Transformer partial discharge ultrasonic sensor field calibration signal denoising method

Abstract

The invention relates to the field of partial discharge ultrasonic sensor verification, in particular to a transformer partial discharge ultrasonic sensor field verification signal denoising method, which comprises the following steps of: acquiring an original noisy verification signal from a to-be-tested ultrasonic sensor; the method comprises the following steps: decomposing an original signal by using MVMD to obtain IMF components, and screening useful IMF components based on a kurtosis criterion so as to remove narrowband noise; performing wavelet decomposition on the useful IMF component to obtain a wavelet detail coefficient, and processing the wavelet detail coefficient based on an improved threshold function and a general threshold to remove white noise; and performing wavelet reconstruction on the processed coefficients of each layer through wavelet inverse transformation to obtain a de-noised signal. According to the method, the white noise and the narrow-band noise contained in the verification signal output by the transformer partial discharge ultrasonic sensor are effectively reduced, and the accuracy of the verification result of the ultrasonic sensor to be tested is improved.

Description

Technical Field

[0001] The present invention relates to the field of partial discharge sensor calibration, and specifically to a method for denoising the on-site calibration signal of a transformer partial discharge ultrasonic sensor. Background Art

[0002] Utilizing partial discharge triggering protection can protect against tripping before the main insulation of a large power transformer breaks down, avoiding explosion and combustion accidents. Therefore, this poses higher requirements for the detection accuracy of partial discharge. Ultrasonic sensors have become one of the core sensing devices for partial discharge detection due to advantages such as non-intrusive measurement and strong anti-electromagnetic interference ability. However, during its long-term service, its performance is easily affected by the on-site environment, and on-site calibration work needs to be carried out regularly. However, when calibrating under on-site working conditions, the signal is often interfered by various noises, directly affecting the accuracy of calibration. Therefore, it is necessary to denoise the calibration signal during the calibration process.

[0003] Currently, the main denoising methods for ultrasonic sensing signals include hardware filtering methods, adaptive filtering methods, and wavelet analysis methods, etc. The noise suppression technology based on hardware filtering eliminates specific frequency band noises through a band-pass filter, but there are problems such as rigid cut-off frequency setting and loss of high-frequency components of useful signals when facing wide-band non-stationary noises. For the algorithms based on adaptive filtering, although such methods can dynamically adjust the filtering parameters, the suppression effect on transient pulse noises and mechanical vibration noises overlapping with the effective signal frequency band is not good. The time-frequency domain processing method based on wavelet transform has the characteristics of multi-resolution analysis, but it faces limitations such as the selection of wavelet basis functions depending on experience and the threshold rule being difficult to adapt to on-site dynamic noises. Therefore, it is necessary to study a denoising method suitable for on-site complex noise environments, with both adaptive processing ability and high-fidelity characteristics. Summary of the Invention

[0004] To solve the technical problems existing in the above background art, the present invention provides a method for denoising the on-site calibration signal of a transformer partial discharge ultrasonic sensor.

[0005] The technical solution of the present invention is as follows:

[0006] A method for denoising the on-site calibration signal of a transformer partial discharge ultrasonic sensor, the specific method comprising the following steps:

[0007] S1. Install an acoustic emission sensor on the transformer 200 mm away from the sensor to be tested, install a standard ultrasonic sensor 200 mm away from the acoustic emission sensor, connect the standard ultrasonic sensor and the ultrasonic sensor to be tested to the host computer, and connect the acoustic emission sensor to the signal generator. The signal generator excites the acoustic emission sensor to send a calibration signal, and obtains the original noise calibration signal y1(t) received by the sensor to be tested through the host computer. Expand the single-channel noise calibration signal y1(t) to a three-channel noise calibration signal matrix X(t)∈R 3*N , where N is the length of the original noise-stained calibration signal, and channel 1 uses the original noise-stained calibration signal, that is:

[0008] X1(t)=y1(t)

[0009] Channel 2 uses the instantaneous phase information extracted from y1(t) after Hilbert transformation, namely:

[0010]

[0011] Channel 3 is constructed by phase shifting y1(t), namely:

[0012]

[0013] The three-channel noise-stained calibration signal X(t) is decomposed using MVMD to obtain the IMF components.

[0014] S2. Based on the improved kurtosis criterion, the IMF components dominated by the noise signal are screened out and eliminated, and the useful IMF components are retained, that is, the components with kurtosis values less than 2 are eliminated, and the components with kurtosis values greater than 2 are retained. The calculation formula of the improved kurtosis K is:

[0015]

[0016] In the formula, x i is the i-th IMF component obtained by decomposing the original noise-stained check signal, μ i and σ i are the mean and standard deviation of the ith IMF component, E(x i -μ i ) 4 is the fourth-order mathematical expectation of the signal, E i is the energy of the ith IMF component, calculated as:

[0017]

[0018] Where T is the length of the IMF component signal, E total is the total energy of all IMF components, calculated as:

[0019]

[0020] In the formula, n is the number of IMF components. When the modal component is dominated by the noise signal, the kurtosis value will be less than 2. When the modal component is dominated by the calibration signal, the kurtosis value will increase, so as to eliminate the IMF components dominated by the noise signal.

[0021] S3. Perform wavelet decomposition on the useful IMF components retained in S2. After obtaining the wavelet detail coefficients, use the improved wavelet threshold function for processing, aiming to separate and eliminate the residual white noise interference in the IMF components. The calculation formula of the improved wavelet threshold function is:

[0022]

[0023] In the formula, λ is the wavelet threshold, and the general threshold is adopted. Its expression is:

[0024]

[0025] In the formula, N is the length of the original noisy calibration signal y1(t), ω 1,k represents the k-th wavelet coefficient in the first-layer wavelet decomposition coefficients

[0026] S4. Perform wavelet reconstruction on the processed coefficients of each layer through inverse wavelet transform to obtain the denoised signal.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. By constructing a three-channel signal matrix and combining MVMD for multivariate variational mode decomposition, using the information redundancy and complementarity between multi-channel signals, the ability to suppress narrowband noise and white noise is significantly enhanced in the transformer operating environment.

[0029] 2. The improved kurtosis criterion can dynamically distinguish the characteristics of noise and calibration signals, overcome the problem of insufficient sensitivity of the traditional kurtosis criterion to non-Gaussian noise, and avoid the problem of signal distortion caused by mis-eliminating useful components in the traditional method.

[0030] 3. According to the characteristics of the calibration signal, an exponentially decaying improved threshold function is proposed, which overcomes the problems of "oversmoothing" or "under-denoising" easily generated by the existing wavelet threshold function in the processing of IMF components, effectively separates the residual white noise and retains the high-frequency details of the signal, and ensures that the waveform of the reconstructed signal does not distort. Description of the Drawings

[0031] In the drawings:

[0032] Figure 1 is the overall step flow chart of a method for denoising the on-site calibration signal of a transformer partial discharge ultrasonic sensor;

[0033] Figure 2 It is the layout diagram of the sensor on the transformer;

[0034] Figure 3 It is the original noisy verification signal diagram;

[0035] Figure 4 It is the IMF decomposition diagram;

[0036] Figure 5 It is the screening diagram of the improved kurtosis criterion;

[0037] Figure 6 It is the denoised signal diagram. Specific implementation manner

[0038] The present invention will be further described in detail below in conjunction with the specific implementation manner and the accompanying drawings. It should be emphasized that the specific embodiments described herein are intended to explain the present invention and should not be construed as limiting the present invention.

[0039] As Figure 1 shown, it is a method for denoising the on-site verification signal of the partial discharge ultrasonic sensor of a transformer. The specific method includes the following steps:

[0040] S1. As Figure 2 shown, install an acoustic emission sensor 200 mm away from the sensor to be measured on the transformer, install a standard ultrasonic sensor 200 mm away from the acoustic emission sensor. The standard ultrasonic sensor and the sensor to be measured are connected to the upper computer, and the acoustic emission sensor is connected to the signal generator. The signal generator stimulates the acoustic emission sensor to emit a verification signal, and the original noisy verification signal y1(t) received by the sensor to be measured is obtained through the upper computer, as Figure 3 shown. Expand the single-channel noisy verification signal y1(t) into a three-channel noisy verification signal matrix X(t) ∈ R 3*N , where N is the length of the original noisy verification signal. Channel 1 uses the original noisy verification signal, that is:

[0041] X1(t) = y1(t)

[0042] Channel 2 uses the instantaneous phase information extracted from y1(t) after Hilbert transform, that is:

[0043]

[0044] Channel 3 is constructed by phase-shifting y1(t), that is:

[0045]

[0046] Use MVMD to decompose the three-channel noisy verification signal X(t) to obtain 10 IMF components, as Figure 4 shown.

[0047] S2. Screen out and remove the IMF components dominated by noise signals based on the improved kurtosis criterion, retain the useful IMF components, remove the components with kurtosis values less than 2, and retain the components with kurtosis values greater than 2. As shown in Figure 5 , remove IMF2, IMF4, IMF5, and IMF7. The calculation formula for the improved kurtosis K is:

[0048]

[0049] In the formula, x i is the i-th IMF component obtained by decomposing the original noisy calibration signal, μ i and σ i are the mean and standard deviation of the i-th IMF component respectively, E(x i -μ i ) 4 is the fourth-order mathematical expectation of the signal, E i is the energy of the i-th IMF component, and the calculation formula is:

[0050]

[0051] In the formula, T is the length of the IMF component signal, E total is the total energy of all IMF components, and the calculation formula is:

[0052]

[0053] In the formula, n is the number of IMF components. When the modal component is dominated by noise signals, the kurtosis value will be less than 2. When the modal component is dominated by calibration signals, the kurtosis value will increase, so as to remove the IMF components containing narrowband interference.

[0054] S3. Perform wavelet decomposition on the useful IMF components retained in S2, select the sym8 wavelet basis function for 4-layer wavelet decomposition, and use the improved wavelet threshold function to process the obtained wavelet detail coefficients, aiming to separate and remove the residual white noise interference in the IMF components. The calculation formula for the improved wavelet threshold function is:

[0055]

[0056] In the formula, λ is the wavelet threshold, and the general threshold is used, and its expression is:

[0057]

[0058] In the formula, N is the length of the original noisy calibration signal y1(t), ω 1,k represents the k-th wavelet coefficient in the first-layer wavelet decomposition coefficients

[0059] S4. Perform wavelet reconstruction on the processed coefficients of each layer through inverse wavelet transform to obtain the denoised signal, as shown in Figure 6 as follows.

Claims

1. A method for denoising the on-site calibration signal of a partial discharge ultrasonic sensor of a transformer, characterized in that, The specific method includes the following steps: S1. Collect the original noisy calibration signal y1(t) from the ultrasonic sensor to be measured installed on the transformer, expand the single-channel noisy calibration signal y1(t) into a three-channel noisy calibration signal, and use MVMD to decompose the multi-channel noisy calibration signal to obtain IMF components; S2. Screen out and remove the IMF components dominated by noise signals based on the improved kurtosis criterion, and retain the useful IMF components; S3. Based on the improved wavelet threshold function, process the wavelet detail coefficients obtained after wavelet decomposition of the useful IMF components retained in S2, aiming to separate and remove the residual white noise interference in the IMF components; S4. Perform wavelet reconstruction on the processed coefficients of each layer through inverse wavelet transform to obtain the denoised signal; In S1, the single-channel noisy verification signal y1(t) is extended to a three-channel noisy verification signal matrix X(t) ∈ R 3*N , where N is the length of the original noisy verification signal. Channel 1 uses the original noisy verification signal, that is: X1(t) = y1(t) Channel 2 uses the instantaneous phase information extracted after the Hilbert transform of y1(t), that is: Channel 3 is constructed by performing phase shift on y1(t), that is:

2. A method for denoising the on-site calibration signal of a transformer partial discharge ultrasonic sensor according to claim 1, characterized in that, In S2, the improved kurtosis criterion is used to remove the IMF components containing narrowband signals. The calculation formula of the improved kurtosis K is: where x i is the i-th IMF component obtained by decomposing the original contaminated and noise-added verification signal, μ i and σ i are the mean and standard deviation of the i-th IMF component respectively, E(x i - μ i ) 4 is the fourth-order mathematical expectation of the signal, and E i is the energy of the i-th IMF component, and the calculation formula is: where T is the length of the IMF component signal, and E total is the total energy of all IMF components, and the calculation formula is: In the formula, n is the number of IMF components; when the modal component is dominated by noise, the kurtosis value will be less than 2, and when the modal component is dominated by the calibration signal, the kurtosis value will increase, so as to remove the IMF components dominated by noise.

3. A method for denoising the on-site calibration signal of a partial discharge ultrasonic sensor of a transformer according to claim 1, characterized in that, In S3, the wavelet threshold and the improved threshold function are used to process the wavelet detail coefficients. The calculation formula of the improved wavelet threshold function is: In the formula, λ is the wavelet threshold, and the general threshold is used, and its expression is: where N is the length of the original noisy parity-check signal y1(t), ω 1,k represents the k-th wavelet coefficient in the wavelet decomposition coefficients of the first layer.

Images