Partial discharge fault identification method and device, storage medium and electronic equipment

Abstract

The application discloses a partial discharge fault identification method and device, a storage medium and an electronic device. The method comprises the following steps: obtaining a frequency spectrum data sequence set based on a signal sample set of a gas insulated metal enclosed combined electric appliance, wherein the frequency spectrum data sequence set is used for representing a partial discharge feature, and the signal sample set is obtained by processing a partial discharge signal of the gas insulated metal enclosed combined electric appliance; obtaining a wavelet feature map set of the gas insulated metal enclosed combined electric appliance based on the frequency spectrum data sequence set; and inputting the wavelet feature map set into a preset parallel convolutional neural network for processing, so as to obtain a fault mode identification result of the gas insulated metal enclosed combined electric appliance. The application solves the technical problem of low efficiency of partial discharge fault identification in the related art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power operation and maintenance, in particular to a partial discharge fault identification method and device, a storage medium and an electronic device. BACKGROUND

[0002] As a high-voltage electrical device, gas insulated switchgear (GIS) is widely used in substations of various voltage levels. GIS equipment is a sealed device, and when a fault occurs inside the GIS equipment, the closed external structure and complex internal structure make it difficult to find and eliminate the fault. The commonly used partial discharge signal feature extraction methods include statistical feature parameter method, time domain feature parameter method and frequency domain feature parameter method. Since the on-site partial discharge detection is often in a complex electromagnetic environment, it brings difficulties to the feature information extraction of GIS partial discharge, resulting in the problem of unsatisfactory GIS partial discharge fault identification efficiency.

[0003] At present, no effective solution has been proposed for the above problems. SUMMARY

[0004] The embodiments of the present application provide a partial discharge fault identification method and device, a storage medium and an electronic device to at least solve the technical problem of unsatisfactory partial discharge fault identification efficiency in the related art.

[0005] According to an aspect of the embodiments of the present application, a partial discharge fault identification method is provided, including: obtaining a frequency spectrum data sequence set based on a signal sample set of a gas insulated switchgear, wherein the frequency spectrum data sequence set is used to represent partial discharge characteristics, and the signal sample set is obtained by processing a partial discharge signal of the gas insulated switchgear; obtaining a wavelet feature map set of the gas insulated switchgear based on the frequency spectrum data sequence set; inputting the wavelet feature map set into a preset parallel convolutional neural network for processing to obtain a fault mode identification result of the gas insulated switchgear.

[0006] According to another aspect of the embodiments of the present application, a partial discharge fault identification device is provided, comprising: an acquisition module configured to obtain a spectrum data sequence set based on a signal sample set of a gas insulated metal-enclosed combined electric appliance, wherein the spectrum data sequence set is configured to represent partial discharge characteristics, and the signal sample set is obtained by processing a partial discharge signal of the gas insulated metal-enclosed combined electric appliance; an obtaining module configured to obtain a wavelet feature map set of the gas insulated metal-enclosed combined electric appliance based on the spectrum data sequence set; and an identification module configured to input the wavelet feature map set into a preset parallel convolutional neural network for processing to obtain a fault mode identification result of the gas insulated metal-enclosed combined electric appliance.

[0007] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided, which stores a plurality of instructions adapted to be loaded and executed by a processor to implement any of the partial discharge fault identification methods.

[0008] According to another aspect of the embodiments of the present application, an electronic device is provided, comprising: one or more processors and a memory configured to store one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement any of the partial discharge fault identification methods.

[0009] In the embodiments of the present application, by obtaining a spectrum data sequence set based on a signal sample set of a gas insulated metal-enclosed combined electric appliance, wherein the spectrum data sequence set is configured to represent partial discharge characteristics, and the signal sample set is obtained by processing a partial discharge signal of the gas insulated metal-enclosed combined electric appliance, obtaining a wavelet feature map set of the gas insulated metal-enclosed combined electric appliance based on the spectrum data sequence set, and inputting the wavelet feature map set into a preset parallel convolutional neural network for processing to obtain a fault mode identification result of the gas insulated metal-enclosed combined electric appliance, the purpose of improving the accuracy of fault mode identification by using a parallel convolutional neural network is achieved, the technical effect of improving the efficiency of partial discharge fault identification is realized, and the technical problem of non-ideal partial discharge fault identification efficiency in the related art is solved. BRIEF DESCRIPTION OF DRAWINGS

[0010] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:

[0011] Figure 1 is a flowchart of an optional partial discharge fault identification method provided according to the embodiments of the present application;

[0012] Figure 2 is a schematic flow chart of an optional partial discharge fault identification method according to an embodiment of the present application;

[0013] Figure 3 is a reconfiguration schematic diagram of an optional partial discharge fault identification method according to an embodiment of the present application;

[0014] Figure 4 is a feature map construction schematic diagram of an optional partial discharge fault identification method according to an embodiment of the present application;

[0015] Figure 5 is a wavelet feature schematic diagram of an optional partial discharge fault identification method according to an embodiment of the present application;

[0016] Figure 6 is a pattern recognition schematic diagram of an optional partial discharge fault identification method according to an embodiment of the present application;

[0017] Figure 7 is a schematic diagram of an optional partial discharge fault identification device according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the personnel in the art better understand the present application scheme, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0020] As a high-voltage electrical device, gas insulated switchgear (GIS) is widely used in substations of various voltage levels. GIS equipment is a sealed device, and when a fault occurs inside the GIS equipment, the closed external structure and complex internal structure make it difficult to find and eliminate the fault. Defects formed during production, transportation, installation and long-term operation can cause insulation failure of the GIS equipment, thereby causing partial discharge (PD) phenomenon. According to the public data, particle defects and insulator defects account for as high as 30% in GIS faults. The partial discharge generated by different types of defects also has different degrees of harm to the equipment. The partial discharge detection technology is introduced to regularly monitor the GIS equipment, the discharge signals obtained are subjected to feature extraction research, and the types of insulation faults of the sealed GIS equipment are identified, so as to find defects and eliminate factors affecting the safe and stable operation of the GIS equipment.

[0021] In engineering, GIS detection is realized by first extracting features and then identifying the extracted features. The commonly used partial discharge signal feature extraction methods include statistical feature parameter method, time domain feature parameter method and frequency domain feature parameter method. Since the on-site partial discharge detection is often in a complex electromagnetic environment, it brings difficulties to the feature information extraction of GIS partial discharge. The commonly used partial discharge pattern recognition methods can be divided into two categories, namely artificial neural network method and support vector machine method. In recent years, the artificial neural network is continuously improved and applied in partial discharge recognition. The artificial neural network method has strong feature extraction and recognition ability, but the implementation is relatively complex and has high requirements for the hardware of the running algorithm. The traditional support vector machine has low accuracy in partial discharge recognition, and needs to analyze the discharge characteristics of the partial discharge in multiple dimensions in advance.

[0022] To solve the above problems, the embodiment of the present application provides a partial discharge fault recognition method. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0023] Figure 1 The flowchart of the partial discharge fault recognition method according to the embodiment of the present application is shown in FIG. 1, which includes the following steps: Figure 1

[0024] ​Step S102, obtaining a frequency spectrum data sequence set based on the signal sample set of the gas insulated metal enclosed combined electric appliance, wherein the frequency spectrum data sequence set is used to characterize the partial discharge feature, and the signal sample set is obtained by processing the partial discharge signal of the gas insulated metal enclosed combined electric appliance.

[0025] It can be understood that the partial discharge signal of the gas insulated metal enclosed combined electric appliance (GIS) is processed to obtain a signal sample set, and then a frequency spectrum data sequence set is obtained.

[0026] In an optional embodiment, the method further comprises: detecting the gas insulated metal enclosed combined electric appliance by using an ultrasonic sensor to obtain a signal detection result; performing signal interception by using a preset rectangular window function based on the signal detection result to obtain an initial sequence; performing wavelet decomposition on the initial sequence to obtain a high-frequency component and a low-frequency component; reconstructing based on the high-frequency component and the low-frequency component to obtain a partial discharge signal sequence; and obtaining the signal sample set based on the partial discharge signal sequence.

[0027] It can be understood that the ultrasonic sensor is used to detect the partial discharge signal of the GIS, and the sampled partial discharge signal is intercepted by using the rectangular window function. The data sequence obtained by intercepting the sliding step of the rectangular window function according to the GIS discharge physicality is used as the signal sample set for subsequent processing. According to the characteristics that the periodic narrowband interference in the partial discharge signal is distributed in the form of a narrow band in the frequency domain, has continuous time domain distribution, and the amplitude is generally higher than that of the partial discharge signal, the partial discharge signal can be completely submerged in the narrowband interference. The signal sample set is decomposed by using the wavelet transform to obtain a high-frequency component and a low-frequency component, and then reconstructed to obtain a partial discharge signal sequence, which is used to obtain the signal sample set.

[0028] Optionally, the ultrasonic sensor is a coupled ultrasonic sensor. The coupled ultrasonic sensor is used to detect the partial discharge signal of the GIS. The working bandwidth range of the ultrasonic sensor is 30-220 Hz, and the sensitivity of the ultrasonic sensor is above 80 dB.

[0029] Optionally, there are various ways to perform signal processing by using the rectangular window function, for example: the partial discharge time domain continuous signal is denoted as x(t) (unit: millivolt; t is time, unit: millisecond), the GIS partial discharge discrete signal sampled based on the ultrasonic detection method is denoted as x(n) (unit: millivolt; n=0, 1, 2… is the sampling amount), and the signal is intercepted by using the rectangular window function w(n). a

[0030] ​

[0031] Wherein, n = t x f, set GIS partial discharge signal sampling frequency f = 500 KHz (Kilohertz), sampling time t = 2000 ms (millisecond). W(n) is a rectangular window function, N1, N2 are the starting point and the end point of the rectangular window function respectively, when intercepting the first signal, set N1 = 0, N2 = 9999. z L is the sliding step of the rectangular window function, and is specifically set to L = 10000. z = 10000, then the intercepted GIS partial discharge signal sequence set is:

[0032] x b (m) = [x b1 (m), x b2 (m), …, x bi (m), …, x bM (m)]

[0033] Wherein, the number of the intercepted GIS partial discharge signal sequence is M, M = 100, each sequence contains the intercepted sampling time m, m = 0, 1, 2, …, 9999, x bi (m) (unit: millivolt) represents the i-th GIS partial discharge signal sequence intercepted from x a (n), each sequence includes m = 0, 1, 2, …, 9999 sampling times.

[0034] Alternatively, there are many ways of wavelet decomposition, for example: the partial discharge signal contains periodic narrowband interference, which is distributed in narrowband in the frequency domain and has continuous time domain distribution, and its amplitude is generally higher than that of the partial discharge signal, which can completely submerge the partial discharge signal in the narrowband interference. The characteristics of the above, each sequence in the intercepted GIS partial discharge signal sequence set x b (m) is respectively decomposed by lifting wavelet; the lifting wavelet decomposition method is: setting "db4" wavelet base for 3-level decomposition.

[0035] In an alternative embodiment, the signal sample set of the gas-insulated metal-enclosed combined electric apparatus is obtained to generate the spectrum data sequence set, including: in the case that the signal sample set includes a partial discharge sequence, and the partial discharge sequence includes a plurality of sampling time points respectively corresponding to sampling quantities, determining energy accumulation results of each time point in the plurality of sampling time points in the partial discharge sequence respectively corresponding to a predetermined multi-order frequency resolution; grouping the energy accumulation results of the plurality of sampling time points respectively corresponding to the multi-order frequency resolution to obtain a plurality of groups of spectrum data corresponding to each order of the multi-order frequency resolution; performing compression processing on the plurality of groups of spectrum data corresponding to each order respectively to obtain a first data sequence, wherein the first data sequence is composed of the plurality of groups of spectrum data corresponding to each order after compression in the multi-order frequency resolution; and obtaining the spectrum data sequence set based on the first data sequence.

[0036] It can be understood that a partial discharge sequence includes a plurality of sampling time points, and each sampling time point corresponds to a sampling quantity. In order to highlight the characteristics of the partial discharge, the energy accumulation results of each time point in a predetermined multi-order frequency resolution are determined respectively. For each order of frequency resolution, the energy accumulation results of the plurality of sampling time points are grouped according to time to obtain a plurality of groups of spectrum data corresponding to each order, wherein each group of spectrum data is obtained by grouping according to the sequence of sampling time points. The plurality of groups of spectrum data corresponding to each order are compressed respectively. For each group of spectrum data in each order, compression is performed. The multi-order frequency resolution is also subjected to such processing to obtain a first data sequence. The spectrum data sequence set is obtained on the basis of the first data sequence.

[0037] For ease of understanding, specific examples are given, for example: wavelet time-frequency transformation is performed on each sequence in the signal sample set of the GIS partial discharge after interception and reconstruction to obtain a spectrum data sequence set F i (m).

[0038]

[0039] wherein the i-th GIS partial discharge signal sequence after reconstruction is denoted as x ci (m), F i (m) indicates the spectrum data sequence obtained by wavelet time-frequency transformation of the GIS partial discharge signal sequence x ci (m). ij (m) indicates the energy accumulation result of the GIS partial discharge signal sequence x ci (m) in the j-th order frequency resolution. F(m) = [F1(m), F2(m), …, F i (m), …, F M(m) represents the frequency spectrum data sequence set obtained by wavelet time-frequency transform of the reconstructed GIS partial discharge signal sequence set. The frequency order P of the wavelet time-frequency transform is set to 250, including 1000 sampling time points m = 0, 1, 2,..., 9999, then F i (m) is 250 x 10000.

[0040] Each sequence F i (m) in the frequency spectrum data sequence set F i (m) is compressed to obtain a compressed sequence set FY i1 (r). i1 For example, each order includes 10000 sampling time points m = 0, 1, 2,..., 9999, and f i1 (0)... f i1 (19) represents the energy accumulation result of the 0-9999 time points at the first frequency resolution in the i-th partial discharge signal sequence. Grouping is performed in 20 samples, and the first group includes the energy accumulation results of f i1 (0)... f i (19), and the average value of the above 20 data is obtained to obtain fy i (0). Each group of each order is processed in the above manner to obtain a first data sequence, denoted as FY i (r), i is the sequence identifier.

[0041]

[0042]

[0043] wherein FY i (r) is a compressed sequence of F i (m), and the size of FY i (r) is 250 x 500, r = 0, 1, 2,..., 499 is the sampling amount of the compressed time-frequency signal sequence, and the above 500 is obtained by taking the average value of every 20 sampling time points from 10000 sampling time points, and in the case of varying number of time points included in each group, the above 500 value will also change. Through the above processing, the i-th first data sequence is obtained, and in the case of multiple first data sequences, multiple first data sequences are obtained in the same manner, and the following expression is generated:

[0044] FY(r) = [FY1(r), FY2(r),..., FY i (r),..., FY M (r)]

[0045] wherein FY(r) represents the frequency spectrum data sequence set obtained by compressing F(m).

[0046] In an optional embodiment, the obtaining the spectrum data sequence set based on the first data sequence comprises: in a case where the first data sequence is multiple, performing normalization processing on the multiple first data sequences to obtain normalized multiple first data sequences; determining a second data sequence less than a preset first threshold value in the normalized multiple first data sequences; performing zero processing on the second data sequence to obtain a processed second data sequence; and obtaining the spectrum data sequence set based on the processed second data sequence.

[0047] It can be understood that the first data sequence of the method also needs to be normalized and threshold processed. In a case where the first data sequence is multiple, the normalized multiple first data sequences are obtained by performing normalization processing. In order to highlight the characteristics of partial discharge and achieve better recognition efficiency, threshold processing is needed, a first threshold value is set, and the normalized first data sequence is selected. The second data sequence less than the first threshold value is processed by zero processing, and the zero processing corresponds to the physical meaning of noise reduction. The spectrum data sequence set is obtained based on the processed second data sequence. Through the above processing, the characteristics of partial discharge are highlighted by threshold processing, and noise reduction processing is performed, which is beneficial to improve the recognition efficiency.

[0048] Optionally, the normalization processing mode can be various, for example: the i-th compressed sequence set FY i (r) is linearly normalized to obtain the i-th normalized first data sequence FG i (r).

[0049]

[0050] FG i (r) represents the first data sequence normalized from FY i (r), FG i (r) has a size of 250x500, FY imin represents the maximum value of the elements in FY i (r), and FY imax represents the minimum value of the elements in FY i (r). In a case where the first data sequence is multiple, FG(r) = [FG1(r), FG2(r), …, FG i (r), …, FG M (r)] is generated as a normalized first data sequence set based on the normalized multiple first data sequences.

[0051] In an alternative embodiment, based on the processed second data sequence, the set of spectrum data sequences: determining third data sequences greater than or equal to the first threshold value and less than or equal to a preset second threshold value, and fourth data sequences greater than the second threshold value in the plurality of normalized first data sequences, wherein the first threshold value is less than the second threshold value; performing a one processing on the fourth data sequences to obtain processed fourth data sequences; and obtaining the set of spectrum data sequences based on the processed second data sequence, the third data sequences, and the processed fourth data sequences.

[0052] It can be understood that multi-layer threshold processing is performed, and the first threshold value and the second threshold value are set, so that there are three cases: less than the first threshold value, between the first threshold value and the second threshold value, and greater than the second threshold value, and different processing is performed respectively. For the third data sequences greater than or equal to the first threshold value and less than or equal to the preset second threshold value, no processing is performed and they are directly reserved. For the fourth data sequences greater than the second threshold value, a one processing is performed. The set of spectrum data sequences is obtained based on the processed second data sequence, the third data sequences, and the processed fourth data sequences.

[0053] It should be noted that the processed second data sequence, the third data sequence, and the processed fourth data sequence are included in the set of spectrum data sequences as a plurality of spectrum data sequences.

[0054] In order to facilitate understanding, specific examples are given, for example: comparing each normalized first data sequence with the first threshold value and the second threshold value, and performing corresponding processing to highlight the local features of the set of time-frequency signal sequences.

[0055]

[0056] FZ(r) = [FZ1(r), FZ2(r), …, FZn(r)] (1) i (r) represents FG i (r) the second data sequence after threshold processing, FZ i (r) is 250x500, and the above processing is performed on each sequence to generate FZ(r) = [FZ1(r), FZ2(r), …, FZn(r)] (1) i (r), …, FZ M (r)] is the set of spectrum data sequences after threshold processing.

[0057] Step S104, obtaining the set of wavelet feature maps of the gas insulated metal enclosed combined electric appliance based on the set of spectrum data sequences.

[0058] It can be understood that in order to input the convolutional neural network for processing, it is necessary to generate a wavelet feature map set, which includes the wavelet feature map corresponding to the partial discharge signal sequence. The set of spectrum signal sequences with prominent partial discharge features is mapped to a gray image to obtain the wavelet feature map of GIS partial discharge, and then the wavelet feature map set is generated.

[0059] In an optional embodiment, based on the above-mentioned spectrum data sequence set, the wavelet feature map set of the gas insulated metal enclosed combined electric appliance is obtained, including: in the case that the spectrum data sequence set includes spectrum data sequences, respectively performing square normalization processing on the plurality of spectrum data sequences to obtain a first matrix corresponding to each of the plurality of spectrum data sequences; using a linear mapping method, mapping the first matrix corresponding to each of the plurality of spectrum data sequences to a gray image matrix corresponding to each of the plurality of spectrum data sequences; and based on the gray image matrix corresponding to each of the plurality of spectrum data sequences, obtaining a wavelet feature map set.

[0060] It can be understood that in the case that the spectrum data sequence set includes a plurality of spectrum data sequences, square normalization processing is performed on the plurality of spectrum data sequences to obtain a first matrix, i.e., a corresponding wavelet feature map matrix. Using a linear mapping method, each spectrum data sequence is processed to map the first matrix to a gray image matrix; and based on the gray image matrix, a wavelet feature map is obtained. According to the wavelet feature matrix corresponding to each spectrum data sequence, a wavelet feature map set is obtained.

[0061] Optionally, the gray image matrix is a single-channel gray image.

[0062] Optionally, the square normalization processing method can be various, for example: square normalization processing is performed on the threshold-processed i-th spectrum data sequence FZ i (r) to generate a wavelet feature map matrix. Wherein, fz ij (r) is the spectrum data at the jth frequency resolution of the i-th spectrum data sequence, r represents the compressed sampling time, and P is the order of the multi-order frequency resolution.

[0063]

[0064] Wherein, H i (m) is the spectrum data sequence processed by FZ i (r) through matrix processing, i.e., a wavelet feature map matrix, and the matrix size is 500x500, indicating that the wavelet feature map matrix is constructed by the i-th sequence in the threshold-processed time-frequency signal sequence set FZ(r). H(x) = [H1(m), H2(m), …, H i (m), …, H M (m)] is a constructed wavelet feature map matrix set.

[0065] Step S106: Input the above wavelet feature map set into a preset parallel convolutional neural network for processing to obtain the fault mode recognition result of the above gas-insulated metal-enclosed combined electrical appliance.

[0066] It is understandable that, in order to improve the accuracy and efficiency of fault mode recognition, a parallel convolutional neural network (CNN) is used for processing. The wavelet feature map obtained above is input to obtain the fault mode recognition result for gas-insulated metal-enclosed combined electrical appliances. The parallel convolutional neural network is a low-layer parallel CNN method to complete the CNN algorithm structure and construct a parallel CNN algorithm.

[0067] Optionally, parallel convolutional neural networks (CNNs) can have various structures, such as constructing a parallel CNN algorithm consisting of 3 convolutional layers, 2 max-pooling layers, and 3 fully connected layers. The mathematical expression of the parallel CNN algorithm is established as follows: PY is the output of the parallel CNN algorithm, Tan is the activation function, W is the weight matrix, PX is the input of the parallel CNN algorithm, and K is the offset.

[0068] PY = Tan(W*PX + K)

[0069] Parallel convolutional neural networks, also known as parallel CNN algorithms, are intelligent recognition algorithms that are improved by designing parallel convolutional layers on the basis of standard CNN algorithms. Convolutional layers, max pooling layers, and fully connected layers are one of the operation methods in CNN algorithms.

[0070] Optionally, activation functions are designed for parallel convolutional neural networks to improve the robustness of the algorithm. The mathematical expression of the activation function is established as follows:

[0071]

[0072] Here, Tan(PZ) is the activation function, and PZ represents the output after batch normalization following the convolutional layer operation. Batch normalization is a calculation method that can improve the speed of the algorithm.

[0073] Optionally, the learning rate of the parallel convolutional neural network is set to 0.01, and the number of iterations is 100. Training refers to the process of inputting the GIS partial discharge wavelet feature map into the parallel CNN algorithm for learning, which is a self-learning method for intelligent recognition algorithms.

[0074] Optionally, 70% of the wavelet feature maps in the preset GIS partial discharge wavelet feature map set are used as the test set to train the parallel convolutional neural network, and 30% of the wavelet feature maps are used as the test set to test the processing capability of the parallel convolutional neural network. The test set is used to test the processing capability of the parallel convolutional neural network after training to ensure that the processing capability meets the requirements (e.g., the accuracy reaches the preset recognition threshold).

[0075] In an optional embodiment, the above-mentioned inputting the wavelet feature map set into a preset parallel convolutional neural network for processing to obtain the fault mode recognition result of the gas-insulated metal-enclosed combined electrical appliance includes: when the wavelet feature map set includes multiple wavelet feature maps, inputting the multiple wavelet feature maps into the parallel convolutional neural network for processing to obtain fault mode recognition results corresponding to the multiple wavelet feature maps respectively, wherein the fault mode recognition result includes at least one or more of the following: metal protrusion defect mode, free particle defect mode, insulator surface metal contaminant defect mode, air gap defect mode, and needle-to-needle discharge mode; and determining the fault mode recognition result of the gas-insulated metal-enclosed combined electrical appliance based on the fault mode recognition results corresponding to the multiple wavelet feature maps respectively.

[0076] It is understandable that when the wavelet feature map set of GIS partial discharge includes multiple wavelet feature maps, these multiple wavelet feature maps are input into a parallel convolutional neural network for processing. The parallel convolutional neural network is implemented using a low-layer parallel CNN method to complete the CNN algorithm structure. This parallel CNN algorithm is then constructed to obtain fault mode recognition results corresponding to each of the multiple wavelet feature maps. The fault mode recognition results include at least one or more of the following: metal protrusion defect mode, free particle defect mode, insulator surface metal contaminant defect mode, air gap defect mode, and discharge mode between needles. The corresponding GIS partial discharge mode is given based on the input wavelet feature maps. Based on the fault mode recognition results corresponding to the multiple wavelet feature maps, a comprehensive judgment is made to determine the GIS fault mode recognition result.

[0077] Through the above steps S102 to S106, the goal of improving the accuracy of fault mode recognition can be achieved by using parallel convolutional neural networks, thereby improving the technical effect of partial discharge fault recognition efficiency and solving the technical problem of unsatisfactory partial discharge fault recognition efficiency in related technologies.

[0078] Based on the above embodiments and optional embodiments, the present invention proposes an optional implementation method. Figure 2 This is a schematic flowchart of an optional partial discharge fault identification method provided by an embodiment of the present invention, such as... Figure 2 As shown, it mainly includes 4 steps, which are explained in detail below:

[0079] Step S1 involves preprocessing the GIS partial discharge signal acquired using an ultrasonic detection method, outputting a denoised and reconstructed sample. The GIS partial discharge signal acquired using ultrasonic detection is a one-dimensional signal, while the input to the parallel CNN is a two-dimensional image; therefore, it needs to be converted from a one-dimensional signal to a two-dimensional image. Signal truncation in step S1 prepares the data for the CNN input being a two-dimensional image. The truncation needs to include as many GIS partial discharge features as possible without being too long. A rectangular window function is used here to truncate the original GIS partial discharge signal using a sliding method, which can quickly complete the truncation work. Simultaneously, the truncated GIS partial discharge signal is subjected to lifting wavelet decomposition to obtain high and low frequency components, preparing for subsequent analysis and denoising of the GIS partial discharge signal to highlight its features.

[0080] The data extraction and decomposition process in step S1 mainly consists of the following two sub-steps:

[0081] Step S11, the continuous time-domain signal of partial discharge is denoted as: x(t) (unit: millivolts; t is time, unit: milliseconds), and the discrete partial discharge signal of GIS sampled based on ultrasonic detection is denoted as: x a (n) (unit is millivolt; n = 0, 1, 2... is the number of samples), and the signal is truncated using a rectangular window function w(n);

[0082]

[0083] Where n = t × f, the sampling frequency of the GIS partial discharge signal is set to f = 500 kHz, and the sampling time is t = 2000 ms. W(n) is a rectangular window function, and N1 and N2 are the start and end points of the rectangular window function, respectively. When the first signal is captured, N1 = 0 and N2 = 9999 are set. L z Let L be the sliding step size of the rectangular window function. z =10000, then the extracted GIS partial discharge signal sequence set is:

[0084] x b (m)=[x b1 (m),x b2 (m),…,x bi (m),…,x bM (m)]

[0085] The number of extracted GIS partial discharge signal sequences is denoted as M, where M = 100. Each sequence contains the extracted sampling time denoted as m, where m = 0, 1, 2, ..., 9999. bi (m) (unit: millivolt) represents the amount of x a(n) The i-th GIS partial discharge signal sequence is extracted. Each sequence includes m = 0, 1, 2, ..., 9999 sampling times.

[0086] Step S12, process the extracted GIS partial discharge signal sequence set x b Each sequence within (m) undergoes lifting wavelet decomposition. Lifting wavelet decomposition is a mathematical method for decomposing signals. The wavelet order of the lifting wavelet decomposition method is set to "db4", and the decomposition order is 3. X b Each sequence within (m) will be decomposed into 6 components, namely 3 high-frequency components C. d1 (m), C d2 (m), C d3 (m) and 3 low-frequency components C a1 (m), C a2 (m), C a3 (m)(all components are in millivolts).

[0087] Step S2 involves designing a three-layer wavelet threshold to denoise the wavelet decomposition signal and then reconstructing the signal. Feature transformation and data enhancement are applied to the denoised and reconstructed samples to obtain a spectral data sequence set with prominent partial discharge characteristics. Due to the uncertainty of the on-site GIS partial discharge detection environment, which is often in a complex electromagnetic environment, the actual acquired GIS partial discharge signals contain a large amount of mixed noise, affecting the detection effect. Therefore, filtering out mixed noise from GIS partial discharge signals has become a hot issue in current partial discharge detection. Mixed noise mainly consists of two types: white noise and periodic narrowband interference. The heating of electrical equipment inevitably causes white noise, which has a small amplitude and wide distribution in the time domain and a wide frequency spectrum, making it the most common noise in actual detection. System high-order harmonics, radio interference, and carrier communication can cause periodic narrowband interference. This periodic narrowband interference has a narrow band distribution in the frequency domain and a continuous time domain distribution. Its amplitude is generally higher than that of the partial discharge signal, and it can completely submerge the partial discharge signal in narrowband interference. Therefore, effectively suppressing noise is crucial for partial discharge detection. By setting multiple thresholds, noise reduction is achieved on the high-frequency components obtained from lifting wavelet decomposition. Simultaneously, during signal reconstruction, first- and second-order low-frequency components are removed, collectively filtering out white noise and periodic narrowband interference contained in the GIS partial discharge signal.

[0088] Figure 3 This is a reconstructed schematic diagram of an optional partial discharge fault identification method provided by an embodiment of the present invention, as shown below. Figure 3As shown, the process in step S2 can be summarized as follows: wavelet decomposition is performed on the truncated GIS partial discharge signal, assuming that 6 components are obtained. The 6 components obtained by decomposition include the high-frequency component C. d1 (m), C d2 (m), C d3 (m) and low-frequency component C a1 (m), C a2 (m), C a3 (m), the three high-frequency components C after thresholding d1 (m), C d2 (m), C d3 (m) and low-frequency component C a1 (m), C a2 (m), C a3 (m) Perform signal reconstruction to obtain the reconstructed GIS partial discharge signal sample set.

[0089] Step S3 involves mapping the spectral signal sequence set highlighting the local features of PD to a grayscale image, obtaining the wavelet feature map of GIS partial discharge. Since the time-domain waveform of the GIS partial discharge signal has limited features, wavelet time-frequency transform is used to obtain the frequency domain information of the GIS partial discharge. Using wavelet time-frequency transform achieves higher frequency resolution, meaning more detailed information about the frequency domain features of the GIS partial discharge signal is obtained. Then, the frequency domain features of the GIS partial discharge signal are reconstructed to build a larger feature map matrix, highlighting its detailed features. The feature matrix is ​​output as an image to meet the input data format requirements of parallel CNNs.

[0090] Figure 4 This is a schematic diagram illustrating the feature map construction of an optional partial discharge fault identification method according to an embodiment of the present invention, such as... Figure 4 The diagram illustrates the wavelet feature map construction process, specifically the process of obtaining GIS partial discharge wavelet feature maps that satisfy the input data format of parallel CNN. Step S3 can be summarized as follows: the reconstructed GIS partial discharge signal is obtained, which is then used for wavelet time-frequency transformation. The resulting wavelet video transformation is recombined to obtain a wavelet feature map that has not yet been normalized. After normalization, a wavelet feature map with a size of 28*28 is obtained. Steps S31 to S36 are explained in detail below.

[0091] Step S31: Perform wavelet time-frequency transform on each sequence in the extracted and reconstructed GIS partial discharge signal sample set to obtain the frequency domain signal generation spectrum data sequence set F. i (m).

[0092]

[0093] Here, the reconstructed i-th GIS partial discharge signal sequence is denoted as x. ci (m), F i (m) represents the reconstructed GIS partial discharge signal sequence x ci (m) is the spectral data sequence obtained by performing wavelet time-frequency transform, f ij (m) represents the reconstructed GIS partial discharge signal sequence x ci (m) represents the energy accumulation result at the j-th order frequency resolution. F(m) = [F1(m), F2(m), ..., F i (m),…,F M [m] represents the spectral data sequence obtained by wavelet time-frequency transforming the reconstructed GIS partial discharge signal sequence set. Setting the frequency order of the wavelet time-frequency transform P = 250, including 1000 sampling times m = 0, 1, 2, ..., 9999, then F... i (m) The size is 250×10000.

[0094] Step S32, for each sequence F in the spectrum data sequence set F(m) i Each row of (m) is divided into segments of 20 samples, and the mean is calculated to generate the compressed sequence set FY. i (r). For example: each order includes 10,000 sampling times, m = 0, 1, 2, ..., 9999, f i1 (0)……f i1 (9999) represents the energy accumulation result at time 0-9999 in the i-th partial discharge signal sequence at the first-order frequency resolution. Grouping the data into segments of 20 samples, the first group includes f... i1 (0)…f i1 (19) The cumulative energy result is used to calculate the average of the above 20 data points to obtain fy. i1 (0). Using the above processing method, each group of each order is processed to obtain the first data sequence denoted as FY. i (r), where i is the identifier of the sequence.

[0095]

[0096]

[0097] Among them, FY i (r) is from F i (m) Compressed sequence, FY i(r) represents the sample size of the compressed time-frequency signal sequence, which is 250 × 500, where r = 0, 1, 2, ..., 499. The 500 is the average of 20 samples taken from 10,000 sampling times. This value of 500 will change as the number of times in each group varies. Through the above processing, the i-th first data sequence is obtained. If multiple first data sequences exist, they are obtained using the same method, generating the following expression:

[0098] FY(r) = [FY1(r), FY2(r), ..., FY i (r),…,FY M (r)]

[0099] Where FY(r) represents the set of spectral data sequences obtained by compressing F(m).

[0100] Step S33, for the i-th compressed sequence set FY i (r) Perform linear normalization to obtain the i-th normalized first data sequence FG i (r).

[0101]

[0102] Among them, FG i (r) indicates that FY i (r) The first data sequence after normalization, FG i (r) is 250×500, FY imin FY i The maximum value of the elements in (r), FY imax FY i The minimum value of an element within (r). In the case of multiple first data sequences, based on the normalized multiple first data sequences, generate FG(r) = [FG1(r), FG2(r), ..., FG...]. i (r),…,FG M [r] represents the first set of data sequences after normalization.

[0103] Step S34: Compare each normalized first data sequence with the first threshold and the second threshold, and perform corresponding processing to highlight the local features of the time-frequency signal sequence set.

[0104]

[0105] Among them, FZ i (r) represents FG i (r) The second data sequence after thresholding, FZ iThe size of (r) is 250×500. The above processing is performed on each sequence to generate FZ(r) = [FZ1(r), FZ2(r), ..., FZ...]. i (r),…,FZ M [r] represents the set of spectral data sequences after thresholding.

[0106] Step S35, the i-th spectral data sequence FZ after threshold processing i (r) is processed into a matrix to generate a wavelet feature map matrix. Where fz ij (r) represents the spectral data at the j-th frequency resolution in the i-th spectral data sequence, where r represents the compressed sampling time and P is the order of the multi-order frequency resolution.

[0107]

[0108] Among them, H i (m) is from FZ i (r) is the spectral data sequence after matrix processing, i.e., the wavelet feature map matrix, with a matrix size of 500×500. It represents the wavelet feature map matrix constructed from the i-th sequence in the time-frequency signal sequence set FZ(r) after thresholding. H(x)=[H1(m),H2(m),…,H i (m),…,H M [m] represents the set of wavelet feature maps constructed.

[0109] Step S36: Linearly map the wavelet feature map matrix to the grayscale image matrix and output it as an image with a length and width of 28×28, i.e., the GIS partial discharge wavelet feature map. Figure 5 This is a schematic diagram of wavelet features for an optional partial discharge fault identification method according to an embodiment of the present invention. The wavelet feature map matrix is ​​linearly mapped to a grayscale image matrix and output as an image with a length and width of 28×28, i.e., a GIS partial discharge wavelet feature map. The horizontal axis of the obtained wavelet feature image represents the time axis, and the vertical axis represents the frequency axis. Since it is a normalized coordinate system, neither the horizontal nor vertical axis has units. Figure 5 The depth of each pixel in the image represents the amount of energy accumulated at the corresponding frequency over time; the lighter the color, the more energy is accumulated, and the darker the color, the less energy is accumulated. For example... Figure 5 The pixel indicated by the middle arrow is the pixel at coordinates (0, 27), which means that at time 0, with a normalized frequency of 27, the energy accumulation is represented by the gray level of the pixel at position (0, 27).

[0110] Step S4 involves designing a low-level parallel convolutional neural network to identify the wavelet feature map of GIS partial discharge, ultimately obtaining the types of defects causing GIS partial discharge faults. To ensure high recognition accuracy while reducing hardware requirements and single-run time, a low-layer parallel CNN approach is used to construct the CNN algorithm structure, consisting of 3 convolutional layers, 2 max-pooling layers, and 3 fully connected layers. The mathematical expression of the parallel CNN algorithm is established as follows: PY is the output of the parallel CNN algorithm, Tan is the activation function, W is the weight matrix, PX is the input of the parallel CNN algorithm, and K is the offset.

[0111] PY = Tan(W*PX + K)

[0112] Parallel convolutional neural networks, also known as parallel CNN algorithms, are intelligent recognition algorithms that are improved by designing parallel convolutional layers on the basis of standard CNN algorithms. Convolutional layers, max pooling layers, and fully connected layers are one of the operation methods in CNN algorithms.

[0113] Activation functions are designed for parallel convolutional neural networks to improve the algorithm's robustness. The mathematical expression of the activation function is established as follows:

[0114]

[0115] Here, Tan(PZ) is the activation function, and PZ represents the output after batch normalization following the convolutional layer operation. Batch normalization is a calculation method that can improve the speed of the algorithm.

[0116] The learning rate for the parallel convolutional neural network was set to 0.01, and the number of iterations was 100. Training refers to the process of inputting the GIS partial discharge wavelet feature map into the parallel CNN algorithm for learning; this is a self-learning method for intelligent recognition algorithms. 70% of the wavelet feature maps in the preset set were used as the test set for training the parallel convolutional neural network, while 30% were used as the test set to test the processing capability of the parallel convolutional neural network. The test set was used to ensure that the processing capability after training met the requirements (e.g., accuracy reached the preset recognition threshold).

[0117] The trained and tested parallel convolutional neural network is processed by inputting the wavelet feature map set obtained in step S3 above, resulting in the fault mode recognition mode of GIS partial discharge. The aforementioned fault modes refer to at least one or more of the following: metal protrusion defect mode, free particle defect mode, insulator surface metal contaminant defect mode, air gap defect mode, and needle-to-needle discharge mode.

[0118] Figure 6This is a pattern recognition schematic diagram of an optional partial discharge fault identification method provided by an embodiment of the present invention, such as... Figure 6 The diagram illustrates the processing of a trained parallel CNN algorithm. The horizontal axis represents the predicted value, and the vertical axis represents the true value. Labels 1-5 represent the labels: label 1 indicates a metal protrusion defect mode, label 2 indicates a free particle defect mode, label 3 indicates a metal contaminant defect mode on the insulator surface, label 4 indicates an air gap defect mode, and label 5 indicates a discharge mode between needles. Figure 6 The value in the figure represents the accuracy rate of the test sample. For a 30% test sample, Figure 6 The value "30" in the cell indicates that the corresponding label defect pattern classification accuracy is 100%. Using the recognition method provided in this embodiment of the invention, a recognition accuracy of 99.33% can be achieved with a training time of 6 seconds.

[0119] The above-described optional implementation methods achieve at least the following effects: Because GIS partial discharge identification features are pre-extracted, the complexity requirements of the CNN used are reduced, allowing for the use of a simpler, low-layer CNN to identify GIS partial discharge patterns. The low layer count and simple structure of the parallel CNN ensures high accuracy in identifying GIS partial discharge patterns while requiring less hardware. Furthermore, the parallel CNN algorithm, due to its parallel structure, runs faster and achieves higher accuracy in GIS partial discharge pattern identification in a shorter time.

[0120] 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, and 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.

[0121] This embodiment also provides a partial discharge fault identification device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0122] According to embodiments of the present invention, an apparatus embodiment for implementing a partial discharge fault identification method is also provided. Figure 7 This is a schematic diagram of a partial discharge fault identification device according to an embodiment of the present invention, such as... Figure 7 As shown, the above-mentioned partial discharge fault identification device includes: a data acquisition module 702, an acquisition module 704, and an identification module 706. The device will be described below.

[0123] The acquisition module 702 is used to obtain a spectrum data sequence set based on the signal sample set of the gas-insulated metal-enclosed combined electrical appliance, wherein the spectrum data sequence set is used to characterize the partial discharge characteristics, and the signal sample set is obtained by processing the partial discharge signal of the gas-insulated metal-enclosed combined electrical appliance.

[0124] The acquisition module 704, connected to the acquisition module 702, is used to obtain the wavelet feature map set of the gas-insulated metal-enclosed combined electrical appliance based on the above-mentioned spectrum data sequence set;

[0125] The identification module 706, connected to the acquisition module 704, is used to input the above wavelet feature map set into a preset parallel convolutional neural network for processing, so as to obtain the fault mode identification result of the above gas-insulated metal-enclosed combined electrical appliance.

[0126] In a partial discharge fault identification device provided by this invention, a data acquisition module 702 is used to obtain a spectral data sequence set based on a signal sample set of a gas-insulated metal-enclosed combined electrical appliance. The spectral data sequence set is used to characterize partial discharge features, and the signal sample set is obtained by processing the partial discharge signal of the gas-insulated metal-enclosed combined electrical appliance. An acquisition module 704, connected to the data acquisition module 702, is used to obtain a wavelet feature map set of the gas-insulated metal-enclosed combined electrical appliance based on the spectral data sequence set. An identification module 706, connected to the acquisition module 704, is used to input the wavelet feature map set into a preset parallel convolutional neural network for processing, thereby obtaining a fault mode identification result for the gas-insulated metal-enclosed combined electrical appliance. This achieves the goal of improving the accuracy of fault mode identification by utilizing a parallel convolutional neural network, realizing the technical effect of improving the efficiency of partial discharge fault identification, and thus solving the technical problem of unsatisfactory partial discharge fault identification efficiency in related technologies.

[0127] 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.

[0128] It should be noted that the aforementioned acquisition module 702, obtaining module 704, and identification module 706 correspond to steps S102 to S106 in the embodiments. The instances and application scenarios implemented by these modules and their corresponding steps are the same, but they are not limited to the content disclosed in the above embodiments. It should also be noted that these modules, as part of the device, can run on a computer terminal.

[0129] 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.

[0130] The aforementioned partial discharge fault identification device may also include a processor and a memory. The acquisition module 702, the acquisition module 704, the identification module 706, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0131] The processor contains a core that retrieves the corresponding program unit from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, 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.

[0132] This invention provides a non-volatile storage medium storing a program that, when executed by a processor, implements a partial discharge fault identification method.

[0133] This invention provides an electronic device, comprising a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: obtaining a spectral data sequence set based on a signal sample set of a gas-insulated metal-enclosed combined electrical appliance, wherein the spectral data sequence set is used to characterize partial discharge features, and the signal sample set is obtained by processing the partial discharge signal of the gas-insulated metal-enclosed combined electrical appliance; obtaining a wavelet feature map set of the gas-insulated metal-enclosed combined electrical appliance based on the spectral data sequence set; and inputting the wavelet feature map set into a preset parallel convolutional neural network for processing to obtain a fault mode recognition result for the gas-insulated metal-enclosed combined electrical appliance. The device in this document can be a server, PC, etc.

[0134] The present invention also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having the following method steps: obtaining a spectral data sequence set based on a signal sample set of a gas-insulated metal-enclosed combined electrical appliance, wherein the spectral data sequence set is used to characterize partial discharge features, and the signal sample set is obtained by processing the partial discharge signal of the gas-insulated metal-enclosed combined electrical appliance; obtaining a wavelet feature map set of the gas-insulated metal-enclosed combined electrical appliance based on the spectral data sequence set; and inputting the wavelet feature map set into a preset parallel convolutional neural network for processing to obtain a fault mode recognition result for the gas-insulated metal-enclosed combined electrical appliance.

[0135] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0137] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 Figure 1 The steps of the function specified in one or more boxes.

[0139] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0140] 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 is an example of computer-readable media.

[0141] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0142] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0143] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0144] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for identifying partial discharge faults, characterized in that, include: Based on the signal sample set of the gas-insulated metal-enclosed combined electrical appliance, a spectral data sequence set is obtained, wherein the spectral data sequence set is used to characterize the partial discharge characteristics, and the signal sample set is obtained by processing the partial discharge signal of the gas-insulated metal-enclosed combined electrical appliance. Based on the aforementioned spectral data sequence set, the wavelet feature map set of the gas-insulated metal-enclosed combined electrical appliance is obtained; The wavelet feature map set is input into a preset parallel convolutional neural network for processing to obtain the fault mode recognition result of the gas-insulated metal-enclosed combined electrical appliance. The parallel convolutional neural network is an improvement on the standard CNN algorithm by designing parallel convolutional layers, and consists of 3 convolutional layers, 2 max pooling layers and 3 fully connected layers. The method further includes: using an ultrasonic sensor to detect a gas-insulated metal-enclosed combined electrical appliance to obtain a signal detection result; based on the signal detection result, using a preset rectangular window function to truncate the signal to obtain an initial sequence; performing wavelet decomposition on the initial sequence to obtain high-frequency components and low-frequency components; using a three-layer wavelet threshold to denoise and reconstruct the high-frequency components and the low-frequency components to obtain a partial discharge signal sequence; and obtaining the signal sample set based on the partial discharge signal sequence. The method further includes: processing the signal sample set to obtain a first data sequence; if there are multiple first data sequences, normalizing the multiple first data sequences to obtain normalized multiple first data sequences; determining a second data sequence among the normalized multiple first data sequences that is less than a preset first threshold; setting the second data sequence to zero to obtain a processed second data sequence; determining a third data sequence among the normalized multiple first data sequences that is greater than or equal to the first threshold and less than or equal to a preset second threshold, and a fourth data sequence that is greater than the second threshold, wherein the first threshold is less than the second threshold; setting the fourth data sequence to one to obtain a processed fourth data sequence; and obtaining the spectrum data sequence set based on the processed second data sequence, the third data sequence, and the processed fourth data sequence.

2. The method according to claim 1, characterized in that, The signal sample set based on the gas-insulated metal-enclosed combined electrical appliance yields a spectral data sequence set, including: When the signal sample set includes a partial discharge sequence, and the partial discharge sequence includes the sampling amount corresponding to multiple sampling times, the energy accumulation result corresponding to each of the multiple sampling times in the partial discharge sequence at a predetermined multi-order frequency resolution is determined. Based on the energy accumulation results corresponding to multiple sampling times, multiple groups of spectral data corresponding to each order in the multi-order frequency resolution are obtained by grouping the data. The first data sequence is obtained by compressing the multiple sets of spectral data corresponding to each order, wherein the first data sequence is composed of multiple sets of compressed spectral data corresponding to each order of the multi-order frequency resolution. Based on the first data sequence, the spectral data sequence set is obtained.

3. The method according to claim 1, characterized in that, Based on the aforementioned spectral data sequence set, a wavelet feature map set for the gas-insulated metal-enclosed combined electrical appliance is obtained, including: When the set of spectrum data sequences includes multiple spectrum data sequences, the multiple spectrum data sequences are respectively matrixed to obtain the first matrix corresponding to each of the multiple spectrum data sequences. A linear mapping method is used to map the first matrix corresponding to the multiple spectral data sequences to their respective grayscale image matrices; Based on the grayscale image matrices corresponding to the multiple spectral data sequences, a wavelet feature map set is obtained.

4. The method according to any one of claims 1 to 3, characterized in that, The step of inputting the wavelet feature map set into a preset parallel convolutional neural network for processing to obtain the fault mode recognition result of the gas-insulated metal-enclosed combined electrical appliance includes: When the wavelet feature map set includes multiple wavelet feature maps, the multiple wavelet feature maps are input into the parallel convolutional neural network for processing to obtain fault mode recognition results corresponding to the multiple wavelet feature maps respectively. The fault mode recognition results include at least one or more of the following: metal protrusion defect mode, free particle defect mode, insulator surface metal contaminant defect mode, air gap defect mode, and discharge mode between needles. Based on the fault mode identification results corresponding to the multiple wavelet feature maps, the fault mode identification results of the gas-insulated metal-enclosed combined electrical appliance are determined.

5. A partial discharge fault identification device, characterized in that, include: The acquisition module is used to obtain a spectral data sequence set based on the signal sample set of the gas-insulated metal-enclosed combined electrical appliance, wherein the spectral data sequence set is used to characterize the partial discharge characteristics, and the signal sample set is obtained by processing the partial discharge signal of the gas-insulated metal-enclosed combined electrical appliance. The acquisition module is used to obtain the wavelet feature map set of the gas-insulated metal-enclosed combined electrical appliance based on the spectrum data sequence set; The identification module is used to input the wavelet feature map set into a preset parallel convolutional neural network for processing to obtain the fault mode identification result of the gas-insulated metal-enclosed combined electrical appliance. The parallel convolutional neural network is an improvement on the standard CNN algorithm by designing parallel convolutional layers, and consists of 3 convolutional layers, 2 max pooling layers and 3 fully connected layers. The device is further configured to use an ultrasonic sensor to detect gas-insulated metal-enclosed combined electrical appliances and obtain signal detection results; based on the signal detection results, a preset rectangular window function is used to truncate the signal to obtain an initial sequence; wavelet decomposition is performed on the initial sequence to obtain high-frequency components and low-frequency components; noise reduction and reconstruction of the high-frequency components and low-frequency components are performed using a three-layer wavelet threshold to obtain a partial discharge signal sequence; and the signal sample set is obtained based on the partial discharge signal sequence. The apparatus is further configured to: process the signal sample set to obtain a first data sequence; when there are multiple first data sequences, normalize the multiple first data sequences to obtain normalized multiple first data sequences; determine a second data sequence among the normalized multiple first data sequences that is less than a preset first threshold; set the second data sequence to zero to obtain a processed second data sequence; determine a third data sequence among the normalized multiple first data sequences that is greater than or equal to the first threshold and less than or equal to a preset second threshold, and a fourth data sequence that is greater than the second threshold, wherein the first threshold is less than the second threshold; set the fourth data sequence to one to obtain a processed fourth data sequence; and obtain the spectrum data sequence set based on the processed second data sequence, the third data sequence, and the processed fourth data sequence.

6. 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 fault identification method according to any one of claims 1 to 4.

7. An electronic device, characterized in that, include: 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 fault identification method according to any one of claims 1 to 4.

Images