A transformer fault diagnosis method and system

Abstract

The present invention relates to a transformer fault diagnosis method and system in the technical field of transformer detection. The method and system comprise the following steps: collecting partial discharge signal training samples of the transformer and preprocessing the partial discharge signal training samples to obtain first feature extraction data; optimizing a neural network based on the first feature extraction data and a fruit fly algorithm to generate a neural network optimization model; collecting partial discharge signal test samples of the transformer and preprocessing the partial discharge signal test samples to obtain second feature extraction data; and inputting the second feature extraction data into the neural network optimization model to obtain a fault diagnosis result. The method solves the problem that, during the fault diagnosis process of the transformer, its feature extraction energy is lost and the formal signal characteristics cannot be objectively reflected.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer detection, and in particular to a transformer fault diagnosis method and system. Background Art

[0002] Partial discharge fault type is one of the important technical indicators for evaluating the insulation condition of high-voltage power equipment. It not only reflects the insulation condition level but also can estimate the service life of power equipment insulation. Therefore, research on transformer partial discharge fault identification algorithms is particularly important.

[0003] In recent years, the processing and analysis of discharge signals has mostly shifted from Fourier transforms to wavelet transforms, which can perform time-frequency localized analysis. Both methods have good processing effects on stationary signals, but these algorithms cannot adapt to changes in the time and frequency of the signal itself and cannot objectively reflect the true signal characteristics. BP neural networks suffer from unavoidable problems such as slow convergence and susceptibility to local minima. Furthermore, support vector machines are not suitable for large-sample, multi-class classification problems. Summary of the Invention

[0004] The present invention addresses the shortcomings of the prior art and provides a transformer fault diagnosis method and system, which solves the problem that during the transformer fault diagnosis process, its feature extraction energy loss cannot objectively reflect the formal signal characteristics.

[0005] In order to solve the above technical problems, the present invention is solved by the following technical solutions:

[0006] A transformer fault diagnosis method comprises the following steps:

[0007] Collecting partial discharge signal training samples of the transformer and preprocessing the partial discharge signal training samples to obtain first feature extraction data;

[0008] Generating a neural network optimization model based on the first feature extraction data and the fruit fly algorithm to optimize the neural network;

[0009] Collecting partial discharge signal test samples of the transformer and preprocessing the partial discharge signal test samples to obtain second feature extraction data;

[0010] The second feature extraction data is input into a neural network optimization model to obtain a fault diagnosis result.

[0011] Optionally, preprocessing the partial discharge signal training sample to obtain first feature extraction data includes the following steps:

[0012] Obtaining a signal time series 1 of the partial discharge signal training sample and performing a Hilbert-Huang transform to obtain a Hilbert training spectrum;

[0013] Based on the Hilbert training spectrum, data clustering calculation is performed to obtain first feature extraction data.

[0014] Optionally, obtaining a signal time series 1 of the partial discharge signal training sample and performing a Hilbert-Huang transform to obtain a Hilbert training spectrum comprises the following steps:

[0015] Performing empirical mode decomposition on the signal time series to obtain a plurality of intrinsic mode training functions;

[0016] Performing Hilbert transform on a plurality of the intrinsic mode training functions, and calculating instantaneous frequencies to generate a plurality of Hilbert training spectra.

[0017] Optionally, performing data clustering calculation to obtain first feature extraction data includes the following steps:

[0018] Summarizing a number of the Hilbert training spectra to obtain a training marginal spectrum;

[0019] The training marginal spectrum is clustered according to the index, and combined with the Hilbert energy spectrum to calculate the first feature extraction data.

[0020] Optionally, generating a neural network optimization model based on the first feature extraction data and the fruit fly algorithm to optimize the neural network includes the following steps:

[0021] Constructing a neural network, generating a feature matrix from the first feature extraction data and inputting the matrix into the neural network for training;

[0022] The fruit fly algorithm is used to obtain the optimal value of the set of neural network weights and thresholds, and a neural network optimization model is obtained.

[0023] Optionally, extracting the first feature data to generate a feature matrix and inputting the matrix into a neural network for training includes the following steps:

[0024] The first feature data is used to generate first feature vector data, and the first feature vector data is normalized to obtain a feature matrix, which is input into a neural network.

[0025] Optionally, collecting partial discharge signal test samples of the transformer and preprocessing the partial discharge signal test samples to obtain second feature extraction data includes the following steps:

[0026] Obtaining a second signal time series of the partial discharge signal test sample and performing a Hilbert-Huang transform to obtain a Hilbert test spectrum;

[0027] Based on the Hilbert test spectrum, data clustering calculation and feature vectorization are performed to obtain second feature extraction data.

[0028] Optionally, obtaining a second signal time series of the partial discharge signal test sample and performing a Hilbert-Huang transform to obtain a Hilbert test spectrum includes the following steps:

[0029] Performing empirical mode decomposition on the signal time series 2 to obtain a number of intrinsic mode test functions;

[0030] A plurality of the intrinsic modal test functions are subjected to Hilbert transformation, and the instantaneous frequencies are calculated to generate a plurality of Hilbert test spectra.

[0031] Optionally, performing data clustering calculation and feature vectorization based on the Hilbert test spectrum to obtain second feature extraction data includes the following steps:

[0032] Aggregating a number of the Hilbert test spectra to obtain a test marginal spectrum;

[0033] The test marginal spectrum is clustered according to the index, and the Hilbert energy value is calculated in combination with the Hilbert energy spectrum, and the feature vector is quantized to obtain the second feature extraction data.

[0034] A transformer fault diagnosis system includes a first preprocessing unit, an optimization unit, a second preprocessing unit and a diagnosis unit;

[0035] The first preprocessing unit is used to collect partial discharge signal training samples of the transformer and preprocess the partial discharge signal training samples to obtain first feature extraction data;

[0036] The optimization unit is used to optimize the neural network based on the first feature extraction data and the fruit fly algorithm to generate a neural network optimization model;

[0037] The second preprocessing unit is used to collect partial discharge signal test samples of the transformer and preprocess the partial discharge signal test samples to obtain second feature extraction data;

[0038] The diagnosis unit is used to input the second feature extraction data into a neural network optimization model to obtain a fault diagnosis result.

[0039] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

[0040] The signal characteristics of partial discharge signals are analyzed by Hilbert spectrum, and the Hilbert energy clustering algorithm is proposed to improve the separability of the eigenvalue itself, making the partial discharge fault classification effect more obvious. On the other hand, the PNN neural network is combined as a classifier for partial discharge fault identification, which has faster calculation and recognition speed and higher recognition rate, and has better scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0042] Figure 1 This is a flow chart of a transformer fault diagnosis method proposed in the first embodiment;

[0043] Figure 2 This is a schematic diagram of the probabilistic neural network topology proposed in the first embodiment. DETAILED DESCRIPTION

[0044] The present invention will be further described in detail below with reference to the examples. The following examples are intended to explain the present invention but the present invention is not limited to the following examples.

[0045] Example 1

[0046] like Figure 1 As shown, a transformer fault diagnosis method includes the following steps: collecting partial discharge signal training samples of the transformer and preprocessing the partial discharge signal training samples to obtain first feature extraction data. Specifically, the method includes the following steps: obtaining a signal time series 1 of the partial discharge signal training samples and performing a Hilbert-Huang transform to obtain a Hilbert training spectrum; and performing data clustering calculation based on the Hilbert training spectrum to obtain the first feature extraction data.

[0047] First, the partial discharge signal of the transformer is acquired through a sensor. Specifically, a high-frequency partial discharge signal acquisition device is used to collect four typical partial discharge defect signals: suspended discharge in oil, surface discharge in oil, air gap discharge in oil, and needle-plate discharge in air. Then, the signal time series F(t) of these four partial discharge signals is subjected to Hilbert-Huang transform. Specifically, the signal time series is subjected to empirical mode decomposition to obtain several inherent mode training functions. The several inherent mode training functions are subjected to Hilbert transform, and the instantaneous frequencies are calculated to generate several Hilbert training spectra.

[0048] A single Hilbert transform will filter out the harmonics in the signal. Therefore, empirical mode decomposition is combined to filter out the intrinsic mode training functions that meet the conditions from the original signal. The formula for performing the Hilbert transform on several intrinsic mode training functions is: And Y(t) is X(t) and Convolution, PV is the Cauchy principal value. By this definition, F(t) and Y(t) form a conjugate complex pair, so we can get an analytical signal Z(t), as follows:

[0049] Z(t)=F(t)+iY(t)=a(t)e iθ(t) ;

[0050]

[0051] Where F and Y are complex numbers, i indicates that Z is a complex number, Y is the imaginary part of Z, a(t) is the amplitude of the following trend, and θ(t) is the phase function, from which the instantaneous frequency ω can be calculated. The calculation formula is

[0052] After calculating the instantaneous frequency, the original function can be expressed as a time-frequency function, and the expression formula is: Where X is the original signal, which is decomposed into n intrinsic modal training functions. The Hilbert transform of the modal training functions that meet the conditions is used to obtain Z. Z is a complex function of frequency and time. Therefore, to obtain X again, it is necessary to integrate the time of multiple modal training functions and then accumulate them. In this way, X is the dependent variable, and frequency and time are independent variables. The partial discharge signal is further expressed as a function of amplitude a, instantaneous frequency ω and time t, that is, the Hilbert training spectrum H(ω, t).

[0053] Then, based on the Hilbert training spectrum, data clustering calculation is performed to obtain the first feature extraction data. Specifically, the following steps are included: aggregating several Hilbert training spectra to obtain a training marginal spectrum; clustering the training marginal spectrum according to the index, and combining the Hilbert energy spectrum calculation to obtain the first feature extraction data.

[0054] The Hilbert training spectrum is summarized into the training marginal spectrum. The calculation method is to integrate the Hilbert spectrum on the time axis, which describes the distribution of amplitude or energy on the frequency axis. Then, the training marginal spectrum with time-frequency information is split according to the exponential power from low to high frequency. Combined with the calculation formula of the Hilbert energy spectrum, that is, the Hilbert energy clustering algorithm, we have: Among them, H(ω, t) is a discrete signal. After squaring H(ω, t) and integrating it with time, it can be expressed as H 2The accumulation of discrete data points of (ω, t) removes the time variable. It should be noted that the energy value represented by ES is only related to the frequency, and the energy value is the amplitude of the electrical signal. ES(ω) is the Hilbert energy value. After sampling, the discrete points are summed and clustered. At the same time, combined with the Hilbert energy spectrum, the energy eigenvalue of the Hilbert training spectrum is extracted according to the exponential distribution family, that is, the first feature extraction data.

[0055] Then, based on the first feature extraction data and the fruit fly algorithm, the neural network is optimized to generate a neural network optimization model. Specifically, a neural network is constructed, and the first feature extraction data is used to generate a feature matrix and input into the neural network for training, that is, the first feature data is generated into the first feature vector data, and the first feature vector data is normalized to obtain the feature matrix, and input into the neural network. The PNN neural network is constructed using the newpnn function in Matlab. The function call format is net=newpnn(P,T,spread), where P is the input matrix, T is the target matrix, spread is the expansion speed, and net is the trained neural network model.

[0056] like Figure 2 As shown, further, the first feature extraction data are all represented by feature vectors, and then all feature vectors constitute the feature matrix Xmn as the input of the neural network, m is the eigenvalue dimension, the corresponding input layer has m neurons, and n is the number of input samples. It should be noted that the core of the pattern layer is based on the Gaussian kernel function under the Euclidean distance. The distance between the input feature sample and the training sample is obtained by the radial basis nonlinear mapping of the Gaussian kernel function. It is an important step to achieve the high-precision characteristics of the PNN nonlinear algorithm. The expression is as follows: Among them, δ j is the smoothing factor, which determines the different shapes of the bell curve; W j It can be regarded as the weight of the connection from the input layer to the pattern layer.

[0057] On the other hand, the summation layer summarizes the probabilities of samples of the same category output by the pattern layer. Therefore, the number of neurons in the summation layer corresponds to the fault category k of the training sample. The probability density function of each category using the Parzen window is: Among them, f kn That is the probability summary of the summation layer, n k is the number of samples in the Kth category; X mi is the i-th input vector; X kj is the j-th training vector for fault mode k.

[0058] The number of neurons l in the competitive layer depends on the type or combination of output classifications. The competitive layer takes a group of types with the maximum posterior probability density obtained by the summation layer as the final output of the network.

[0059] For example, set the category vectors of suspended discharge in oil, surface discharge in oil, air gap discharge in oil and needle-plate discharge in air, (1, 0, 0, 0) corresponds to the suspended discharge state in oil, (0, 1, 0, 0) corresponds to the surface discharge state in oil, (0, 0, 1, 0) corresponds to the air gap discharge state in oil, and (0, 0, 0, 1) corresponds to the needle-plate discharge state in air. Then, the characteristic vectors of these four groups of states constitute the input matrix I, the corresponding category vectors constitute the target matrix T, and training samples are selected to train the PNN neural network.

[0060] The fruit fly algorithm is used to obtain the optimal value of the set of neural network weights and thresholds to obtain the neural network optimization model. Specifically, the fruit fly group size is set, the number of generations is optimized, and the flight distance Fr is randomly initialized to the fruit fly group position (Init X_axis, Y_axis). The fruit fly group position is the set of weights and thresholds in the current iteration of the PNN neural network. Then, the fruit fly individual is given a random direction and distance to search for food using its sense of smell. The calculation expression is: Among them, X j Y is the distance that the fruit fly moves in the random direction of the horizontal coordinate when it uses its sense of smell to find food. j The vertical coordinate distance that a fruit fly moves in a random direction when it uses its sense of smell to find food.

[0061] Since the location of the food is unknown, the distance from the origin (D) is estimated first, and then the taste concentration judgment value (S i ), the formula is as follows: Then the taste concentration judgment value (S i ) is substituted into the smell concentration determination function (Fitnessfunction) to find the smell concentration (smell i ), the formula is as follows smell i =Function(S i ), then find the fruit fly with the highest flavor concentration in this group of fruit flies, retain the optimal flavor concentration value and x, y coordinates, and then the fruit fly group uses vision to fly to this location, forming a new cluster. Then, the fruit fly iterative optimization process begins, repeating the steps to obtain the optimal flavor concentration value and determining whether the flavor concentration is better than the previous iteration's flavor concentration. If so, it is retained. For the PNN model, the root mean square error (RMSE) between the output value of the training sample and the target value is used as the fitness function to find the optimal δ. Thus, the PNN neural network is optimized through the fruit fly algorithm, resulting in a neural network optimization model.

[0062] Then the staff collected partial discharge signal test samples of the transformer and preprocessed the partial discharge signal test samples to obtain the second feature extraction data. Specifically, the signal time series 2 of the partial discharge signal test sample was obtained and Hilbert-Huang transform was performed to obtain the Hilbert test spectrum; based on the Hilbert test spectrum, data clustering calculation and feature vectorization were performed to obtain the second feature extraction data. Furthermore, the signal time series 2 was subjected to empirical mode decomposition to obtain several inherent modal test functions; several inherent modal test functions were subjected to Hilbert transform, and the instantaneous frequency was calculated to generate several Hilbert test spectra; several Hilbert test spectra were summarized to obtain a test marginal spectrum; the test marginal spectrum was clustered according to the index, and the Hilbert energy value was calculated in combination with the Hilbert energy spectrum, and the feature vectorization was performed to obtain the second feature extraction data.

[0063] It should be noted that the preprocessing of the partial discharge signal test sample is the same as the preprocessing method of the partial discharge signal test sample, and will not be described in detail here. Then, the second feature extraction data is input into the neural network optimization model to obtain the fault diagnosis result. Through the method of this embodiment, during the transformer fault diagnosis process, the energy loss in the feature extraction process is avoided, the resolution of the characteristic value is improved, and the calculation speed is fast, thereby improving the speed and accuracy of partial discharge fault identification.

[0064] Example 2

[0065] A transformer fault diagnosis system includes a first preprocessing unit, an optimization unit, a second preprocessing unit, and a diagnosis unit; the first preprocessing unit is used to collect partial discharge signal training samples of the transformer and preprocess the partial discharge signal training samples to obtain first feature extraction data; the optimization unit is used to optimize a neural network based on the first feature extraction data and a fruit fly algorithm to generate a neural network optimization model; the second preprocessing unit is used to collect partial discharge signal test samples of the transformer and preprocess the partial discharge signal test samples to obtain second feature extraction data; the diagnosis unit is used to input the second feature extraction data into the neural network optimization model to obtain a fault diagnosis result. It should be noted that the transformer fault diagnosis system of this embodiment can be used to execute the transformer fault diagnosis method described in Example 1.

[0066] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the transformer fault diagnosis method described in the first embodiment is executed.

[0067] More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wire segments, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0068] In this application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. Furthermore, in this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, electrical, optical, RF, or any suitable combination thereof.

[0069] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules, subassemblies, or units is merely a logical functional division. In actual implementation, other division methods may be used, such as multiple units, subassemblies, or components can be combined or integrated into another device, or some features can be ignored or not performed.

[0070] The units may or may not be physically separate, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0071] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0072] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above-mentioned functions defined in the method of the present application are executed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of an electrical, magnetic, optical, electromagnetic, infrared segment, or semiconductor, or any combination of the above.

[0073] The flow charts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0074] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention shall be covered by the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A transformer fault diagnosis method, characterized in that: The following steps are involved: Collecting partial discharge signal training samples of a transformer and preprocessing the partial discharge signal training samples to obtain first feature extraction data specifically includes the following steps: obtaining a signal time series 1 of the partial discharge signal training samples and performing a Hilbert-Huang transform to obtain a Hilbert training spectrum; performing data clustering calculation based on the Hilbert training spectrum to obtain first feature extraction data; wherein obtaining a signal time series 1 of the partial discharge signal training samples and performing a Hilbert-Huang transform to obtain a Hilbert training spectrum includes the following steps: performing an empirical mode decomposition on the signal time series 1 to obtain a plurality of intrinsic mode training functions; performing a Hilbert transform on the plurality of intrinsic mode training functions and calculating instantaneous frequencies to generate a plurality of Hilbert training spectra; The calculation formula of Hilbert training spectrum is H( , t) = , n is the number of intrinsic mode training functions; is the trend amplitude; i represents a complex number of the analytical signal; represents the instantaneous frequency; represents instantaneous frequency; t represents time; Generating a neural network optimization model based on the first feature extraction data and the fruit fly algorithm to optimize the neural network; Collecting partial discharge signal test samples of the transformer and preprocessing the partial discharge signal test samples to obtain second feature extraction data; The second feature extraction data is input into a neural network optimization model to obtain a fault diagnosis result.

2. A transformer fault diagnosis method according to claim 1, characterized in that: Performing data clustering calculation to obtain first feature extraction data includes the following steps: Summarizing a number of the Hilbert training spectra to obtain a training marginal spectrum; The training marginal spectrum is clustered according to the index, and combined with the Hilbert energy spectrum to calculate the first feature extraction data.

3. A transformer fault diagnosis method according to claim 1, characterized in that: Generating a neural network optimization model based on the first feature extraction data and the fruit fly algorithm to optimize the neural network includes the following steps: Constructing a neural network, generating a feature matrix from the first feature extraction data and inputting the matrix into the neural network for training; The fruit fly algorithm is used to obtain the optimal value of the set of neural network weights and thresholds, and a neural network optimization model is obtained.

4. A transformer fault diagnosis method according to claim 3, characterized in that: The first feature extraction data generates a feature matrix and inputs it into a neural network for training, comprising the following steps: The first feature extraction data is used to generate first feature vector data, and the first feature vector data is normalized to obtain a feature matrix, which is input into a neural network.

5. A transformer fault diagnosis method according to claim 1, characterized in that: Collecting partial discharge signal test samples of the transformer and preprocessing the partial discharge signal test samples to obtain second feature extraction data includes the following steps: Obtaining a second signal time series of the partial discharge signal test sample and performing a Hilbert-Huang transform to obtain a Hilbert test spectrum; Based on the Hilbert test spectrum, data clustering calculation and feature vectorization are performed to obtain second feature extraction data.

6. A transformer fault diagnosis method according to claim 5, characterized in that: Obtaining a second signal time series of the partial discharge signal test sample and performing a Hilbert-Huang transform to obtain a Hilbert test spectrum includes the following steps: Performing empirical mode decomposition on the signal time series 2 to obtain a number of intrinsic mode test functions; A plurality of the intrinsic modal test functions are subjected to Hilbert transformation, and the instantaneous frequencies are calculated to generate a plurality of Hilbert test spectra.

7. A transformer fault diagnosis method according to claim 5, characterized in that: Based on the Hilbert test spectrum, data clustering calculation and feature vectorization are performed to obtain second feature extraction data, including the following steps: Aggregating a number of the Hilbert test spectra to obtain a test marginal spectrum; The test marginal spectrum is clustered according to the index, and the Hilbert energy value is calculated in combination with the Hilbert energy spectrum, and the feature vector is quantized to obtain the second feature extraction data.

8. A transformer fault diagnosis system, characterized in that: The transformer fault diagnosis system implements the transformer fault diagnosis method according to any one of claims 1 to 7, comprising a first preprocessing unit, an optimization unit, a second preprocessing unit, and a diagnosis unit; The first preprocessing unit is used to collect partial discharge signal training samples of the transformer and preprocess the partial discharge signal training samples to obtain first feature extraction data; The optimization unit is used to optimize the neural network based on the first feature extraction data and the fruit fly algorithm to generate a neural network optimization model; The second preprocessing unit is used to collect partial discharge signal test samples of the transformer and preprocess the partial discharge signal test samples to obtain second feature extraction data; The diagnosis unit is used to input the second feature extraction data into a neural network optimization model to obtain a fault diagnosis result.

Images