-scale features in the following way: ,in, Representing multi-scale features, This represents the mean of the output feature maps after splicing within a predetermined batch. This indicates the number of output feature maps after stitching within a predetermined batch. This represents the index of any concatenated output feature map within a predetermined batch. , This represents the first learnable parameter. This represents the second learnable parameter. It is obtained through training and optimization of a convolutional neural network. This represents a preset small value. For example, the dimension of multi-scale features can be obtained through the above process. (384 channels, where H represents the height of the multi-scale feature and W represents the width of the multi-scale feature). The obtained multi-scale features can reflect not only the detailed features of the target partial discharge audio signal (such as the narrow pulse of corona discharge), but also the macroscopic structural features of the signal (such as the low-frequency trend of surface discharge) and the periodic features of the signal (such as the pulse interval of floating potential discharge).

[0038] In one optional embodiment, convolution processing is performed on the target partial discharge audio signal to obtain an output feature map of the target partial discharge audio signal, including: performing two rounds of convolution processing on the target partial discharge audio signal, and obtaining the output feature map of any round of convolution corresponding to the target partial discharge audio signal in the following manner:

[0039] ;

[0040] in, Let m represent the output feature map of any round of convolution; i represent the index of any input channel of any round of convolution; and j represent the index of the position of the output feature map of any round of convolution. The number of input channels in any round of convolution is represented by K, where K represents the kernel size of the target partial discharge audio signal. The weight parameters represent the weights of any convolution kernel. This represents the index of the position of any convolution kernel in the row direction; This represents the index of the column direction of any convolution kernel. This represents the bias parameter for any round of convolution. This represents the activation function. The input feature map of any round of convolution represents a local region, where the input feature map is obtained based on the target partial discharge audio signal; the output feature maps of the target partial discharge audio signal are obtained by using the method of obtaining the output feature map of any round of convolution; wherein the output feature map of the target partial discharge audio signal includes the output feature maps of the target partial discharge audio signal corresponding to the two rounds of convolution.

[0041] Optionally, convolution is one of the core operations in convolutional neural networks. In signal processing, such as the identification of partial discharge signals, convolution operations can automatically learn and extract meaningful features from the input signal. These features are often key information in determining the type of partial discharge signal. When the target partial discharge audio signal includes multiple signals, such as a first partial discharge audio signal, a second partial discharge audio signal, and a third partial discharge audio signal, the output feature map of each target partial discharge audio signal is obtained in the same way as described above.

[0042] Step S106: Perform feature fusion on the multi-scale features to obtain the fused features of the target partial discharge audio signal;

[0043] Optionally, feature fusion plays a crucial role in processing multimodal or multi-scale data, including audio. For the analysis of partial discharge audio signals, the goal of multi-scale feature fusion is to integrate different features extracted from small-scale, medium-scale, and large-scale branches to obtain a fused feature that comprehensively represents multiple aspects of the signal. The fused feature contains richer and more comprehensive information, helping to more accurately distinguish different types of partial discharges.

[0044] In one optional embodiment, feature fusion is performed on multi-scale features to obtain fused features of the target partial discharge audio signal, including: splitting the multi-scale features to obtain time-domain correlated features and frequency-domain correlated features; compressing the time-domain correlated features and frequency-domain correlated features respectively to obtain time-domain sub-features and frequency-domain sub-features; and obtaining fused features based on the time-domain sub-features and frequency-domain sub-features.

[0045] Optionally, firstly, the multi-scale features are divided into two categories based on temporal correlation and frequency correlation. Temporal-related features correspond to the results extracted by small-scale convolutions, focusing on the details of temporal pulses; frequency-related features come from the outputs of medium- and large-scale convolutions, focusing on the dominant frequency distribution information in the frequency domain. For example, through this splitting process, the dimensions of the multi-scale features... Decomposed into time-domain related feature dimensions Frequency domain related feature dimensions Next, the time-domain correlation features and frequency-domain correlation features are respectively derived from... The two-dimensional space dimension is compressed into L= By using a one-dimensional length, we can obtain time-domain and frequency-domain sub-features, thus simplifying computational complexity and facilitating subsequent calculations. For example, time-domain sub-features can be obtained as follows: ,in, The time-domain sub-features are represented; the frequency-domain sub-features are obtained as follows: ,in, The temporal sub-features are represented. Finally, based on the obtained temporal and frequency sub-features, the resulting fused features help to more accurately distinguish various types of partial discharges.

[0046] In one optional embodiment, obtaining fused features based on temporal sub-features and frequency sub-features includes: determining a correlation matrix, wherein the correlation matrix is ​​used to quantify the correlation between temporal sub-features and frequency sub-features; obtaining temporal attention weights and frequency attention weights based on the correlation matrix, wherein the temporal attention weights are used to quantify the importance of temporal sub-features, and the frequency attention weights are used to quantify the importance of frequency sub-features; performing a weighted operation on the temporal sub-features and temporal attention weights to obtain weighted temporal features; performing a weighted operation on the frequency sub-features and frequency attention weights to obtain weighted frequency features; and obtaining fused features based on the weighted temporal features and weighted frequency features.

[0047] Optionally, firstly, cosine similarity is used to calculate the correlation between time-domain sub-features and frequency-domain sub-features to obtain the correlation matrix, which is obtained as follows: ,in, Let represent the correlation matrix, where i represents the index of any row element in the correlation matrix, j represents the index of any column element in the correlation matrix, and represents the correlation strength between any channel of the time-domain sub-feature and the corresponding channel of the frequency-domain sub-feature, with a value range of [-1, 1]. The L2 norm represents the square root of the sum of squares of all elements in the correlation matrix. Preset a small value. For example, the dimensions of the correlation matrix can be obtained through the above calculation. .

[0048] Next, by normalizing the row and column dimensions of the correlation matrix, time-domain attention weights and frequency-domain attention weights are generated. The attention weight of any row element to any column element in the correlation matrix is ​​obtained as follows: ,in, The attention weights represent the attention weights of any row element to any column element in the correlation matrix, where k represents the index of any column element and m represents the number of columns in the correlation matrix. The temporal attention weights are obtained by calculating the attention weights of any row element to any column element in the correlation matrix. For example, the dimension of the temporal attention weights can be obtained using the above calculation method. The attention weights of any column element to any row element in the correlation matrix are obtained as follows: ,in, The frequency domain attention weights are obtained by representing the attention weight of any column element to any row element in the correlation matrix, where h represents the index of any row element in the correlation matrix. For example, the dimension of the frequency domain attention weights can be obtained using the above calculation method. .

[0049] Subsequently, using the obtained temporal and frequency-domain attention weights, the temporal and frequency-domain sub-features are weighted to emphasize features highly correlated with the target partial discharge type, thereby reducing the influence of redundant information and enhancing the representational power of effective features. The weighted temporal features are obtained as follows: , Indicates the weighted time-domain features, The temporal attention weights are represented; the weighted frequency domain features are obtained as follows: , Represents the weighted frequency domain characteristics. This represents the frequency domain attention weights. For example, the dimension of the weighted time-domain features can be obtained using the above calculation method. Dimensions of weighted frequency domain features Finally, the weighted time-domain features and weighted frequency-domain features are reshaped into a two-dimensional spatial dimension, and then spliced ​​and fused along the channels to obtain the fused features. For example, the fused features are obtained through the above calculation method: ,in, Indicates fusion characteristics, , Dimensions of fused features The fusion features obtained through the above methods can take into account both time-domain pulse details and frequency-domain dominant frequency distribution, laying the foundation for accurate determination of target partial discharge. Table 2 shows the core parameters in an optional process of determining fusion features according to an embodiment of the present invention. As can be seen from the table, the finally generated fusion features can integrate multi-scale and cross-domain signal characteristics, providing a feature representation with rich information and strong anti-interference ability for partial discharge signal identification.

[0050] Table 2

[0051]

[0052] Step S108: Based on the fusion characteristics, determine the target partial discharge type of the power equipment.

[0053] Optionally, classifying partial discharge types based on fusion features is key to achieving accurate identification of partial discharges in power equipment. This process can improve the accuracy of partial discharge identification, enhance anti-interference capabilities, and reduce the cost of partial discharge identification. Table 3 shows the characteristics of various optional partial discharge audio signals according to an embodiment of the present invention. The table reveals key information about the types of partial discharges in power equipment and their audio signal characteristics, providing a theoretical basis for assessing the health status of power equipment. By identifying and classifying partial discharge types, the potential threat posed by target partial discharge types to power equipment can be assessed, providing a scientific basis for the maintenance and repair of power equipment and guiding the safe operation of the power system.

[0054] Table 3

[0055]

[0056] In one optional embodiment, determining the target partial discharge type of the power equipment based on fused features includes: performing global average pooling on the fused features to obtain an average feature vector; obtaining an original score vector based on the average feature vector, wherein any element in the original score vector represents the predicted score of the target partial discharge audio signal belonging to any partial discharge type, and multiple elements in the original score vector correspond one-to-one with multiple partial discharge types; normalizing the original score vector to obtain the probability corresponding to each of the multiple partial discharge types; and determining the target partial discharge type of the power equipment based on the probabilities corresponding to each of the multiple partial discharge types.

[0057] Optionally, global average pooling is a commonly used deep learning technique to reduce the dimensionality of fused features while preserving the channel information of the fused features. The average feature value of any channel is obtained as follows: V[c] = Where V[c] represents the average eigenvalue of any channel, and c represents the number of channels. The dimension representing the fused features. Let represent the fusion feature of any channel, where 'i' represents the index of any height position in the dimension of the fusion feature, and 'j' represents the index of any width position in the dimension of the fusion feature. By obtaining the average feature value of any channel, the average feature values ​​corresponding to each of the multiple channels are obtained, thus yielding the average feature vector. For example, the dimension of the average feature vector can be obtained through the above calculation method. .

[0058] After obtaining the average feature vector, a fully connected layer and an activation function are used to obtain the original score vector. The fully connected layer is a component in the neural network used to map the average feature vector to the output space of the deep learning model. The activation function enables the deep learning model to learn and represent complex functional relationships. Multiple elements in the original score vector correspond to the predicted scores of the four identified partial discharge types (such as corona discharge, surface discharge, internal discharge, and floating potential discharge). Table 4 shows the core parameters in an optional process of determining the target partial discharge type according to an embodiment of the present invention. As can be seen from the table, the original score vector can be obtained in the following way: First, a nonlinear transformation is performed using an activation function to map the 384-dimensional average feature vector to a 128-dimensional activated average feature vector, A = ReLU(Z), Z = Where A represents the average feature vector after activation; the dimension is... ReLU represents the activation function. This represents the weight matrix of the fully connected layer; its dimension is... , Represents the bias vector; dimension is Next, the original score vector is output using the fully connected output layer, L= Where L represents the original score vector; the dimension is , This represents the output fully connected layer weight matrix; its dimension is... , This represents the output bias vector; its dimension is... The original score vector is normalized and converted into a probability distribution to obtain the probability corresponding to each of the various partial discharge types. The maximum probability value among the various partial discharge types is compared with a preset probability threshold. If the maximum probability value exceeds the preset probability threshold, the partial discharge type corresponding to the maximum probability value is determined as the target partial discharge type; otherwise, the original partial discharge audio signal is re-acquired and partial discharge identification is performed again.

[0059] Table 4

[0060]

[0061] In one optional embodiment, the original score vector is normalized to obtain the probabilities corresponding to each of the various partial discharge types, including: normalizing the original score vector to obtain the probability of any partial discharge type in the following manner: ;in, This represents the probability of any partial discharge type, and k represents the index of any partial discharge type. This represents any element in the original score vector. It represents the sum of multiple elements in the original score vector; by obtaining the probability of any partial discharge type, the probabilities corresponding to each of the multiple partial discharge types are obtained.

[0062] Optionally, the elements in the original score vector are normalized and transformed into an easily interpretable probabilistic form. The probability value for each partial discharge type reflects the most likely corresponding partial discharge type, providing a basis for decision-making regarding the identification and classification of partial discharge types in power equipment. Through the probability output, the likelihood of each discharge type can be intuitively understood, enabling appropriate measures to be taken to ensure the safety and stability of the power system.

[0063] Through the above steps S102 to S108, the goal of accurately determining the type of partial discharge of power equipment can be achieved by performing multi-scale feature extraction and feature fusion on the acquired target partial discharge audio signal, thereby improving the technical effect of partial discharge identification accuracy. This solves the technical problem of low accuracy in identifying partial discharge of power equipment due to incomplete consideration of factors when facing complex and ever-changing power field environments.

[0064] Based on the above embodiments and optional embodiments, the present invention proposes an optional implementation method. Figure 2This is a schematic diagram of an optional hardware system for power equipment according to an embodiment of the present invention. The hardware system consists of two parts: a signal acquisition terminal and an edge computing terminal, used to acquire, transmit, calculate, and identify partial discharge audio signals. Table 5 shows the specific composition of an optional hardware system according to an embodiment of the present invention. The signal acquisition terminal is deployed next to the power equipment (such as the top of a transformer tank or the door of a switchgear), and the edge computing terminal is deployed in the control room. The two are connected via 5G or Ethernet to achieve a closed loop of "on-site acquisition - remote calculation - real-time monitoring".

[0065] In the process of model development and training, the construction of the dataset is a crucial first step. Constructing a high-quality and diverse dataset is a prerequisite for achieving accurate partial discharge audio signal recognition. The following are the specific steps and details of an optional dataset construction: (1) Select three types of typical power equipment (110kV transformer, 10kV switchgear, 35kV cable terminal), simulate four types of partial discharge faults in the laboratory, collect samples, corona discharge: apply 110kV voltage to the tip of the transformer high-voltage bushing to generate tip corona; surface discharge: coat the surface of the switchgear insulator with oil, apply 10kV voltage to generate surface flashover discharge; internal discharge: inject bubbles (0.5-1mm in diameter) into the transformer oil, apply 110kV voltage to generate oil bubble discharge; floating potential discharge: set a loose metal sheet (floating potential 1-5kV) in the switchgear to generate floating discharge. (2) Dataset size and distribution: A total of 10,000 unlabeled samples (for model training) and 1,000 labeled samples (for fine-tuning and testing) were collected. Table 6 shows an optional sample distribution method according to an embodiment of the present invention. (3) Sample labeling method: The power equipment insulation expert (with more than 5 years of experience in partial discharge detection) adopted "double-blind labeling": Expert A labeled the sample type (0-corona, 1-surface, 2-internal, 3-suspended); Expert B reviewed the labeling results and judged the divergent samples (<5%) by "time domain pulse shape + frequency domain dominant frequency"; the final labeling accuracy rate was ≥99.5%, ensuring the quality of the dataset.

[0066] The preprocessing stage is the first step in preparing the model data input, and its effectiveness directly affects the model's final recognition performance. After denoising, the pulse details (width ≈ 8μs) of the time-domain signal are still preserved, and the 50Hz power frequency harmonic interference (50Hz, 100Hz) of the frequency-domain signal is completely eliminated. The energy proportion of the effective feature region (100-150kHz frequency band) of the Mel spectrum is increased from 35% to 82%, providing high-quality input for subsequent feature extraction. The model training environment in the hardware system can be configured as follows: hardware includes a graphics processor (NVIDIA RTX4090, 48GB VRAM), a central processing unit (Intel i9-12900K), and memory (64GB DDR4); software includes an operating system (Ubuntu 20.04), a deep learning framework (PyTorch 2.0), and data processing libraries (NumPy 1.24, SciPy 1.10).

[0067] Table 5

[0068]

[0069] Table 6

[0070]

[0071] Table 7 shows an optional core training parameter for the model according to an embodiment of the present invention. After 200 rounds of training (pre-training) + 50 rounds (fine-tuning), the accuracy of the training set reached 99.8%, and the accuracy of the test set reached 98.9%. The losses of both the training set and the test set converged stably (the test set loss decreased to 0.12), with no overfitting. Finally, on-site testing and verification were conducted. Three 110kV transformers and five 10kV switchgear cabinets of a 220kV substation were selected as test objects. The test lasted for one month, and the core test indicators are as follows: recognition accuracy, classification accuracy and average accuracy of four types of partial discharge; inference speed, single recognition time of the edge computing unit; anti-interference capability, recognition accuracy under different signal-to-noise ratios (SNR=8dB / 12dB / 16dB). Table 8 shows an optional on-site test result according to an embodiment of the present invention. The test results show that this embodiment can maintain high recognition accuracy and fast inference speed in the complex environment of the power field, fully meeting the needs of on-site monitoring.

[0072] Table 7

[0073]

[0074] Table 8

[0075]

[0076] Based on the above embodiments and optional embodiments, the present invention proposes an optional implementation method. Figure 3 This is a flowchart of an optional partial discharge identification method for power equipment according to an embodiment of the present invention, such as... Figure 3 As shown, the method includes:

[0077] S1: Acquire the original partial discharge audio signal of the power equipment, i.e., partial discharge (PD) audio signal acquisition. The specific implementation process is the same as the previous embodiment, and will not be repeated here.

[0078] S2: Preprocess the original partial discharge audio signal to obtain the target partial discharge audio signal. The preprocessing includes wavelet denoising, frequency domain feature transformation and time-frequency domain feature transformation, i.e., denoising and multi-domain transformation. The specific implementation process is the same as in the previous embodiment, and will not be repeated here.

[0079] S3: Perform multi-scale convolution processing on the target partial discharge audio signal to obtain the multi-scale features of the target partial discharge audio signal. The specific implementation process is the same as the previous embodiment, and will not be repeated here.

[0080] S4: Decompose the multi-scale features to obtain time-domain correlated features and frequency-domain correlated features, and perform weighted fusion of the time-domain correlated features and frequency-domain correlated features to obtain fused features. The specific implementation process is the same as the previous embodiment, and will not be repeated here.

[0081] S5: Perform global average pooling (GAP), fully connected layer and normalization (softmax) on the fused features to obtain the probabilities corresponding to various partial discharge types. The specific implementation process is the same as the previous embodiment, and will not be repeated here.

[0082] S6: Compare the maximum probability value among multiple partial discharge types with a preset probability threshold. If the maximum probability value exceeds the preset probability threshold, the partial discharge type corresponding to the maximum probability value is determined as the target partial discharge type. Otherwise, return to step S1 and re-acquire the original partial discharge audio signal for partial discharge identification. The specific implementation process is the same as in the aforementioned embodiment and will not be repeated here. Table 9 shows a comparison of optional partial discharge identification accuracy results according to an embodiment of the present invention. The data in the table is based on a power equipment partial discharge audio dataset (containing 500 samples of each of 4 types of partial discharge, with a signal-to-noise ratio (SNR) of 20dB). Data analysis from the table shows that the related technologies using fixed-size convolution kernels (such as 3×3 / 5×5) can only cover 1-2 types of partial discharge signal scales, resulting in a difference in identification accuracy of more than 25% between different types of partial discharge. This embodiment employs three parallel multi-scale convolutions to accurately match the scale characteristics of four types of partial discharge signals: a small-scale kernel (1×3) fully extracts the high-frequency narrow pulse features of corona discharge (main frequency 10-50kHz, pulse width <10μs), improving the corona discharge recognition accuracy to over 96%; a medium-scale kernel (1×7) accurately captures the mid-frequency periodic features of suspended potential discharge (main frequency 20-80kHz, stable pulse interval), improving the suspended potential discharge recognition accuracy to over 95%; and a large-scale kernel (1×11) comprehensively covers the low-frequency global trend of surface discharge (main frequency 5-20kHz, strong signal amplitude), improving the surface discharge recognition accuracy to over 94%. Simultaneously, through multi-scale feature splicing, the correlation between high-frequency details (main frequency 50-200kHz) and low-frequency energy distribution of internal discharge is preserved, improving the internal discharge recognition accuracy to over 97%.

[0083] Table 9

[0084]

[0085] This embodiment also provides a partial discharge identification device for power equipment, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0086] According to an embodiment of the present invention, an apparatus embodiment for implementing the above-described partial discharge identification method for power equipment is also provided. Figure 4 This is a schematic diagram of the structure of a partial discharge identification device for power equipment according to an embodiment of the present invention, as shown below. Figure 4As shown, the partial discharge identification device for the aforementioned power equipment includes: a target partial discharge audio signal acquisition module 400, a multi-scale feature determination module 402, a fusion feature determination module 404, and a target partial discharge type determination module 406, wherein:

[0087] The target partial discharge audio signal acquisition module 400 is used to acquire the target partial discharge audio signal of the power equipment, wherein the target partial discharge audio signal is obtained based on the original partial discharge audio signal of the power equipment;

[0088] The multi-scale feature determination module 402 is connected to the target partial discharge audio signal acquisition module 400. It is used to extract multi-scale features from the target partial discharge audio signal to obtain the multi-scale features of the target partial discharge audio signal. The multi-scale features represent the features of the target partial discharge audio signal obtained at multiple scales, and the multiple scales correspond one-to-one with the multiple convolution kernel sizes.

[0089] The fusion feature determination module 404 is connected to the multi-scale feature determination module 402 and is used to perform feature fusion on multi-scale features to obtain the fusion features of the target partial discharge audio signal.

[0090] The target partial discharge type determination module 406 is connected to the fusion feature determination module 404 and is used to determine the target partial discharge type of the power equipment based on the fusion features.

[0091] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0092] It should be noted that the target partial discharge audio signal acquisition module 400, multi-scale feature determination module 402, fusion feature determination module 404, and target partial discharge type determination module 406 mentioned above correspond to steps S102 to S108 in the embodiments. The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run in a computer terminal.

[0093] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0094] The aforementioned partial discharge identification device for power equipment may further include a processor and a memory. The target partial discharge audio signal acquisition module 400, the multi-scale feature determination module 402, the fusion feature determination module 404, and the target partial discharge type determination module 406 are all stored in the memory as program modules. The processor executes the aforementioned program modules stored in the memory to realize the corresponding functions.

[0095] The processor contains a core that retrieves the corresponding program modules from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0096] According to an embodiment of this application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein, when the program runs, it controls the device containing the non-volatile storage medium to execute any of the aforementioned partial discharge identification methods for power equipment.

[0097] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals, and the non-volatile storage medium includes stored programs.

[0098] Optionally, a program may be included to control the device containing the non-volatile storage medium to execute any of the steps of the above-mentioned method for determining the health status of a new energy vehicle battery during program execution.

[0099] According to an embodiment of this application, an embodiment of a processor is also provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the aforementioned partial discharge identification methods for power equipment.

[0100] According to an embodiment of this application, an embodiment of a computer program product is also provided. Optionally, in this embodiment, the computer program product includes a computer program that, when executed by a processor, implements the partial discharge identification method steps of any of the above-described power equipment.

[0101] This invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the method for determining the health status of a new energy vehicle battery as described above.

[0102] The order of the above embodiments of the present invention is merely for description and does not represent the superiority or inferiority of the embodiments.

[0103] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of modules described above can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules 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, or indirect coupling or communication connection between modules, and may be electrical or other forms.

[0105] The modules described above as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0106] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0107] If the aforementioned integrated modules are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned non-volatile storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0108] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying partial discharge in power equipment, characterized in that, include: Acquire a target partial discharge audio signal of a power device, wherein the target partial discharge audio signal is obtained based on the original partial discharge audio signal of the power device; Multi-scale feature extraction is performed on the target partial discharge audio signal to obtain the multi-scale features of the target partial discharge audio signal, wherein the multi-scale features represent the features of the target partial discharge audio signal obtained at multiple scales, and the multiple scales correspond one-to-one with multiple convolution kernel sizes; The multi-scale features are fused to obtain the fused features of the target partial discharge audio signal; Based on the fusion features, the target partial discharge type of the power equipment is determined.

2. The method according to claim 1, characterized in that, When the target partial discharge audio signal includes a first partial discharge audio signal, a second partial discharge audio signal, and a third partial discharge audio signal, acquiring the target partial discharge audio signal of the power equipment includes: The original partial discharge audio signal of the power equipment is subjected to wavelet denoising processing to obtain the first partial discharge audio signal; The first partial discharge audio signal is subjected to frequency domain feature transformation using the short-time Fourier transform method to obtain the second partial discharge audio signal; The second partial discharge audio signal is subjected to time-frequency domain feature transformation using the Mel spectrum conversion method to obtain the third partial discharge audio signal.

3. The method according to claim 1, characterized in that, The step of extracting features from the target partial discharge audio signal to obtain multi-scale features of the target partial discharge audio signal includes: The target partial discharge audio signal is convolved to obtain the output feature map of the target partial discharge audio signal; The output feature map is subjected to pooling processing to obtain a pooled output feature map, wherein the pooling processing is used to adjust the spatial size of the output feature map; The pooled output feature map is then subjected to feature concatenation to obtain a concatenated output feature map. The multi-scale features are obtained based on the spliced ​​output feature map.

4. The method according to claim 3, characterized in that, The step of performing convolution processing on the target partial discharge audio signal to obtain the output feature map of the target partial discharge audio signal includes: The target partial discharge audio signal is subjected to two rounds of convolution processing, and the output feature map of any round of convolution corresponding to the target partial discharge audio signal is obtained in the following way: ； in, Let m represent the index of any round of convolution; i represent the index of any input channel of any round of convolution; and j represent the index of the position of the output feature map of any round of convolution. K represents the number of input channels in any round of convolution, and K represents the kernel size of the target partial discharge audio signal. The weight parameters represent the weights of any convolution kernel. This represents the position index of any of the convolution kernels in the row direction; This represents the position index of any of the convolution kernels in the column direction; This represents the bias parameter of any of the convolution rounds. This represents the activation function. This represents a local region of the input feature map of any round of convolution, wherein the input feature map is obtained based on the target partial discharge audio signal; By obtaining the output feature map of any one round of convolution, the output feature maps corresponding to the two rounds of convolution of the target partial discharge audio signal are obtained respectively. The output feature map of the target partial discharge audio signal includes the output feature maps corresponding to the two rounds of convolution of the target partial discharge audio signal.

5. The method according to claim 1, characterized in that, The process of fusing the multi-scale features to obtain the fused features of the target partial discharge audio signal includes: The multi-scale features are split to obtain time-domain correlated features and frequency-domain correlated features; The time-domain correlation features and the frequency-domain correlation features are compressed respectively to obtain time-domain sub-features and frequency-domain sub-features; The fused feature is obtained based on the time-domain sub-features and the frequency-domain sub-features.

6. The method according to claim 5, characterized in that, The process of obtaining the fused features based on the time-domain sub-features and the frequency-domain sub-features includes: Determine a correlation matrix, wherein the correlation matrix is ​​used to quantify the correlation between the time-domain sub-features and the frequency-domain sub-features; Based on the correlation matrix, temporal attention weights and frequency domain attention weights are obtained, wherein the temporal attention weights are used to quantify the importance of the temporal sub-features, and the frequency domain attention weights are used to quantify the importance of the frequency domain sub-features. The temporal sub-features and the temporal attention weights are weighted to obtain weighted temporal features; The frequency domain sub-features and the frequency domain attention weights are weighted to obtain weighted frequency domain features; The fused features are obtained based on the weighted time-domain features and the weighted frequency-domain features.

7. The method according to any one of claims 1 to 6, characterized in that, Determining the target partial discharge type of the power equipment based on the fusion features includes: The fused features are subjected to global average pooling to obtain an average feature vector; Based on the average feature vector, an original score vector is obtained, wherein any element in the original score vector represents the predicted score of the target partial discharge audio signal belonging to any partial discharge type, and multiple elements in the original score vector correspond one-to-one with multiple partial discharge types. The original score vector is normalized to obtain the probabilities corresponding to each of the various partial discharge types; Based on the probabilities corresponding to each of the various partial discharge types, the target partial discharge type of the power equipment is determined.

8. The method according to claim 7, characterized in that, The normalization process of the original score vector to obtain the probabilities corresponding to each of the various partial discharge types includes: The probability of any partial discharge type is obtained by normalizing the original score vector in the following manner: ； in, represents the probability of any partial discharge type, and k represents the index of any partial discharge type; This represents any element in the original score vector. This represents the sum of multiple elements in the original score vector; The probabilities corresponding to each of the multiple partial discharge types are obtained by using the method of obtaining the probability of any one of the partial discharge types.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores multiple instructions, which are adapted to be loaded by a processor and executed by the partial discharge identification method for power equipment according to any one of claims 1 to 8.

10. An electronic device, characterized in that, The device includes one or more processors and a memory, the memory being used to store one or more programs for execution, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the partial discharge identification method for power equipment according to any one of claims 1 to 8.

Images