Signal noise suppression method and system, storage medium and electronic equipment

Abstract

The invention provides a signal noise suppression method and system, a storage medium and electronic equipment, and the method comprises the steps: obtaining the original data of a partial discharge signal of a switch cabinet, and carrying out the feature extraction of the partial discharge signal; performing noise suppression on the partial discharge signal after feature extraction, determining a signal-to-noise ratio and a mean square error, and judging the quality of a noise suppression effect according to the signal-to-noise ratio and the mean square error; and optimizing a noise suppression effect according to a comparison result of the noise-suppressed data and the noise-free signal. According to the invention, feature extraction is carried out on the partial discharge signal and then noise suppression is carried out, so that the accuracy of signal feature extraction is improved; the noise suppression effect is continuously optimized according to the comparison result of the data after noise suppression and the noiseless signal, and the noise suppression precision of the complex noise feature in the partial discharge signal of the switch cabinet is improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of signal noise suppression, and in particular to a signal noise suppression method, system, storage medium and electronic device. Background Art

[0002] Partial discharge in switchgear refers to the partial discharge phenomenon caused by insulation defects or excessive electric field strength in high-voltage switchgear. This discharge will generate certain electromagnetic signals, which can be detected and monitored by sensors. However, due to the complexity of the on-site environment and the electromagnetic interference of the equipment itself, the detected partial discharge signals are often interfered by various noises. For example, electromagnetic interference, equipment vibration, temperature changes and other factors may affect the signal quality. In order to accurately identify and analyze partial discharge signals, it is necessary to suppress the noise of the signals. In the context of noise suppression, various signal processing techniques, such as filtering, wavelet transform, time-frequency analysis, etc., can be used to denoise partial discharge signals.

[0003] The commonly used noise suppression methods are as follows: wavelet threshold denoising and empirical mode decomposition denoising. The wavelet threshold denoising method uses wavelet transform to decompose the signal into frequency bands of different scales, and then performs threshold processing on each frequency band based on the estimation of the signal-to-noise ratio, setting the low-amplitude wavelet coefficients to zero, thereby achieving the purpose of removing noise; the empirical mode decomposition denoising method is an adaptive signal decomposition method, which decomposes the signal into several intrinsic mode functions (IMFs), and then removes noise by filtering and reconstructing the IMFs.

[0004] Traditional noise suppression methods in the prior art are difficult to adapt to complex, nonlinear noises and may cause signal information loss. At the same time, they also have single requirements for noise types, need to determine the wavelet basis function, are sensitive to noise types and signal characteristics, have weak adaptability, and cannot accurately extract signal characteristics. Summary of the invention

[0005] The present disclosure aims to at least solve the above-mentioned technical problems existing in the prior art.

[0006] To this end, an object of the present disclosure is to provide a signal noise suppression method, the signal noise suppression method comprising:

[0007] Acquire raw data of a partial discharge signal of a switch cabinet, and perform feature extraction on the partial discharge signal;

[0008] Perform noise suppression on the partial discharge signal after feature extraction, determine the signal-to-noise ratio and mean square error, and judge the quality of the noise suppression effect based on the signal-to-noise ratio and mean square error;

[0009] The noise suppression effect is optimized according to the comparison result between the noise suppressed data and the noise-free signal.

[0010] In some embodiments, extracting features from the partial discharge signal includes:

[0011] Decomposing the partial discharge signal to obtain a plurality of intrinsic mode functions;

[0012] The time domain signal in each intrinsic mode function is converted into a frequency domain signal, the spectral centroid of each intrinsic mode function is calculated, and the eigenvector of each intrinsic mode function is determined according to the position of the spectral centroid.

[0013] In some embodiments, decomposing the local discharge signal to obtain a plurality of intrinsic mode functions includes:

[0014] Optimize the parameter combination of the decomposition mode number K and the penalty factor α of the variational mode decomposition according to the weighted inclination;

[0015] The original data is decomposed based on the optimized parameter combination to obtain a plurality of intrinsic mode functions.

[0016] In some embodiments, converting the time domain signal in each intrinsic mode function into a frequency domain signal, calculating the spectral centroid of each intrinsic mode function, and determining the eigenvector of each intrinsic mode function according to the position of the spectral centroid includes:

[0017] The time domain signal in each intrinsic mode function is converted into a frequency domain signal, and the multi-frequency signal is converted into a single source signal;

[0018] The spectral centroid is determined according to the similarity of the single source signal in the time domain and in the frequency domain.

[0019] In some embodiments, performing noise suppression on the partial discharge signal after feature extraction, determining a signal-to-noise ratio and a mean square error, and judging the quality of the noise suppression effect according to the signal-to-noise ratio and the mean square error includes:

[0020] According to the noise suppression result, the signal-to-noise ratio and the mean square error are determined;

[0021] The quality of the noise suppression effect is judged based on the signal-to-noise ratio and the mean square error loss function.

[0022] In some embodiments, optimizing the noise suppression effect according to the comparison result between the noise suppressed data and the noise-free signal includes:

[0023] The raw data of the switchgear partial discharge signal is input into the generator, and the denoised signal is output;

[0024] Comparing the denoised signal with the noise-free signal, optimizing the parameters of the generator and the discriminator through adversarial training, and adjusting the weights of the generator and the discriminator according to the adversarial loss function;

[0025] Compare the similarity between the denoised signal output by the generator and the noise-free signal, and judge the noise suppression effect of the generator based on the peak signal-to-noise ratio.

[0026] In some embodiments, the raw data includes at least one of a voltage signal, a current signal, a temperature value, a humidity level, and a sound signal.

[0027] Another object of the present disclosure is to provide a noise suppression system for partial discharge signals of a switch cabinet, comprising:

[0028] An acquisition module, used to acquire raw data of a partial discharge signal of a switch cabinet and perform feature extraction on the partial discharge signal;

[0029] A judgment module is used to perform noise suppression on the partial discharge signal after feature extraction, determine the signal-to-noise ratio and the mean square error, and judge the quality of the noise suppression effect according to the signal-to-noise ratio and the mean square error;

[0030] An optimization module for optimizing the noise suppression effect according to the quality of the noise suppression effect.

[0031] Another object of the present disclosure is to provide a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the noise suppression method of the above method is performed.

[0032] Another object of the present disclosure is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned noise suppression method when executing the program.

[0033] The signal noise suppression method, system, storage medium and electronic device provided by the embodiments of the present disclosure have the following beneficial effects:

[0034] The signal noise suppression method includes obtaining the original data of the partial discharge signal of the switch cabinet, and extracting the features of the partial discharge signal; performing noise suppression on the partial discharge signal after feature extraction, determining the signal-to-noise ratio and mean square error, and judging the quality of the noise suppression effect according to the signal-to-noise ratio and mean square error; optimizing the noise suppression effect according to the comparison results of the noise-suppressed data and the noise-free signal. The feature extraction of the partial discharge signal is first performed, and then the noise suppression is performed, so as to improve the accuracy of the signal feature extraction; the noise suppression effect is continuously optimized according to the comparison results of the noise-suppressed data and the noise-free signal, so as to improve the noise suppression accuracy of the complex noise features in the partial discharge signal of the switch cabinet. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 is a schematic diagram of steps of a method for suppressing noise of a signal in an embodiment of the present disclosure;

[0037] Figure 2 is a schematic diagram of steps of a method for suppressing noise of a signal in an embodiment of the present disclosure;

[0038] Figure 3 is a schematic diagram of steps of a method for suppressing noise of a signal in an embodiment of the present disclosure;

[0039] Figure 4 is a workflow diagram of a VMD algorithm of a signal noise suppression method in an embodiment of the present disclosure; DETAILED DESCRIPTION

[0040] Various aspects and features of the present disclosure are described herein with reference to the drawings.

[0041] It should be understood that various modifications may be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but only as an example of an embodiment. Other modifications within the scope and spirit of the present disclosure will occur to those skilled in the art.

[0042] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the general description of the present disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the present disclosure.

[0043] These and other characteristics of the present disclosure will become apparent from the following description of a preferred form of embodiment given as a non-limiting example, with reference to the accompanying drawings.

[0044] It should also be understood that, although the present disclosure has been described with reference to some specific examples, those skilled in the art will be able to realize many other equivalent forms of the present disclosure that have the features described in the claims and are therefore within the scope of protection defined thereby.

[0045] The above and other aspects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.

[0046] Specific embodiments of the present disclosure are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments claimed are merely examples of the present disclosure, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid obscuring the present disclosure with unnecessary or redundant details. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present disclosure in a variety of ways with substantially any suitable detailed structure.

[0047] The present disclosure provides a method for suppressing noise of a signal, such as Figure 1-Figure 4 As shown, the noise suppression method of the signal includes:

[0048] S1: Obtain the original data of the partial discharge signal of the switch cabinet and extract the features of the partial discharge signal.

[0049] Specifically, the partial discharge signal of the switch cabinet includes at least one of a voltage signal, a current signal, a temperature value, a humidity level and a sound signal; specifically, data is collected through sensors in the switch cabinet. The collected raw data is further preprocessed and denoised through a filter to make the data signal smoother; the data is then scaled to the same range and scale to eliminate the differences between the data so that different data can be compared and analyzed; and the processed raw data is classified and stored in the database in turn. De-noising and data normalization can improve the quality and accuracy of the data, reduce noise interference, and make the data more reliable; scaling the data to the same range and scale eliminates the differences between the data, ensures the consistency of the data, and facilitates comparison and analysis; the processed data is classified and stored in the database, which is helpful for data organization, management and retrieval, and improves the availability and accessibility of the data.

[0050] Among them, feature extraction of partial discharge signals includes:

[0051] S101: Decomposing the partial discharge signal to obtain a plurality of intrinsic mode functions (IMFs);

[0052] Specifically, it includes:

[0053] S1011: separating the partial discharge signal from the original data according to the parameter combination of the decomposition mode number K and the penalty factor α of the weighted slope optimization variational mode decomposition (VMD).

[0054] The calculation formula of weighted slope is:

[0055]

[0056] Among them, S i represents the skewness of the i-th eigenmode function, Bi represents the Bhattacharyya distance of the i-th eigenmode function;

[0057] S i The calculation formula is:

[0058]

[0059] Among them, x i is the time series of the signal, μ is the mean value of the signal, N is the number of sample points, and σ is the variance of the signal;

[0060] B i The calculation formula is:

[0061]

[0062] Where x is the signal after VMD decomposition, y is the original partial discharge signal, μ x is the average value of the signal after VMD decomposition, μ y is the average value of the original partial discharge signal, P x is the covariance matrix of the signal after VMD decomposition, P y is the covariance matrix of the original partial discharge signal.

[0063] The calculation formula of P is:

[0064]

[0065] Due to the short-term and steep characteristics of the partial discharge signal, the IMF with partial discharge signal has a higher skewness than the IMF without partial discharge signal. Therefore, the skewness can be used to determine whether the IMF contains a partial discharge signal. However, the skewness only depends on the distribution density of the signal. Using skewness as the objective function may lead to over-decomposition of VMD and loss of partial discharge signal features. The Bhattacharyya distance can represent the theoretical distance between two Gaussian distributions. The smaller the Bhattacharyya distance, the more similar the two distributions are. Combining the skewness and the Bhattacharyya distance between the decomposed signal and the original partial discharge signal, a weighted skewness W is proposed. S . Select W S As the objective function of VMD optimization, the partial discharge signal can be effectively separated from the noise and the characteristics of the signal can be fully preserved.

[0066] Specifically, VMD is related to the number of decomposed modes K and the penalty factor α, which determines the bandwidth of the IMF; if K is large, over-decomposition will occur, resulting in the loss of local discharge signal features; when K is small, under-compensation will occur, resulting in the aliasing of local discharge signals and noise in the same IMF. Particle Swarm Optimization (PSO) is a population-based global optimization search algorithm. It works on the principle that there is a population of candidate solutions, called particles. Each particle moves in the search space according to its own historical best position (called personal best) and the best position found by the entire particle (called global best). When an improved position is found, the personal best and the global best update their positions and guide the movement of the particles. The process is repeated until the predefined stopping criteria are met. When optimizing VMD with the particle swarm algorithm, the initial parameters of PSO and VMD (except K and α) are set first, and then the initial optimization range of parameters α and K is set. Finally, the parameter combination of α and K with the maximum weighted inclination is found as the optimal parameter and output. Formally, the problem of optimizing D dimensions, each particle position is X i =(x i1 ,x i2 ,...,x id ,...,x iD ), the speed is

[0067] V i =(v i1 ,v i2 ,...,v id ,...,v iD ), where i = 1, 2, ..., N, N is the particle size. The personal best value of the i-th particle and the global best value of the group are expressed as:

[0068] pbest i =(pbest i1 ,pbest i2 ,...,pbest id ,...,pbest iD )

[0069] and gbest i =(gbest1,gbest2,...,pbest d ,...,gbest D ), v id and x id After being initialized, v is updated according to the following formula id and x id :

[0070]

[0071] in is the velocity of particle i in the t+1th generation and dth dimension; is the current position of particle i in the t+1th generation and dth dimension; ω is the inertia weight, is the d-th dimension component of the individual optimal position of particle i in the t-th generation; is the d-th dimension component of the global optimal position of all particles in the t-th generation; r1 and r2 are two random values ​​independently generated in the range [0, 1], and c1 and c2 are acceleration coefficients that control the influence of personal cognition and global components.

[0072] Variational mode decomposition (VMD) can decompose complex discrete signals into simple modal components. The specific steps are as follows:

[0073] S10111: Construct a variational problem, and the variational problem is to minimize the sum of the estimated bandwidths of the IMFs. The constraint is that the sum of the IMFs is the original signal. The algorithm formula is as follows:

[0074]

[0075] Among them, {u k} is the set of IMFs, {ω k} is the set of center frequencies, δ(t) is the pulse function, j is the imaginary unit, t is the time, u k(t) For each eigenmode function obtained, ω k is the center frequency, f(t) is the original function, and K is the preset decomposition number.

[0076] S10112: The penalty factor α and the Lagrange multiplier λ(t) are used to transform the constrained variational problem into an unconstrained problem. α affects the reconstruction accuracy of the signal, and λ(t) maintains the strictness of the constraint conditions. The algorithm formula is as follows:

[0077]

[0078] Here, λ is the Lagrange multiplier.

[0079] S10113: The alternating direction multiplier method is used to solve the unconstrained problem. The algorithm formula for iterative update and center frequency of IMF is as follows:

[0080]

[0081] in, is the kth IMF with center frequency ω at the (n+1)th iteration; is the Fourier transform of the original function with variable ω, is the Fourier transform of the eigenmode function with variable ω and not the kth one, is the Fourier transform of the Lagrange multiplier with variable ω, ωk is the center frequency of the kth eigenmode function;

[0082]

[0083] in, is the center frequency of the kth IMF at the (n+1)th iteration, is the Fourier transform of the intrinsic mode function with variable ω.

[0084] The use of variational mode decomposition (VMD) is beneficial to the effective separation of signal frequencies.

[0085] S1012: Decomposing the original data based on the optimized parameter combination to obtain a plurality of intrinsic mode functions.

[0086] S102: converting the time domain signal in each intrinsic mode function into a frequency domain signal, calculating the spectral centroid of each intrinsic mode function, and determining the eigenvector of each intrinsic mode function according to the position of the spectral centroid.

[0087] Specifically include:

[0088] S1021: converting the time domain signal in each intrinsic mode function into a frequency domain signal, and converting the multi-frequency signal into a single source signal;

[0089] Specifically, the signals obtained from various switchgears may show significant variation. The amplitude peaks and positions of these signals are similar in the time domain, but vary greatly in the frequency domain. In addition, these signals are captured in a very short period of time using a high-frequency data acquisition module. Therefore, the position of the peaks and troughs in the pulse train alone is not a decisive factor in determining whether the signal represents a partial discharge type detection. The conversion of the time domain signal in each intrinsic mode function to the frequency domain signal is performed by the following formula,

[0090]

[0091] Where X(k) is the kth frequency domain component in the discrete Fourier transform (DFT) frequency domain, x n is the nth element of the time series signal, and N is the signal length in the IMF.

[0092] S1022: determining a spectral centroid according to similarities between the single source signal in the time domain and in the frequency domain;

[0093] Since the spectral centroid performs well in terms of robustness, is less sensitive to outliers, and is conducive to processing data containing noise values, the spectral centroid (SC) is calculated according to the following formula:

[0094]

[0095] Where |X(k)| is the amplitude at frequency k, and N is the length of the discrete Fourier transform result.

[0096] If the c value is large, it means that the acquired raw data contains a large number of signals with high-frequency components, which means that the corresponding frequency components should receive more attention. The frequency spectrum centroid is used to perform adaptive feature extraction on each IMF to better separate the partial discharge signal from the noise and speed up the feature extraction of the captured data.

[0097] Then, the two parts on both sides of the spectrum centroid c are evenly divided into several parts to construct more informative features. The feature matrix f of the i-th IMF is i is defined as:

[0098]

[0099] Where s is the number of uniform segments of the frequency feature, c is the spectral centroid of |X|, The frequency is The amplitude.

[0100] In S1, the original data of the partial discharge signal of the switch cabinet is obtained, the feature of the partial discharge signal is extracted, and the data of the partial discharge signal after the feature extraction is marked and annotated.

[0101] Specifically, the original data is labeled according to the different characteristics of the partial discharge signal, and annotated with labels to form multiple data sets. Each data set is further divided into a training group, a validation group, and a test group; 70% of each data set is a training group, 15% is a validation group, and 15% is a test group. The data in each data set is feature corrected using the Vott tool. By labeling and labeling data with different characteristics, the characteristics and meaning of the data can be accurately described, which is helpful for data analysis and model training; dividing the data set into a training group, a validation group, and a test group is conducive to evaluating the generalization ability of the model and improving the stability and generalization performance of the model; using the Vott tool for feature correction can further improve the quality and accuracy of the data and ensure the correctness and consistency of the annotation.

[0102] S2: Perform noise suppression on the partial discharge signal after feature extraction, determine the signal-to-noise ratio and the mean square error, and judge the quality of the noise suppression effect based on the signal-to-noise ratio and the mean square error.

[0103] Specifically, when deep learning is performed on the local discharge signal after feature extraction, the long short-term memory network architecture (LSTM) is first selected, and the number of layers and hidden units of the model is defined, and a recurrent neural network encoder is added. Further, the mean square error loss function is defined, and the stochastic gradient descent algorithm is selected as the optimizer, and the signal-to-noise ratio is added as an evaluation indicator; then the training group data is brought into the model for training, and the model parameters are updated through the back propagation algorithm. After each training cycle, the validation group is used for verification; finally, the test group is used to evaluate the model performance, and the signal-to-noise ratio is calculated. At the same time, the model output data is compared with the original data to adjust the model architecture and hyperparameters. The model architecture and hyperparameters are continuously optimized by deep learning of the local discharge signal after feature extraction to improve the noise suppression effect and output a signal with less noise as much as possible. By using architectures suitable for time series data processing, such as long short-term memory networks (LSTM), we can better capture the time series features in the data and improve the performance of the model; defining the mean square error loss function helps to measure the difference between the model output and the true value, and the stochastic gradient descent algorithm as an optimizer can help the model converge faster; adding the signal-to-noise ratio as an evaluation indicator can intuitively evaluate the performance of the model and help understand the effectiveness of the model's denoising; using the test group to evaluate the performance of the model and calculating indicators such as the signal-to-noise ratio can objectively evaluate the denoising effect of the model; comparing the model output data with the original data can help discover the advantages and disadvantages of the model, and then adjust the model architecture and hyperparameters to improve model performance.

[0104] The signal-to-noise ratio and the mean square error are determined, and the quality of the noise suppression effect is judged according to the signal-to-noise ratio and the mean square error, including:

[0105] S201: Determine a signal-to-noise ratio and a mean square error according to a noise suppression result.

[0106] The calculation formula of signal-to-noise ratio SNR is:

[0107]

[0108] Among them, x i is the original signal, y i is the denoised signal;

[0109] The original signal refers to the local discharge signal after feature extraction, and the denoised signal refers to the signal after deep learning (or after adversarial training optimization).

[0110] The calculation formula for mean square error MSE is:

[0111]

[0112] Where n is the number of samples, y i is the denoised signal, Predict values ​​for deep learned models.

[0113] S202: judging the quality of the noise suppression effect according to the signal-to-noise ratio and the mean square error loss function.

[0114] Specifically, the signal-to-noise ratio is the ratio of signal power to noise power, usually expressed in decibels (dB). The higher the signal-to-noise ratio, the clearer the signal is relative to the noise.

[0115] The mean square error loss function can also be used to evaluate the difference between signal and noise. The mean square error indicates the error between the signal and the expected output. The smaller the mean square error, the smaller the error between the signal and the expected output. The clearer the signal is relative to the noise and the smaller the mean square error is, the higher the quality of the noise suppression effect is.

[0116] Both the mean square error and the signal-to-noise ratio can be used as verification criteria. The mean square error between the results predicted by the model and the true label of the verification group data must be within a certain range, that is, it is considered qualified when the mean square error threshold is reached; the signal-to-noise ratio of the verification group data after being processed by the model needs to reach a certain level, and a minimum signal-to-noise ratio can be set to be considered qualified.

[0117] In S2, the use of new data that has not been trained and verified for testing can more realistically reflect the performance of the model in actual application scenarios and avoid overfitting; objective indicators such as signal-to-noise ratio and mean square error are used to evaluate the noise suppression effect of the model, which can quantitatively measure the performance of the model and provide a basis for model improvement; the use of visualization tools to display the evaluation results can intuitively present the processing effect of the model, making it easier for users to understand and analyze; the entire step focuses on the application effect of the model in actual application scenarios, which can provide reliable reference and technical support for the actual application of partial discharge signal detection in switch cabinets.

[0118] S3: Optimize the noise suppression effect according to the comparison result between the noise suppressed data and the noise-free signal.

[0119] Specifically include:

[0120] S301: the original data of the switch cabinet partial discharge signal is input into the generator, and the denoised signal is output;

[0121] S302: Compare the denoised signal with the noise-free signal, optimize the parameters of the generator and the discriminator through adversarial training, and adjust the weights of the generator and the discriminator according to the adversarial loss function;

[0122] Specifically, in the process of optimizing noise suppression, the goal of the generator is to convert the input noisy switchgear partial discharge signal into a denoised signal. The input of the generator can be the noisy switchgear partial discharge signal, that is, the original data after feature extraction. The generator outputs the denoised signal after processing. In adversarial training, the goal of the generator is to match the generated signal with the noise-free signal as much as possible to reduce noise and improve signal quality.

[0123] The calculation formula of the adversarial loss function is as follows:

[0124]

[0125] Among them, x is the denoised signal output by the generator, z is the random noise vector, G is the generator, D is the discriminator, and p data (x) is the distribution of the denoised signal output by the generator, p z (z) is the distribution of random noise signal, is the loss of the denoised signal output by the generator, is the loss of generated data.

[0126] The random noise vector z usually represents the noise-free signal or the input signal of the generator.

[0127] When adjusting the weights of the generator and the discriminator, it can also be implemented based on the adversarial loss function and other functions. Among them, other loss functions can be mean square error loss function (MSE), L1 loss function, perceptual loss function, etc.

[0128] Adversarial loss functions and other loss functions usually act on both the generator and the discriminator at the same time, and optimize the parameters of both through adversarial training. During the training process, the goal of the generator is to generate a denoised signal that is as realistic as possible, and the goal of the discriminator is to distinguish the signal generated by the generator from the real noise-free signal as accurately as possible. Through adversarial training, the weights of the generator and the discriminator are adjusted according to the gradient of the loss function to improve the quality of the signal generated by the generator.

[0129] Secondly, the adversarial loss function usually aims to minimize the difference between the generator and the discriminator, making the signal generated by the generator more and more realistic. Other loss functions can be selected according to specific tasks and needs. For example, the use of the mean square error loss function can help the generator better fit the real noise-free signal. During the training process, by weighted combination of the adversarial loss function and other loss functions, the weights of the generator and the discriminator can be adjusted so that the two are balanced during the optimization process, thereby improving the noise suppression effect of the generator.

[0130] S303: Compare the similarity between the denoised signal output by the generator and the noise-free signal, and determine the noise suppression effect of the generator according to the peak signal-to-noise ratio.

[0131] Specifically, the peak signal-to-noise ratio is calculated based on the mean square error loss function to measure the signal quality, and the similarity between the generated signal and the noise-free signal is compared.

[0132] The calculation formula for peak signal-to-noise ratio PSNR is:

[0133]

[0134] Where P is the maximum value of the signal, which is usually 2 for a noise-free audio signal. 16 -1, MSE is mean square error.

[0135] The larger the PSNR, the more similar the generated signal is to a noise-free signal.

[0136] In S3, the generative adversarial network can learn complex nonlinear relationships, which helps to better handle the complex noise characteristics in the partial discharge signal of the switchgear; the generator can generate realistic partial discharge signals, which helps to imitate real signals and provide more accurate noise suppression effects; unlike supervised learning methods, the use of generative adversarial networks for noise suppression does not require a large amount of labeled noise and signal data, reducing the cost and complexity of data labeling; through adversarial training, the generative adversarial network helps to improve the generalization ability of the model, making the model more adaptable to different types of noise and signals; the generative adversarial network can adaptively adjust the parameters of the generator and discriminator according to the distribution of the input data, so that the model can better adapt to different noise and signal characteristics.

[0137] The LSTM network is an improved version of the recurrent neural network. By introducing memory units and gating mechanisms, it can better process long sequence data and avoid the gradient vanishing problem. The LSTM network focuses more on capturing the grammatical and semantic information of the text, while the GAN network focuses more on generating realistic text data. The LSTM network can be used as a part of the generator. The GAN network can improve its generation ability. Through adversarial training of the GAN network, the LSTM network can be guided to generate more realistic and diverse texts.

[0138] Based on the same inventive concept as the first embodiment described above, the second embodiment of the present disclosure provides a signal noise suppression system, comprising:

[0139] An acquisition module is used to acquire the original data of the partial discharge signal of the switch cabinet and perform feature extraction on the partial discharge signal;

[0140] A judgment module is used to perform noise suppression on the partial discharge signal after feature extraction, determine the signal-to-noise ratio and the mean square error, and judge the quality of the noise suppression effect according to the signal-to-noise ratio and the mean square error;

[0141] The optimization module is used to optimize the noise suppression effect according to the comparison result of the noise suppressed data and the noise-free signal.

[0142] Furthermore, the acquisition module includes:

[0143] A decomposition unit, used for decomposing the partial discharge signal to obtain a plurality of intrinsic mode functions;

[0144] The determination unit is used to convert the time domain signal in each intrinsic mode function into a frequency domain signal, calculate the spectral centroid of each intrinsic mode function, and determine the eigenvector of each intrinsic mode function according to the position of the spectral centroid.

[0145] Further, the determining unit comprises:

[0146] A conversion unit, used for converting a time domain signal in each intrinsic mode function into a frequency domain signal, and converting a multi-frequency signal into a single source signal;

[0147] The spectral centroid determination unit is used to determine the spectral centroid according to the similarity of the single source signal in the time domain and in the frequency domain.

[0148] Furthermore, the judging module includes:

[0149] A first calculation unit, used for determining a signal-to-noise ratio and a mean square error according to a noise suppression result;

[0150] The first judging unit is used to judge the quality of the noise suppression effect according to the signal-to-noise ratio and the mean square error loss function.

[0151] Furthermore, the optimization module includes:

[0152] The second judgment unit is used to compare the denoised signal with the noise-free signal, optimize the parameters of the generator and the discriminator through adversarial training, and adjust the weights of the generator and the discriminator according to the adversarial loss function;

[0153] The second calculation unit is used to compare the similarity between the denoised signal output by the generator and the noise-free signal, and to determine the noise suppression effect according to the peak signal-to-noise ratio.

[0154] In this embodiment, the feature extraction of the partial discharge signal is first performed, and then the noise suppression is performed, thereby improving the accuracy of the signal feature extraction; the noise suppression effect is continuously optimized based on the comparison result of the noise-suppressed data and the noise-free signal, thereby improving the noise suppression accuracy of the complex noise features in the partial discharge signal of the switch cabinet.

[0155] A third embodiment of the present disclosure provides a storage medium storing a computer program. When the computer program is executed by a processor, the steps of implementing the above method include:

[0156] S11: acquiring raw data of partial discharge signals of the switch cabinet and performing feature extraction on the partial discharge signals;

[0157] S12: performing noise suppression on the partial discharge signal after feature extraction, determining a signal-to-noise ratio and a mean square error, and judging the quality of the noise suppression effect according to the signal-to-noise ratio and the mean square error;

[0158] S13: Optimizing the noise suppression effect according to a comparison result between the noise suppressed data and the noise-free signal.

[0159] Of course, other steps of the control method can also be used to implement the above embodiment.

[0160] In this embodiment, the feature extraction of the partial discharge signal is first performed, and then the noise suppression is performed, thereby improving the accuracy of the signal feature extraction; the noise suppression effect is continuously optimized based on the comparison result of the noise-suppressed data and the noise-free signal, thereby improving the noise suppression accuracy of the complex noise features in the partial discharge signal of the switch cabinet.

[0161] A fourth embodiment of the present disclosure provides an electronic device, the electronic device including at least a memory and a processor, the memory storing a computer program, and the processor implementing the steps of the above method when executing the computer program in the memory, specifically including:

[0162] S21: acquiring raw data of a partial discharge signal of a switch cabinet, and performing feature extraction on the partial discharge signal;

[0163] S22: performing noise suppression on the partial discharge signal after feature extraction, determining a signal-to-noise ratio and a mean square error, and judging the quality of the noise suppression effect according to the signal-to-noise ratio and the mean square error;

[0164] S23: Optimizing the noise suppression effect according to a comparison result between the noise suppressed data and the noise-free signal.

[0165] Of course, other steps of the control method can also be used to implement the above embodiment.

[0166] In this embodiment, the feature extraction of the partial discharge signal is first performed, and then the noise suppression is performed, thereby improving the accuracy of the signal feature extraction; the noise suppression effect is continuously optimized based on the comparison result of the noise-suppressed data and the noise-free signal, thereby improving the noise suppression accuracy of the complex noise features in the partial discharge signal of the switch cabinet.

[0167] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0168] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0169] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0170] In the embodiments provided in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0171] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0173] If the integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned motor torque control method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0174] In addition, the features of the embodiments shown in the drawings of the present application or the various embodiments mentioned in this specification are not necessarily understood as independent embodiments. Instead, each feature described in one example of an embodiment can be combined with one or more other desired features from other embodiments to produce other embodiments not described in words or with reference to the drawings.

[0175] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for suppressing noise of a signal, characterized in that: include: Acquire raw data of a partial discharge signal of a switch cabinet, and perform feature extraction on the partial discharge signal; Perform noise suppression on the partial discharge signal after feature extraction, determine the signal-to-noise ratio and mean square error, and judge the quality of the noise suppression effect based on the signal-to-noise ratio and mean square error; The noise suppression effect is optimized according to the comparison result between the noise suppressed data and the noise-free signal.

2. The signal noise suppression method according to claim 1, characterized in that: Extracting features from the partial discharge signal includes: Decomposing the partial discharge signal to obtain a plurality of intrinsic mode functions; The time domain signal in each intrinsic mode function is converted into a frequency domain signal, the spectral centroid of each intrinsic mode function is calculated, and the eigenvector of each intrinsic mode function is determined according to the position of the spectral centroid.

3. The signal noise suppression method according to claim 2, characterized in that: The partial discharge signal is decomposed to obtain a plurality of intrinsic mode functions including: Optimizing the number of decomposed modes of variational mode decomposition according to weighted inclination K With penalty factor α Parameter combination; The original data is decomposed based on the optimized parameter combination to obtain a plurality of intrinsic mode functions.

4. The signal noise suppression method according to claim 2, characterized in that: The time domain signal in each intrinsic mode function is converted into a frequency domain signal, the spectral centroid of each intrinsic mode function is calculated, and the eigenvector of each intrinsic mode function is determined according to the position of the spectral centroid, including: The time domain signal in each intrinsic mode function is converted into a frequency domain signal, and the multi-frequency signal is converted into a single source signal; The spectral centroid is determined according to the similarity of the single source signal in the time domain and in the frequency domain.

5. The signal noise suppression method according to claim 1, characterized in that: The noise suppression is performed on the partial discharge signal after feature extraction, the signal-to-noise ratio and mean square error are determined, and the quality of the noise suppression effect is judged based on the signal-to-noise ratio and mean square error. The following are included: According to the noise suppression result, the signal-to-noise ratio and the mean square error are determined; The quality of the noise suppression effect is judged based on the signal-to-noise ratio and the mean square error loss function.

6. The signal noise suppression method according to claim 1, characterized in that: According to the comparison result of the noise suppressed data and the noise-free signal, optimizing the noise suppression effect includes: The raw data of the switchgear partial discharge signal is input into the generator, and the denoised signal is output; Comparing the denoised signal with the noise-free signal, optimizing the parameters of the generator and the discriminator through adversarial training, and adjusting the weights of the generator and the discriminator according to the adversarial loss function; Compare the similarity between the denoised signal output by the generator and the noise-free signal, and judge the noise suppression effect based on the peak signal-to-noise ratio.

7. The signal noise suppression method according to claim 1, characterized in that: The raw data includes at least one of a voltage signal, a current signal, a temperature value, a humidity level, and a sound signal.

8. A signal noise suppression system, characterized in that: include: An acquisition module, used to acquire raw data of a partial discharge signal of a switch cabinet and perform feature extraction on the partial discharge signal; A judgment module is used to perform noise suppression on the partial discharge signal after feature extraction, determine the signal-to-noise ratio and the mean square error, and judge the quality of the noise suppression effect according to the signal-to-noise ratio and the mean square error; The optimization module is used to optimize the noise suppression effect according to the comparison result of the noise-suppressed data and the noise-free signal.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the noise suppression method described in any one of claims 1 to 7 is performed.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the noise suppression method described in any one of claims 1 to 7 is implemented.

Images