Ultrasonic sensor field calibration signal denoising method based on improved VMD-wavelet

Abstract

The invention discloses an ultrasonic sensor field calibration signal denoising method based on improved VMD-wavelet. When a partial discharge ultrasonic sensor is verified on site, a verification signal output by the ultrasonic sensor comprises narrow-band noise and white noise. In order to suppress the two kinds of noise, firstly, an original noisy signal output by the ultrasonic sensor to be detected is collected and decomposed by using VMD to obtain IMF components, secondly, the IMF components containing narrow-band signals are screened out based on a kurtosis joint correlation coefficient criterion and removed, effective IMF components are reserved, the effective IMF components are decomposed by adopting discrete wavelet transform, and then the two kinds of noise are suppressed. And processing the high-frequency detail coefficient by using an improved threshold criterion and a self-adaptive threshold function, and finally performing wavelet reconstruction on each layer of coefficient after processing through wavelet inverse transformation to obtain a de-noised signal. According to the method, adaptive denoising can be carried out on narrow-band noise and white noise, and the algorithm efficiency is high.

Description

Technical Field

[0001] The present invention relates to the technical field of sensor calibration, and particularly to a denoising method for on-site calibration signals of ultrasonic sensors based on improved VMD-wavelet. Background Technique

[0002] Partial discharge monitoring is an effective means to predict the faults of large power equipment. The accuracy of on-site calibration of ultrasonic sensors used for monitoring directly determines the judgment of the sensor performance by the staff, and further affects the credibility of partial discharge quantitative evaluation. Due to the existence of composite noise pollution such as breaker operation vibration and carrier communication band crosstalk in the substation site, traditional denoising methods are difficult to effectively separate the microsecond-level ultrasonic pulses of the calibration signal from strong background interference, resulting in systematic deviation in the calibration of the sensor sensitivity.

[0003] In the existing signal denoising technologies, the non-stationary signal processing method based on empirical mode decomposition (EMD) is vulnerable to mode mixing, resulting in the fracture of the characteristic waveform of the ultrasonic signal; while the classical variational mode decomposition (VMD) realizes frequency band segmentation by presetting the number of modes, but it is difficult to effectively extract low-frequency noise components; the wavelet threshold denoising method can suppress some broadband noise through time-frequency localization analysis, but its performance is limited by aspects such as basis functions, fixed threshold rules, and traditional threshold functions, resulting in amplitude attenuation or oscillation residue during signal reconstruction. Moreover, the types of on-site noise are complex, and the noise frequency may be mixed with the calibration signal frequency. When the above methods are used alone, the effect on composite noise is poor, and the amplitude of the calibration signal will attenuate and the waveform will be distorted after denoising. Summary of the Invention

[0004] To solve the technical problems existing in the above background technique, the present invention provides a denoising method for on-site calibration signals of ultrasonic sensors based on improved VMD-wavelet.

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

[0006] A denoising method for on-site calibration signals of ultrasonic sensors based on improved VMD-wavelet, the specific method includes the following steps:

[0007] S1. Collect the original noisy calibration signal y(t) output by the ultrasonic sensor to be measured installed on the transformer or gas insulated switchgear (GIS), and use VMD to decompose y(t) to obtain n IMF components.

[0008] S2. Calculate the kurtosis and correlation coefficient of each IMF component, and based on the kurtosis combined with the correlation coefficient criterion J m Screen out the IMF components containing narrowband signals and eliminate them, and retain the effective IMF components, that is, the IMF components with J m greater than 3 are effective IMF components. The calculation formula of the kurtosis combined with the correlation coefficient criterion is:

[0009] J m = αKu m + (1 - α)R m

[0010] In the formula, Ku m and R m are the kurtosis and correlation coefficient of the m-th IMF component respectively, and the calculation formulas are respectively:

[0011]

[0012] In the formula, x m (t) is the m-th IMF component obtained after the original noisy calibration signal is decomposed by VMD, μ m and σ m are the mean and standard deviation of the m-th IMF component respectively, E and D represent the expectation and variance operations respectively, and the weight parameter α can be calculated from the information entropy of the kurtosis and correlation coefficient, and its calculation formula is:

[0013]

[0014] In the formula, H(R) and H(Ku) are the information entropies of the correlation coefficient and kurtosis respectively, and their calculation formulas are respectively:

[0015]

[0016] In the formula, r i and k i are the correlation coefficient and kurtosis value of the i-th IMF component respectively.

[0017] S3. Select the wavelet basis function and the decomposition level, and perform wavelet decomposition on the effective IMF components retained in S2 by using discrete wavelet transform to obtain the high-frequency detail coefficients and low-frequency approximation coefficients at each level. For the high-frequency detail coefficients where the noise is mainly distributed, use the improved threshold criterion and the adaptive threshold function to process them to separate the Gaussian white noise mixed in the IMF components. The expression of the improved threshold is:

[0018]

[0019] σ, σ s represent the standard deviation and the modified variance of the noise signal respectively,

[0020]

[0021] σ s = max(0, Var(ω 1,j ) - σ 2 )

[0022] In the formula, ω1,k Denote the \(k\)-th wavelet coefficient in the first-layer wavelet decomposition coefficients, and \(Var(\cdot)\) represents the variance operation. The expression of the adaptive threshold function is:

[0023]

[0024] where \(\omega\) j,k is the \(k\)-th wavelet coefficient at the \(j\)-th scale of wavelet decomposition, \(\alpha\) is the steepness parameter, \(\beta\) is the shrinkage parameter, and the values of \(\alpha\) and \(\beta\) are related to the kurtosis value of the wavelet coefficients at the \(j\)-th scale. Its expression is:

[0025]

[0026] where \(ku\) is the kurtosis value of the set of wavelet coefficients at the \(j\)-th scale.

[0027] S4. Perform wavelet reconstruction on the processed coefficients of each layer through inverse wavelet transform to obtain the denoised signal.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] 1. Propose the kurtosis combined with correlation coefficient criterion, fuse the double criteria, and avoid the problem of signal mis-removal caused by a single criterion. At the same time, dynamically adjust the weight parameter through information entropy to enhance the adaptability of the algorithm to different noise scenarios and different signals.

[0030] 2. Propose an improved threshold criterion and an adaptive threshold function. The improved threshold criterion can dynamically calculate the threshold magnitude to ensure that the threshold is adaptively adjusted with the noise intensity. In the adaptive threshold function, introduce the steepness parameter and the shrinkage parameter, which are determined by the kurtosis value of the wavelet coefficients at the current scale. When the distribution of wavelet coefficients is sharp, increase the shrinkage parameter to strengthen the noise shrinkage; when the distribution is flat, reduce the steepness parameter to reduce signal distortion. This mechanism can remove the residual white noise to the greatest extent without reducing the amplitude of the verification signal and without causing distortion of the original signal. Description of the Drawings

[0031] In the drawings:

[0032] Figure 1 is a flowchart of a denoising method for on-site calibration signals of ultrasonic sensors based on improved VMD-wavelet;

[0033] Figure 2 is a diagram of the original noisy calibration signal;

[0034] Figure 3 is a diagram of the \(J\) m value of each IMF component;

[0035] Figure 4 is a diagram of the denoised signal. Detailed implementation manners

[0036] The present invention will be further described in detail below in combination with the specific implementation manners 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 a limitation of the present invention.

[0037] As Figure 1 shown is a flowchart of a denoising method for on-site calibration signals of ultrasonic sensors based on improved VMD-wavelet. The specific method includes the following steps:

[0038] S1. Collect the original noisy calibration signal y(t) output by the ultrasonic sensor to be measured installed on the transformer or gas-insulated switchgear (GIS), as Figure 2 shown. Decompose y(t) using VMD to obtain 12 IMF components.

[0039] S2. Calculate the kurtosis and correlation coefficient of each IMF component, and screen out and remove the IMF components containing narrowband signals based on the kurtosis combined with the correlation coefficient criterion J m as m shown. When J Figure 3 is less than 3, it is considered that the IMF component is a component containing a narrowband signal, that is, IMF2, IMF3, IMF5, and IMF10 are removed. At the same time, retain the effective IMF components, that is, the IMF components with J m greater than 3 are effective IMF components. The calculation formula of the kurtosis combined with the correlation coefficient criterion is: m J

[0040] J m =αKu m +(1 - α)R m

[0041] In the formula, Ku m and R m are the kurtosis and correlation coefficient of the m-th IMF component respectively, and the calculation formulas are respectively:

[0042]

[0043] In the formula, x m (t) is the m-th IMF component obtained by decomposing the original noisy calibration signal using VMD, μ m and σ m are the mean and standard deviation of the m-th IMF component respectively, E and D represent the expectation and variance operations respectively, and the weight parameter α can be calculated from the information entropy of the kurtosis and the correlation coefficient, and its calculation formula is:

[0044]

[0045] Wherein, H(R) and H(Ku) are the information entropy of the correlation coefficient and kurtosis respectively, and their calculation formulas are as follows:

[0046]

[0047] Wherein, r i and k i are the correlation coefficient and kurtosis value of the i-th IMF component respectively.

[0048] S3. Select the db45 wavelet as the wavelet basis function, select 6-layer wavelet decomposition, and perform wavelet decomposition on the effective IMF components retained in S2 by using discrete wavelet transform to obtain the high-frequency detail coefficients and low-frequency approximation coefficients at each level. For the high-frequency detail coefficients where the noise is mainly distributed, use the improved threshold criterion and adaptive threshold function to process them and separate the Gaussian white noise mixed in the IMF components. The expression of the improved threshold is:

[0049]

[0050] σ, σ s represent the standard deviation and corrected variance of the noise signal respectively,

[0051]

[0052] σ s = max(0, Var(ω 1,j ) - σ 2 )

[0053] Wherein, ω 1,k represents the k-th wavelet coefficient in the first-layer wavelet decomposition coefficients, Var(·) represents the variance operation, and the expression of the adaptive threshold function is:

[0054]

[0055] Wherein, ω j,k is the k-th wavelet coefficient at the j-th scale of the wavelet decomposition, α is the steepness parameter, β is the shrinkage parameter, and the values of α and β are related to the kurtosis value of the wavelet coefficients at the j-th scale. Their expressions are:

[0056]

[0057] Wherein, ku is the kurtosis value of the wavelet coefficient set at the j-th scale.

[0058] S4. Perform wavelet reconstruction on the processed coefficients of each layer through inverse wavelet transform to obtain the denoised signal, as Figure 4 shown.

Claims

1. A method for denoising ultrasonic sensor field calibration signals based on improved VMD-wavelet, characterized in that: The specific method includes the following steps: S1, collect the original noise-stained calibration signal y(t) output by the ultrasonic sensor to be tested installed on the transformer or gas insulated switchgear (GIS), and use VMD to decompose y(t) to obtain n IMF components; S2. Calculate the kurtosis and correlation coefficient of each IMF component, based on the kurtosis combined correlation coefficient criterion J m Filter out the IMF components containing narrowband signals and remove them, retaining the effective IMF components, i.e., J m IMF components greater than 3 are effective IMF components; S3, select the wavelet basis function and the number of decomposition levels, use discrete wavelet transform to perform wavelet decomposition on the effective IMF components retained in S2, obtain high-frequency detail coefficients and low-frequency approximate coefficients at each level through decomposition, use improved threshold criterion and adaptive threshold function to process the high-frequency detail coefficients where the noise is mainly distributed, and separate the mixed Gaussian white noise in the IMF components; S4. Perform wavelet reconstruction on the processed coefficients of each layer through inverse wavelet transform to obtain a denoised signal.

2. According to claim 1, a method for denoising ultrasonic sensor field calibration signal based on improved VMD-wavelet, characterized in that: S2 kurtosis joint correlation coefficient criterion J m The calculation formula is: J m =αKu m +(1-α)R m In the formula, Ku m and R m are the kurtosis and correlation coefficient of the mth IMF component, respectively, and the calculation formulas are: In the formula, x m (t) is the mth IMF component obtained after VMD decomposition of the original noise-stained check signal, μ m and σ m are the mean and standard deviation of the mth IMF component, E and D represent the expectation and variance operations, respectively. The weight parameter α can be calculated from the information entropy of the kurtosis and correlation coefficient, and its calculation formula is: In the formula, H(R) and H(Ku) are the information entropy of correlation coefficient and kurtosis, respectively, and their calculation formulas are: In the formula, r i and k i are the correlation coefficient and kurtosis value of the i-th IMF component respectively.

3. The method for denoising ultrasonic sensor field calibration signals based on improved VMD-wavelet according to claim 1 is characterized in that: In S3, the improved threshold criterion and adaptive threshold function are used to process the high-frequency detail coefficients. The expression of the improved threshold is: σ, σ s represent the standard deviation and corrected variance of the noise signal respectively, s s =max(0,Var(ω 1,j )-s 2 ) In the formula, ω 1,j represents the jth wavelet coefficient in the first layer of wavelet decomposition coefficients, Var(·) represents the variance operation, and the expression of the adaptive threshold function is: In the formula, ω j,k is the kth wavelet coefficient at the jth scale of wavelet decomposition, α is the steepness parameter, β is the contraction parameter, and the values ​​of α and β are related to the kurtosis value of the wavelet coefficient at the jth scale. The expression is: Where ku is the kurtosis value of the wavelet coefficient set at the jth scale.

Images