Partial discharge defect diagnosis method and device based on HHT marginal spectrum and convolutional neural network

Abstract

The invention discloses a partial discharge defect diagnosis method and device based on an HHT marginal spectrum and a convolutional neural network, and the method comprises the steps: carrying out the decomposition of a collected partial discharge signal through an EMD method, and obtaining a plurality of IMFs; respectively carrying out Hilbert solution on the plurality of obtained intrinsic mode components IMFs to obtain an HHT marginal spectrum; dividing the HHT marginal spectrums of all samples according to a set proportion to obtain a training set and a test set; and training a preset CNN model by using the training set, and performing defect detection on partial discharge by using the trained CNN model. The method is suitable for intelligent analysis of partial discharge signals and automatic identification of insulation defects, and can accurately determine the type of the insulation fault of the electrical equipment, thereby helping inspection personnel to quickly determine the insulation damaged part of the equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of partial discharge defect recognition, and in particular relates to a partial discharge defect diagnosis method and device based on HHT marginal spectrum and convolutional neural network. Background Art

[0002] Partial discharge refers to the breakdown discharge phenomenon that occurs in the local area of ​​the insulating medium. It is the main cause of insulation failure in high-voltage power distribution equipment. When partial discharge occurs, it will damage and corrode the surrounding insulating medium. Although weak partial discharge does little damage to the insulation of the equipment and will not be immediately apparent, long-term or severe partial discharge will quickly weaken the insulation strength. Eventually, it will lead to insulation breakdown, causing equipment failure and safety accidents. Therefore, partial discharge detection and analysis is an important means to judge the insulation status of equipment and promptly discover equipment insulation defects. However, traditional manual patrol and maintenance has the disadvantages of "blind time, blind area" and high personnel costs, and the quality of operation and maintenance varies from person to person. The measurement data has not formed a full life cycle model, resulting in equipment failures from time to time.

[0003] With the popularization and development of artificial intelligence and the Internet of Things, a large amount of electrical and non-electrical data can be collected through sensors and stored in computers. Then, artificial intelligence technologies such as machine learning and deep learning are used to process these data to complete tasks such as pattern recognition and target detection. Among them, machine learning and deep learning are two important components of artificial intelligence algorithms. They can not only be used to extract data features, but also have superior pattern recognition capabilities. Therefore, combining artificial intelligence technologies such as machine learning and deep learning with partial discharge detection technology for distribution equipment can judge the insulation status of equipment more real-time and accurately, and maintain the safety and reliability of the power grid.

[0004] Feature selection is a crucial link in pattern recognition. Selecting appropriate feature quantities as inputs for machine learning or deep learning classifiers can effectively improve the accuracy of recognition. Hilbert-Huang Transform (HHT) is an adaptive time-frequency analysis method for non-stationary and nonlinear signals. It includes two parts. One is to perform empirical mode decomposition (EMD) on the signal to obtain intrinsic mode functions (IMFs). The second is to perform Hilbert transform on IMFs to obtain a complete and accurate time-frequency distribution of signal energy, so that the corresponding marginal spectrum can be further constructed. The HHT marginal spectrum can accurately reflect the variation of signal amplitude with frequency, and has higher accuracy and resolution than the traditional Fourier spectrum. Convolutional Neural Networks (CNN) extracts local features and deep features from local receptive fields through convolution and pooling operations, and has strong learning ability. Therefore, the present invention proposes a local discharge pattern recognition method based on HHT marginal spectrum and convolutional neural network CNN. Summary of the invention

[0005] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and provide a partial discharge defect diagnosis method and device based on HHT marginal spectrum and convolutional neural network.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network, comprising the following steps:

[0008] Performing empirical mode decomposition (EMD) operation on the collected partial discharge signal to obtain multiple intrinsic mode components (IMFs);

[0009] Performing Hilbert solutions on the obtained multiple intrinsic modal components IMFs respectively to obtain HHT marginal spectra; the HHT marginal spectra reflect the amplitude distribution of the signal at each instantaneous frequency point and are used to characterize the frequency characteristics of the response signal;

[0010] The HHT marginal spectra of all samples are divided according to the set ratio to obtain the training set and the test set;

[0011] The preset CNN model is trained using the training set, and the trained CNN model is used to perform defect detection on partial discharge; the CNN model includes five convolutional layers, five pooling layers, one dropout layer and one fully connected layer, which are convolutional layer C1, convolutional layer C2, pooling layer P1, convolutional layer C3, pooling layer P2, convolutional layer C4, pooling layer P3, convolutional layer C5, pooling layer P4, pooling layer P5, dropout layer and fully connected layer, respectively.

[0012] As a preferred technical solution, the collected partial discharge signal is subjected to EMD decomposition to obtain intrinsic modal components IMFs, specifically:

[0013] The local maximum and minimum points of the original signal x(t) are fitted into the upper envelope U(t) and lower envelope L(t) of the original signal x(t), and the mean m of the two envelopes is calculated. 11 (t);

[0014]

[0015] Calculate the original signal x(t) and the mean m 11 The difference h of (t) 11 (t):

[0016] h 11 (t) = x(t) - m 11 (t)

[0017] Judge h 11 (t) Whether the IMF conditions are met, if h 11 (t) is not an IMF, repeat the above process k times until the difference h 1k (t) Satisfy IMF conditions;

[0018] h 1k (t) = h 1(k-1) (t)-m 1k (t)

[0019] Among them, h 1(k-1) (t) represents the signal obtained by the k-1th screening; h 1k (t) represents the signal obtained by the kth screening; m 1k (t) represents h 1(k-1) (t) average envelope;

[0020] The specific method of judging whether the IMF condition is met is to use the standard deviation coefficient SD value to judge whether each result is an IMF component; that is:

[0021]

[0022] Where T is the total time length of the signal, and t represents the time point;

[0023] When h 1k When the SD value requirement is met, let c1 = h 1k is the first IMF component of the original signal x(t). Subtracting c1 from the original signal x(t) yields the residual signal r1=x(t)-c1;

[0024] Repeat the above process with r1 as the signal to be decomposed, r2 = r1-c2, r3 = r2-c3, ... r n =r n-1 -c n ; until the remaining signal r n is a monotonic function;

[0025] After the above process, the signal x(t) is expressed as:

[0026]

[0027] That is, after EMD, any signal can be decomposed into the sum of n IMFs and residuals, and the IMFs contain components of different frequency bands of the signal from high to low, and the frequency components contained in each frequency band vary with different signals.

[0028] As a preferred technical solution, the value of SD is 0.2 to 0.3.

[0029] As a preferred technical solution, Hilbert transform is performed on the obtained multiple intrinsic modal components IMFs respectively to obtain HHT marginal spectra, which are specifically:

[0030] Using the EMD method to decompose the signal s(t) we get the following formula:

[0031]

[0032] c1(t),c2(t),…,c n (t) are the n IMF components obtained by decomposition; each IMF component contains components of different frequency bands of s(t), and their center frequencies are different and change with the signal itself;

[0033] For IMF component c i (t) respectively perform Hilbert transform H[·], that is:

[0034]

[0035] Define the analytic signal z in the complex domain i (t) is:

[0036]

[0037] The instantaneous amplitude A i (t) is:

[0038]

[0039] Instantaneous phase for:

[0040]

[0041] Instantaneous frequency ω i (t) is:

[0042]

[0043] Then the expression of the original signal s(t) is:

[0044]

[0045] Among them, Re means taking the real part of the complex number;

[0046] The expansion of the above formula is the Hilbert spectrum, that is,

[0047]

[0048] The Hilbert spectrum accurately describes the variation of the signal amplitude over time and frequency. The HHT marginal spectrum is further defined as follows:

[0049]

[0050] After the partial discharge signal is decomposed by EMD, the frequency is divided into multiple frequency intervals, and the HHT marginal spectrum of each sample is calculated and statistically obtained.

[0051] As a preferred technical solution, a preset CNN model is trained using a training set, specifically:

[0052] The marginal spectrum data of each discharge sample has a total of 800 data points, which are finally classified into 4 types of discharges. Therefore, the number of input layer nodes is 800 and the number of output layer nodes is 4. The input vector dimension is (800, 1). After the convolution layer C1 with a convolution kernel size of 8 and a step size of 2 and the convolution layer C2 with a convolution kernel size of 8, a step size of 2 and a padding operation of the same, an output mapping of size (199, 16) is obtained. After a downsampling operation of size 2, the pooling layer P1 output of (99, 16) is obtained. Then, after convolution and pooling operations with kernel sizes of 64×4 and 64×2, the output is obtained. The features of (25, 64) are input into the convolution layer C4 with a size of 256×4 and the same expansion operation, and are pooled using the pooling layer P3 with a size of 2 to obtain an output map of (6, 256). Subsequently, a 512×2 convolution operation and a 512×2 pooling operation are performed to obtain an output map of (3, 512). Finally, four classified partial discharge defect states are output after global average pooling, a dropout layer, a fully connected layer, and a softmax classifier. The partial discharge defect states include tip discharge, suspended discharge, internal discharge, and surface discharge.

[0053] As a preferred technical solution, when training the CNN model,

[0054] The optimizer is Adam, the learning rate is 0.0002, and the number of iterations is 50.

[0055] As a preferred technical solution, the model verification step is also included, specifically:

[0056] The training set is further divided into a training set and a validation set according to a set ratio. The training set is sent to the designed CNN model for training, and the model with the best performance in the validation set is selected as the final CNN model. The generalization ability of the model is verified by using the accuracy, precision, recall and F1 value of the test set.

[0057] In a second aspect, the present invention provides a partial discharge defect diagnosis system based on HHT marginal spectrum and convolutional neural network, which is applied to the partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network, and includes an EMD decomposition module, a Hilbert solution module, a data set division module and a model training module;

[0058] The EMD decomposition module is used to perform an empirical mode decomposition (EMD) operation on the collected partial discharge signal to obtain a plurality of intrinsic mode components (IMFs);

[0059] The Hilbert solution module is used to perform Hilbert solutions on the obtained multiple intrinsic modal components IMFs respectively to obtain HHT marginal spectra; the HHT marginal spectra reflect the amplitude distribution of the signal at each instantaneous frequency point and are used to characterize the frequency characteristics of the reaction signal;

[0060] The data set partitioning module is used to partition the HHT marginal spectra of all samples according to a set ratio to obtain a training set and a test set;

[0061] The model training module is used to train a preset CNN model using a training set, and use the trained CNN model to perform defect detection on partial discharge; the CNN model includes five convolutional layers, five pooling layers, a dropout layer and a fully connected layer, which are convolutional layer C1, convolutional layer C2, pooling layer P1, convolutional layer C3, pooling layer P2, convolutional layer C4, pooling layer P3, convolutional layer C5, pooling layer P4, pooling layer P5, dropout layer and fully connected layer in sequence.

[0062] In a third aspect, the present invention provides an electronic device, the electronic device comprising:

[0063] at least one processor; and,

[0064] a memory communicatively connected to the at least one processor; wherein,

[0065] The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network.

[0066] In a fourth aspect, the present invention provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the local discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network.

[0067] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0068] The advantages of the partial discharge pattern recognition method based on HHT marginal spectrum and convolutional neural network CNN of the present invention are: (1) The time-frequency characteristics of the partial discharge defect signal are extracted by using HHT marginal spectrum. This not only reduces the input vector dimension of the convolutional neural network, but also effectively retains the key characteristics of the partial discharge signal. (2) The recognition of partial discharge is realized by using a simple convolutional neural network containing five layers of convolution, which has a high recognition accuracy even when the number of samples is small. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0070] Figure 1 This is a flow chart of a partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network according to an embodiment of the present invention;

[0071] Figure 2 This is a schematic diagram of the structure of a CNN according to an embodiment of the present invention;

[0072] Figure 3 It is a block diagram of a partial discharge defect diagnosis system based on HHT marginal spectrum and convolutional neural network according to an embodiment of the present invention.

[0073] Figure 4 4 is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0075] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0076] This embodiment provides a partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network, which performs EMD decomposition on the collected partial discharge signal to obtain intrinsic modal components IMFs, and then performs Hilbert transform on the IMFs to obtain HHT marginal spectrum. The HHT marginal spectrum of all samples is divided according to a set ratio to obtain a training set and a test set. The built CNN model is trained using the training set, and the model is tested using the test set to test the generalization ability of the model.

[0077] like Figure 1As shown, the partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network in this embodiment includes the following steps:

[0078] S1. Decompose the collected partial discharge signal using the empirical mode decomposition (EMD) algorithm to obtain multiple intrinsic mode components (IMFs).

[0079] Empirical mode decomposition (EMD) is one of the most representative adaptive time-frequency analysis methods, which is suitable for the analysis and processing of nonlinear and non-stationary signals. EMD is essentially an adaptive binary filter, which adaptively decomposes complex signals into several IMF components with different frequencies and amplitudes and a residual based on the time scale characteristics of the signal, based on the concepts of instantaneous frequency and intrinsic mode function, and can characterize the inherent vibration mode of the data.

[0080] Furthermore, step S1 is specifically as follows:

[0081] S11. Use the cubic spline function to fit the local maximum and minimum points of the original signal x(t) into the upper envelope U(t) and lower envelope L(t) of the original signal x(t), and calculate the mean m of the two envelopes. 11 (t), mean m 11 (t) is calculated as follows:

[0082]

[0083] S12. Find the original signal x(t) and mean m 11 The difference h 11 (t): h 11 (t) = x(t) - m 11 (t). Under ideal conditions h 11 (t) satisfies the IMF condition, but since the collected signal is nonlinear and non-stationary, there will be asymmetric waves. In this case, the above formula is repeated k times, h 1k (t) = h 1(k-1) (t)-m 1k (t), until the difference h 1k (t) satisfies the IMF conditions, where h 1(k-1) (t) represents the signal obtained by the k-1th screening; h 1k (t) represents the signal obtained by the kth screening; m 1k (t) represents h 1(k-1) (t) The average envelope.

[0084] Because in practice, h 1k (t) = 0 is usually difficult to satisfy. The standard deviation (SD) value can be used to determine whether each result is an IMF component, that is:

[0085]

[0086] The value of SD is generally 0.2 to 0.3, and T is the total time length of the signal.

[0087] When h 1k When the SD value requirement is met, let c1 = h 1k is the first IMF component of the original signal x(t). Subtracting c1 from the original signal x(t) yields the residual signal r1=x(t)-c1.

[0088] S13, taking r1 as the signal to be decomposed, repeat the above process, r2 = r1-c2, r3 = r2-c3, ... r n =r n-1 -c n Until the remaining signal r n If the information in the function has little significance for the content being studied or is already a monotonic function, stop.

[0089] After the above process, the signal can be expressed as:

[0090]

[0091] That is, after EMD, any signal can be decomposed into the sum of n IMFs and residuals, and the IMFs contain components of different frequency bands of the signal from high to low, and the frequency components contained in each frequency band vary with different signals.

[0092] S2. Perform Hilbert solutions on the obtained multiple intrinsic mode components IMFs respectively to obtain HHT marginal spectra.

[0093] Furthermore, step S2 is specifically as follows:

[0094] S21. Decompose the signal s(t) using the EMD method to obtain the following formula:

[0095]

[0096] c1(t),c2(t),…,c n (t) are the n IMF components obtained by decomposition. Each IMF contains components of different frequency bands of s(t). Their center frequencies are different and change with the signal itself.

[0097] S22, for IMF component c i (t) respectively perform Hilbert transform H[·], that is:

[0098]

[0099] Define the analytic signal z in the complex domain i (t) is:

[0100]

[0101] The instantaneous amplitude A i (t) is:

[0102]

[0103] Instantaneous phase for:

[0104]

[0105] Instantaneous frequency ω i (t) is:

[0106]

[0107] Then the expression of the original signal s(t) is:

[0108]

[0109] Here, Re represents the real part of a complex number.

[0110] The expansion of the above formula is the Hilbert spectrum, that is:

[0111]

[0112] It can be understood that the Hilbert spectrum accurately describes the variation of the signal amplitude over time and frequency. Further definition of the HHT marginal spectrum yields:

[0113]

[0114] The HHT marginal spectrum reflects the amplitude distribution of the signal at each instantaneous frequency point, that is, h(ω) can accurately reflect the true situation of the frequency components in the signal, so it can be used to reflect the frequency characteristics of the signal.

[0115] Furthermore, after performing empirical mode decomposition (EMD) on the local discharge signal, the frequency is divided into 800 frequency intervals, and the HHT marginal spectrum of each sample is calculated and statistically obtained; finally, the dimension of each discharge sample is (800, 1).

[0116] S3. Divide the HHT marginal spectra of all samples according to the set ratio to obtain the training set and the test set.

[0117] In a specific example, the HHT marginal spectrum set is divided into 4:1 to obtain a training set and a test set; the common partial discharge types of distribution equipment are tip discharge, suspension discharge, internal discharge, and surface discharge. The HHT marginal spectrum information of these four discharge types is shown in Table 1. After the 4:1 division, 245 training sets and 61 test sets are obtained.

[0118] Table 1 HHT marginal spectrum information

[0119] Partial discharge type Sample size Frequency interval number Tip discharge 106 800 Suspension discharge 110 800 Internal discharge 36 800 Surface discharge 54 800 total 306

[0120] S4. Use the training set to train the preset CNN model, and use the trained CNN model to perform defect detection on partial discharge.

[0121] 1D-CNN consists of input layer, convolution layer, pooling layer, fully connected layer and output layer. Classification results:

[0122] x output =f Softmax (f fc (f pooling (f conv (x input ))))

[0123] x input --Input set; f conv ()--convolutional layer, including convolution operation and nonlinear activation; f pooling ()--Pooling layer; f fc ()--Fully connected layer; f softmax ()--The output result obtained by using function calculation.

[0124] Since the marginal spectrum data of each discharge sample totals 800 data points, which are finally classified into 4 types of discharges, the number of input layer nodes is 800 and the number of output layer nodes is 4. The one-dimensional convolutional neural network structure is finally determined as follows Figure 1 As shown in Figure 2, the network consists of five convolutional layers, five pooling layers, one dropout layer, and one fully connected layer. The optimizer is Adam, the learning rate is 0.0002, and the number of iterations is 50.

[0125] The input vector dimension is (800, 1). After the convolution layer C1 with a convolution kernel size of 8 and a stride of 2, and the convolution layer C2 with a convolution kernel size of 8 and a stride of 2 and the same padding operation, the output map of size (199, 16) is obtained. After the downsampling operation of size 2, the output of the pooling layer P1 of size (99, 16) is obtained. Then, after the convolution and pooling operations with kernel sizes of 64×4 and 64×2, the feature (25, 64) is input to the convolution layer C4 with a size of 256×4 and the same expansion operation, and the pooling layer P3 with a size of 2 is used for pooling to obtain the output map of (6, 256). Then, after the convolution operation of 512×2 and the pooling operation of 512×2, the output map of (3, 512) is obtained. Finally, after the global average pooling, dropout layer, fully connected layer, and softmax classifier, 4 classified partial discharge defect states are output.

[0126] The training set is further divided into a training set and a validation set at a ratio of 3:1, that is, 184 samples in the training set, 61 samples in the validation set, and 61 samples in the test set. The training set is sent to the designed 1D-CNN model for training, and the model with the best performance in the validation set is selected as the final CNN model. The generalization ability of the model is verified by using the accuracy, precision, recall rate, and F1 value set of the test set.

[0127] It should be noted that, for the sake of convenience, the aforementioned method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited to the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously.

[0128] Based on the same idea as the partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network in the above embodiment, the present invention also provides a partial discharge defect diagnosis system based on HHT marginal spectrum and convolutional neural network, which can be used to execute the above partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network. For ease of explanation, the structural schematic diagram of the embodiment of the partial discharge defect diagnosis system based on HHT marginal spectrum and convolutional neural network only shows the parts related to the embodiment of the present invention. Those skilled in the art can understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0129] See also Figure 3In another embodiment of the present application, a partial discharge defect diagnosis system 100 based on HHT marginal spectrum and convolutional neural network is provided, the system comprising an EMD decomposition module 101, a Hilbert solution module 102, a data set partitioning module 103 and a model training module 104;

[0130] The EMD decomposition module 101 is used to decompose the collected partial discharge signal using the empirical mode decomposition (EMD) method to obtain a plurality of intrinsic mode components (IMFs);

[0131] The Hilbert solution module 102 is used to perform Hilbert solutions on the obtained multiple intrinsic modal components IMFs respectively to obtain HHT marginal spectra; the HHT marginal spectra reflect the amplitude distribution of the signal at each instantaneous frequency point, and are used to characterize the frequency characteristics of the reaction signal;

[0132] The data set division module 103 is used to divide the HHT marginal spectra of all samples according to a set ratio to obtain a training set and a test set;

[0133] The model training module 104 is used to train a preset CNN model using a training set, and use the trained CNN model to perform defect detection on partial discharge; the CNN model includes five convolutional layers, five pooling layers, one dropout layer and one fully connected layer, which are convolutional layer C1, convolutional layer C2, pooling layer P1, convolutional layer C3, pooling layer P2, convolutional layer C4, pooling layer P3, convolutional layer C5, pooling layer P4, pooling layer P5, dropout layer and fully connected layer in sequence.

[0134] See also Figure 4 In one embodiment, an electronic device for implementing a partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network is provided. The electronic device 200 may include a first processor 201, a first memory 202 and a bus, and may also include a computer program stored in the first memory 202 and executable on the first processor 201, such as a partial discharge defect diagnosis program 203 based on HHT marginal spectrum and convolutional neural network.

[0135] The first memory 202 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the first memory 202 can be an internal storage unit of the electronic device 200, such as a mobile hard disk of the electronic device 200. In other embodiments, the first memory 202 can also be an external storage device of the electronic device 200, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 200. Further, the first memory 202 can also include both an internal storage unit of the electronic device 200 and an external storage device. The first memory 202 can not only be used to store application software and various types of data installed in the electronic device 200, such as the code of the partial discharge defect diagnosis program 203 based on HHT marginal spectrum and convolutional neural network, but also can be used to temporarily store data that has been output or is to be output.

[0136] In some embodiments, the first processor 201 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The first processor 201 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes various functions and processes data of the electronic device 200 by running or executing programs or modules stored in the first memory 202, and calling data stored in the first memory 202.

[0137] Figure 4 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 4 The structure shown does not constitute a limitation on the electronic device 200 , and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0138] The partial discharge defect diagnosis program 203 based on HHT marginal spectrum and convolutional neural network stored in the first memory 202 of the electronic device 200 is a combination of multiple instructions. When running in the first processor 201, it can achieve:

[0139] The collected partial discharge signal is decomposed using empirical mode decomposition (EMD) to obtain multiple intrinsic mode components (IMFs);

[0140] Performing Hilbert solutions on the obtained multiple intrinsic modal components IMFs respectively to obtain HHT marginal spectra; the HHT marginal spectra reflect the amplitude distribution of the signal at each instantaneous frequency point and are used to characterize the frequency characteristics of the response signal;

[0141] The HHT marginal spectra of all samples are divided according to the set ratio to obtain the training set and the test set;

[0142] The preset CNN model is trained using the training set, and the trained CNN model is used to perform defect detection on partial discharge; the CNN model includes five convolutional layers, five pooling layers, one dropout layer and one fully connected layer, which are convolutional layer C1, convolutional layer C2, pooling layer P1, convolutional layer C3, pooling layer P2, convolutional layer C4, pooling layer P3, convolutional layer C5, pooling layer P4, pooling layer P5, dropout layer and fully connected layer, respectively.

[0143] Furthermore, if the module / unit integrated in the electronic device 200 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0144] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0145] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0146] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network, characterized in that: The steps include: Performing empirical mode decomposition (EMD) operation on the collected partial discharge signal to obtain multiple intrinsic mode components (IMFs); Performing Hilbert solutions on the obtained multiple intrinsic modal components IMFs respectively to obtain HHT marginal spectra; the HHT marginal spectra reflect the amplitude distribution of the signal at each instantaneous frequency point and are used to characterize the frequency characteristics of the response signal; The HHT marginal spectra of all samples are divided according to the set ratio to obtain the training set and the test set; The preset CNN model is trained using the training set, and the trained CNN model is used to perform defect detection on partial discharge; the CNN model includes five convolutional layers, five pooling layers, one dropout layer and one fully connected layer, which are convolutional layer C1, convolutional layer C2, pooling layer P1, convolutional layer C3, pooling layer P2, convolutional layer C4, pooling layer P3, convolutional layer C5, pooling layer P4, pooling layer P5, dropout layer and fully connected layer, respectively.

2. The partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network according to claim 1 is characterized in that: The collected partial discharge signal is subjected to an empirical mode decomposition (EMD) operation to obtain intrinsic mode components (IMFs), which are specifically: The local maximum and minimum points of the original signal x(t) are fitted into the upper envelope U(t) and lower envelope L(t) of the original signal x(t), and the mean m of the two envelopes is calculated. 11 (t); Calculate the original signal x(t) and the mean m 11 The difference h of (t) 11 (t): h 11 (t)=x(t)-m 11 (t) Judge h 11 (t) Whether the IMF conditions are met, if h 11 (t) is not an IMF, repeat the above process k times until the difference h 1k (t) Satisfy IMF conditions; h 1k (t)=h 1(k-1) (t)-m 1k (t) Among them, h 1(k-1) (t) represents the signal obtained by the k-1th screening; h 1k (t) represents the signal obtained by the kth screening; m 1k (t) represents h 1(k-1) (t) average envelope; The specific method of judging whether the IMF condition is met is to use the standard deviation coefficient SD value to judge whether each result is an IMF component; that is: Where T is the total time length of the signal, and t represents the time point; When h 1k When the SD value requirement is met, let c1 = h 1k is the first IMF component of the original signal x(t). Subtracting c1 from the original signal x(t) yields the residual signal r1=x(t)-c1. Repeat the above process with r1 as the signal to be decomposed, r2 = r1-c2, r3 = r2-c3, ... r n =r n-1 -c n ; until the remaining signal r n is a monotonic function; After the above process, the signal x(t) is expressed as: That is, after EMD, any signal can be decomposed into the sum of n IMFs and residuals, and the IMFs contain components of different frequency bands of the signal from high to low, and the frequency components contained in each frequency band vary with different signals.

3. The partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network according to claim 2 is characterized in that: The SD value is 0.2~0.

3.

4. The partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network according to claim 1 is characterized in that: The Hilbert transform is performed on the obtained multiple intrinsic modal components IMFs to obtain the HHT marginal spectrum, which is specifically: Using the EMD method to decompose the signal s(t) we get the following formula: c1(t),c2(t),…,c n (t) are the n IMF components obtained by decomposition; each IMF component contains components of different frequency bands of s(t), and their center frequencies are different and change with the signal itself; For IMF component c i (t) respectively perform Hilbert transform H[·], that is: Define the analytic signal z in the complex domain i (t) is: The instantaneous amplitude A i (t) is: Instantaneous phase for: Instantaneous frequency ω i (t) is: Then the expression of the original signal s(t) is: Among them, Re means taking the real part of the complex number; The expansion of the above formula is the Hilbert spectrum, that is: The Hilbert spectrum accurately describes the variation of the signal amplitude over time and frequency. The HHT marginal spectrum is further defined as follows: After the partial discharge signal is decomposed by EMD, the frequency is divided into multiple frequency intervals, and the HHT marginal spectrum of each sample is calculated and statistically obtained.

5. The partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network according to claim 1 is characterized in that: Use the training set to train the preset CNN model, specifically: The marginal spectrum data of each discharge sample has a total of 800 data points, which are finally classified into 4 types of discharges. Therefore, the number of input layer nodes is 800 and the number of output layer nodes is 4. The input vector dimension is (800, 1). After the convolution layer C1 with a convolution kernel size of 8 and a step size of 2 and the convolution layer C2 with a convolution kernel size of 8, a step size of 2 and a padding operation of the same, an output mapping of size (199, 16) is obtained. After a downsampling operation of size 2, the pooling layer P1 output of (99, 16) is obtained. Then, after convolution and pooling operations with kernel sizes of 64×4 and 64×2, the output is obtained. The features of (25, 64) are input into the convolution layer C4 with a size of 256×4 and the same expansion operation, and are pooled using the pooling layer P3 with a size of 2 to obtain an output map of (6, 256). Subsequently, a 512×2 convolution operation and a 512×2 pooling operation are performed to obtain an output map of (3, 512). Finally, four classified partial discharge defect states are output after global average pooling, a dropout layer, a fully connected layer, and a softmax classifier. The partial discharge defect states include tip discharge, suspended discharge, internal discharge, and surface discharge.

6. The partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network according to claim 1 is characterized in that: When training a CNN model, The optimizer is Adam, the learning rate is 0.0002, and the number of iterations is 50.

7. The partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network according to claim 1 is characterized in that: It also includes the steps of model validation, specifically: The training set is further divided into a training set and a validation set according to a set ratio. The training set is sent to the designed CNN model for training, and the model with the best performance in the validation set is selected as the final CNN model. The generalization ability of the model is verified by using the accuracy, precision, recall and F1 value of the test set.

8. The partial discharge defect diagnosis system based on HHT marginal spectrum and convolutional neural network is characterized by: A partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network applied to any one of claims 1-7, comprising an EMD decomposition module, a Hilbert solution module, a data set partitioning module and a model training module; The EMD decomposition module is used to perform an empirical mode decomposition (EMD) operation on the collected partial discharge signal to obtain a plurality of intrinsic mode components (IMFs); The Hilbert solution module is used to perform Hilbert solutions on the obtained multiple intrinsic modal components IMFs respectively to obtain HHT marginal spectra; the HHT marginal spectra reflect the amplitude distribution of the signal at each instantaneous frequency point and are used to characterize the frequency characteristics of the reaction signal; The data set partitioning module is used to partition the HHT marginal spectra of all samples according to a set ratio to obtain a training set and a test set; The model training module is used to train a preset CNN model using a training set, and use the trained CNN model to perform defect detection on partial discharge; the CNN model includes five convolutional layers, five pooling layers, a dropout layer and a fully connected layer, which are convolutional layer C1, convolutional layer C2, pooling layer P1, convolutional layer C3, pooling layer P2, convolutional layer C4, pooling layer P3, convolutional layer C5, pooling layer P4, pooling layer P5, dropout layer and fully connected layer in sequence.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a program, characterized in that: When the program is executed by a processor, the partial discharge defect diagnosis method based on HHT marginal spectrum and convolutional neural network described in any one of claims 1 to 7 is implemented.

Images