Method and system for denoising acoustic signals of AC filter field reactor

Abstract

The present invention provides a method and system for joint denoising of the acoustic signal of an AC filter reactor, and belongs to the technical field of signal denoising of electric power equipment. The method firstly adaptively decomposes the original acoustic signal containing complex noise through VMD optimized by IVYA algorithm parameters; then divides the decomposed IMF component into a useful signal dominant component and a noise dominant component according to the correlation coefficient; finally, the noise dominant component is further denoised by using an improved parameter-free wavelet threshold denoising algorithm and reconstructed by superposition with the useful signal dominant component to obtain the final denoised signal. The method has excellent denoising performance and high adaptability, and can effectively extract pure acoustic signals of AC filter reactors, which helps to reduce the difficulty of subsequent feature extraction and pattern recognition tasks, and further provides a basis for the state evaluation of AC filters.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment signal denoising, and in particular to a method and system for denoising an acoustic signal of an AC filter field reactor of an ultra-high voltage converter station. Background Art

[0002] AC filter is a key component in UHV converter station. It has important functions such as regulating voltage, providing reactive power compensation and filtering harmonics. It has the characteristics of high heat generation and large size. If a fault occurs, it will directly affect the safe and stable operation of DC transmission system. The AC filter field contains a large number of filter reactors, which are the basic components to support its function realization and are also one of the main noise sources in the converter station. Under long-term mechanical vibration operation, the reactor is prone to structural fatigue, loose fasteners and wear and aging, which reflects the characteristic changes of its external acoustic vibration monitoring signal. Compared with the high cost and mostly contact measurement of vibration signal acquisition, acoustic signal monitoring with non-contact measurement, strong anti-interference and high operating state sensitivity has good application potential. Therefore, it is of great significance to carry out AC filter status assessment for reactor acoustic signals. However, due to the influence of the complex working conditions of the actual converter station, the reactor body acoustic signal collected by the acoustic sensor is inevitably interfered by various noise components, such as voice, environmental noise, corona discharge sound, etc. These noise interferences may drown out the early weak degradation features, which is not conducive to the subsequent further feature extraction and pattern recognition. Therefore, it is necessary to denoise the collected acoustic signals to improve the credibility of the signals.

[0003] At present, there are few studies on the denoising of acoustic signals of AC filter field reactors in UHV converter stations at home and abroad. Since the monitoring signals of power equipment are usually nonlinear, non-deterministic and non-stationary, the means of denoising them are mainly through signal processing technologies such as empirical mode decomposition (EMD) and wavelet transform with time-frequency joint analysis capabilities. The existing technology has the following limitations in practical application: ① EMD can adaptively decompose the noisy signal into multiple modal components and remove the noise component to achieve noise reduction, but there are problems such as lack of mathematical theoretical basis, modal aliasing and endpoint effect. Although the variant of EMD, namely variational mode decomposition (VMD), effectively reduces this shortcoming, it still faces the challenge of empirical estimation of two key parameters, the quadratic penalty factor and the number of decomposed modes; ② When using the threshold function to reduce the noise of the signal, the traditional wavelet threshold denoising technology has the problems of discontinuity of the hard threshold function, constant deviation of the soft threshold function and the need for additional parameters to perform the threshold function; ③ It is difficult to completely remove the noise interference information hidden in the useful signal using a single signal processing denoising technology, and the denoising effect is poor. Summary of the invention

[0004] The technical problem to be solved by the present invention is how to accurately and adaptively extract pure and effective signals from the acoustic signals of the AC filter reactor of an ultra-high voltage converter station containing noisy noise, retain the most important features of the signals, reduce the difficulty of subsequent feature extraction and pattern recognition tasks, and thus provide a basis for the status evaluation of the AC filter.

[0005] The present invention solves the above technical problems through the following technical means:

[0006] A method for denoising acoustic signals of an AC filter field reactor of a UHV converter station comprises the following steps:

[0007] Acquire the reactor acoustic signal in the AC filter field of the converter station;

[0008] The optimized VMD algorithm is used to adaptively and non-recursively decompose the original AC filter reactor noise signal into a series of IMF components.

[0009] Classify multiple IMF components into useful signal-dominated components and noise-dominated components;

[0010] The noise-dominated signal is denoised by improving the wavelet threshold using a threshold function, and the threshold function expression is as follows:

[0011]

[0012] Where: , Respectively represent l Tier d High-frequency wavelet coefficients before and after threshold processing at each position; is the hyperbolic tangent function; is a sign function; Represents the threshold value, which is used to determine the boundary between the useful signal coefficient and the noise coefficient, that is:

[0013]

[0014] Where: N is the length of the input IMF component sequence; is the noise standard deviation; express median of absolute values; j is an imaginary unit;

[0015] The noise-dominant component and the effective signal-dominant component after threshold processing are reconstructed by superposition to obtain the final denoised acoustic signal.

[0016] Furthermore, the optimization process of the VMD algorithm is as follows: the VMD parameters are optimized using the IVYA algorithm. and K The specific steps for adaptive optimization are as follows:

[0017] ① Input signal containing noise ;

[0018] ② Initialize the IVYA algorithm, including the maximum number of iterations, population size, population initial position, etc.; initialize the VMD parameter search range and set the quadratic penalty factor , the number of modal decompositions ;

[0019] ③ The fitness function of the IVYA algorithm is constructed based on the maximum information coefficient and is defined as follows:

[0020]

[0021] Where: x and y is a random variable; is the joint probability density function; , They are x and y The marginal probability density function of ; a , b Indicated by x and y The rows and columns that make up the two-dimensional grid; U is 0.6 of the total amount of data; the MIC between decomposition modes should be minimized to prevent modal aliasing, while the MIC between the decomposition mode and the input signal should be maximized to retain valid information; therefore, the fitness function constructed based on MIC is It can be expressed as:

[0022]

[0023] Where: , Respectively k 、( k +1) IMF components; is the original signal; K is the total number of IMF components;

[0024] ④ Continuously updated through IVYA algorithm K and Optimize towards the minimum fitness value until the maximum number of iterations is reached, and the current parameter combination is considered as the optimal solution and output;

[0025] ⑤ Use the optimal parameter combination result of step ④ to input the noisy signal Perform VMD decomposition.

[0026] Furthermore, the specific process of classifying multiple IMF components is as follows: using the correlation coefficient curvature spectrum according to the critical value It is divided into two categories: useful signal dominant component and noise dominant component; the correlation coefficient curvature spectrum is defined as the curvature value calculated for the correlation coefficient sequence; k The correlation coefficient between the IMF component and the original signal The calculation formula is as follows:

[0027]

[0028] Where: E and D denote mathematical expectation and variance respectively;

[0029] like (constant), indicating that the component has a high similarity with the original signal and is classified as the dominant component of the useful signal; if , indicating that the component has a low similarity with the original signal and is classified as a noise-dominated component; c The value is determined by the position corresponding to the maximum curvature value in the correlation coefficient curve; the curvature calculation formula is as follows:

[0030]

[0031] Where: For the z ( ) positions’ correlation coefficient curvature values; K is the total number of IMF components; , , Respectively , z , the first z +1) correlation coefficient; , Respectively z The first and second order derivatives of the correlation coefficients at each position.

[0032] Furthermore, the noise dominant component denoising method based on improved wavelet threshold is as follows:

[0033] ① Wavelet decomposition: The wavelet base is selected as Dobesi wavelet, and the number of decomposition layers is set between 3 and 6 layers. Multi-scale decomposition is performed on each noise dominant component to obtain the high-frequency wavelet coefficients of each layer. ;

[0034] ② Threshold processing: combining improved threshold function and threshold right Quantitative processing is performed to obtain ;

[0035] ③ Wavelet reconstruction: By applying inverse wavelet transform, the IMF components after further noise reduction are obtained.

[0036] The present invention also provides a UHV converter station AC filter field reactor acoustic signal denoising system, comprising the following steps:

[0037] An acoustic signal acquisition module, used to acquire the acoustic signal of the reactor in the AC filter field of the converter station;

[0038] An IMF component decomposition module is used to adaptively and non-recursively decompose the original noisy acoustic signal of the AC filter reactor into a series of IMF components using an optimized VMD algorithm;

[0039] An IMF component classification module is used to classify multiple IMF components into useful signal-dominated components and noise-dominated components;

[0040] The secondary denoising module is used to denoise the noise-dominated signal by improving the wavelet threshold using a threshold function. The threshold function expression is as follows:

[0041]

[0042] Where: , Respectively represent l Tier d High-frequency wavelet coefficients before and after threshold processing at each position; is the hyperbolic tangent function; is a sign function; Represents the threshold value, which is used to determine the boundary between the useful signal coefficient and the noise coefficient, that is:

[0043]

[0044] Where: N is the length of the input IMF component sequence; is the noise standard deviation; express median of absolute values; j is an imaginary unit;

[0045] The signal superposition and reconstruction module is used to reconstruct the noise-dominant component and the effective signal-dominant component after threshold processing by superposition to obtain the final denoised sound signal.

[0046] Furthermore, the optimization process of the VMD algorithm is as follows: the VMD parameters are optimized using the IVYA algorithm. andK The specific steps for adaptive optimization are as follows:

[0047] ① Input signal containing noise ;

[0048] ② Initialize the IVYA algorithm, including the maximum number of iterations, population size, population initial position, etc.; initialize the VMD parameter search range and set the quadratic penalty factor , the number of modal decompositions ;

[0049] ③ The fitness function of the IVYA algorithm is constructed based on the maximum information coefficient and is defined as follows:

[0050]

[0051] Where: x and y is a random variable; is the joint probability density function; , They are x and y The marginal probability density function of ; a , b Indicated by x and y The rows and columns that make up the two-dimensional grid; U is 0.6 of the total amount of data; the MIC between decomposition modes should be minimized to prevent modal aliasing, while the MIC between the decomposition mode and the input signal should be maximized to retain valid information; therefore, the fitness function constructed based on MIC is It can be expressed as:

[0052]

[0053] Where: , Respectively k 、( k +1) IMF components; is the original signal; K is the total number of IMF components;

[0054] ④ Continuously updated through IVYA algorithm K and Optimize towards the minimum fitness value until the maximum number of iterations is reached, and the current parameter combination is considered as the optimal solution and output;

[0055] ⑤ Use the optimal parameter combination result of step ④ to input the noisy signal Perform VMD decomposition.

[0056] Furthermore, the specific process of classifying multiple IMF components is as follows: using the correlation coefficient curvature spectrum according to the critical value It is divided into two categories: useful signal dominant component and noise dominant component; the correlation coefficient curvature spectrum is defined as the curvature value calculated for the correlation coefficient sequence; k The correlation coefficient between the IMF component and the original signal The calculation formula is as follows:

[0057]

[0058] Where: E and D denote mathematical expectation and variance respectively;

[0059] like (constant), indicating that the component has a high similarity with the original signal and is classified as the dominant component of the useful signal; if , indicating that the component has a low similarity with the original signal and is classified as a noise-dominated component; c The value is determined by the position corresponding to the maximum curvature value in the correlation coefficient curve; the curvature calculation formula is as follows:

[0060]

[0061] Where: For the z ( ) positions’ correlation coefficient curvature values; K is the total number of IMF components; , , Respectively , z , the first z +1) correlation coefficient; , Respectively z The first and second order derivatives of the correlation coefficients at each position.

[0062] Furthermore, the noise dominant component denoising method based on improved wavelet threshold is as follows:

[0063] ① Wavelet decomposition: The wavelet base is selected as Dobesi wavelet, and the number of decomposition layers is set between 3 and 6 layers. Multi-scale decomposition is performed on each noise dominant component to obtain the high-frequency wavelet coefficients of each layer. ;

[0064] ② Threshold processing: combining improved threshold function and threshold right Quantitative processing is performed to obtain ;

[0065] ③ Wavelet reconstruction: By applying inverse wavelet transform, the IMF components after further noise reduction are obtained.

[0066] The present invention also provides a processing device, comprising at least one processor and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor can execute the above method by calling the program instructions.

[0067] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the above method.

[0068] The advantages of the present invention are:

[0069] The present invention provides a joint denoising method for the acoustic signal of the AC filter reactor of an ultra-high voltage converter station. The method firstly adaptively decomposes the original acoustic signal containing complex noise through VMD with optimized IVYA algorithm parameters; then divides the decomposed IMF component into a useful signal dominant component and a noise dominant component according to the correlation coefficient; finally, the noise dominant component is further denoised by using an improved parameter-free wavelet threshold denoising algorithm and reconstructed by superposition with the useful signal dominant component to obtain the final denoised signal. The method has excellent denoising performance and high adaptability, and can effectively extract pure acoustic signals of the AC filter reactor, which helps to reduce the difficulty of subsequent feature extraction and pattern recognition tasks, and further provides a basis for the state evaluation of the AC filter. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a general flow chart of the method in the embodiment of the present invention;

[0071] Figure 2 Flow chart of the IVYA algorithm in an embodiment of the present invention;

[0072] Figure 3 This is a VMD parameter optimization process based on IVYA in an embodiment of the present invention;

[0073] Figure 4 The present invention improves the comparison between the threshold function and the soft and hard threshold functions. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] This embodiment discloses a method for denoising an acoustic signal of an AC filter field reactor of a UHV converter station, comprising the following steps:

[0076] Step 1: Signal acquisition

[0077] The acoustic signal of the reactor in the AC filter field of the converter station is obtained by an acoustic sensor. The microphone itself is required to have low noise and its frequency range should be able to cover the acoustic signals generated by the power equipment and background environment around the reactor. Then, the obtained signal is normalized and preprocessed.

[0078] Step 2: Signal adaptive decomposition

[0079] This embodiment is based on the signal decomposition method of optimizing VMD parameters by the Ivy algorithm. The Ivy algorithm, which constructs the fitness function with the maximum information coefficient, adaptively selects the modal decomposition number and the quadratic penalty factor of the VMD algorithm, and substitutes them into the VMD decomposition, thereby realizing signal adaptive decomposition.

[0080] The VMD algorithm in this embodiment is specifically:

[0081] The basic idea of ​​the variational mode decomposition (VMD) algorithm is to transform the mode estimation into a variational problem, that is, to decompose the original complex signal into intrinsic mode function (IMF) components with optimal center frequency and finite bandwidth by constructing and solving the variational problem model, and at the same time require that the sum of the estimated bandwidths of all IMF components is minimized. VMD can effectively solve the problems of mode aliasing and boundary effects existing in traditional empirical mode decomposition algorithms.

[0082] ① Construction of variational problem

[0083] Assuming the original signal Depend on K IMF components, and the constrained variational mathematical model established is:

[0084] (1)

[0085] Where: Relative to timet The partial derivative operator of ; For the k IMF components; For the k The center frequency corresponding to each IMF; Represents the analytical signal of each IMF component; Estimate the exponential operator corresponding to the center frequency of each analytical signal; is the pulse function; j is an imaginary unit; is the convolution operator; Represents the square of the gradient norm; is the original input signal.

[0086] ② Solution of variational problem

[0087] Introducing a quadratic penalty factor and Lagrangian operator , transform the constrained variational problem of formula (1) into an unconstrained variational problem, namely:

[0088] (2)

[0089] Where: is the inner product operator; is the Lagrangian augmented function.

[0090] The optimal solution of formula (2) is obtained using the alternating direction multiplier iteration algorithm. The core solution process is:

[0091] (3)

[0092] (4)

[0093] (5)

[0094] Where: is the noise tolerance factor; n is the number of iterations of the VMD algorithm; is the angular frequency; , , and They are , , and Frequency domain form after Fourier transformation; represents the (n+1)th iteration k IMF estimated components; Indicates q an estimated IMF component; Indicates the (n+1)th iteration k The estimated center frequency of the IMF.

[0095] The IMF components are continuously updated alternately through formula (3) to formula (5) , center frequency and Lagrangian operator , until the convergence condition shown in formula (6) is met, and the final output is and .

[0096] (6)

[0097] Where: is the discrimination accuracy; Indicates n The iteration k IMF estimated components.

[0098] The ivy algorithm involved in this embodiment is specifically:

[0099] Ivy Algorithm (IVYA) is a new meta-heuristic algorithm based on plant intelligence proposed by Mojtaba Ghasemi et al. in 2024. Its design is inspired by the different life stage behaviors of ivy plants in growth, rise and spread. The main ideas are as follows:

[0100] ① Population initialization

[0101] Given the population of ivy plants is , the dimension of variables to be optimized is D , No. i ( ) individuals are , then the ivy population can be expressed as . Randomly initialize each individual in the ivy population:

[0102] (7)

[0103] Where: , Respectively represent the lower and upper limits of the decision variables of each dimension; rand (1, D ) represents a random number with a uniform distribution in the range [0,1]; is the Hadamard product operator.

[0104] ② Coordinated and orderly population growth

[0105] Assume that the mathematical differential equation for the time course of ivy plant growth and evolution is:

[0106] (8)

[0107] Where: , , Respectively represent the growth rate, growth speed and the correction factor of deviation from growth. Based on a large number of simulation experiments, formula (8) can be modeled as the difference equation shown in formula (9):

[0108] (9)

[0109] Where: , Respectively represent individuals exist t Time and ( t +1) The growth rate at the moment; is a random number; To obey Random numbers from probability density distribution; x is a random variable; is a random vector that follows a standard normal distribution.

[0110] ③ Growth mechanism towards light source

[0111] Inspired by the behavior of ivy plants in nature that often rely on the growth status of neighboring ivy to determine their own growth direction, individual The optimization process is to find the nearest and most important individual To guide, the fitness function value is used to measure the importance of individuals. First, the population According to the fitness function value, the order from the best individual to the worst individual (i.e. ascending order) is rearranged to obtain ,but It can be expressed as:

[0112] (10)

[0113] Where: , Respectively represent the first and individual; The best individual .

[0114] based on Growth knowledge information, individual The mathematical model for moving in the direction of the light source can be expressed as:

[0115] (11)

[0116] Where: For individuals use The new state after widening; For vector The absolute value of the growth rate It can be expressed as:

[0117] (12)

[0118] Where: Iter is the current iteration number; is the Hadamard division operator.

[0119] ④ Spread and evolution

[0120] exist Through formula (10) global search to After that, a possible stage is, Directly towards the best individual in the population The surrounding area climbs upward, that is:

[0121] (13)

[0122] Where: For individuals use The new state after climbing. Then the new growth rate It can be expressed as:

[0123] (14)

[0124] ⑤ Survivor Choice

[0125] make , , Respectively and The fitness function value of , executing formula (11), represents the widening stage of new branch expansion and flowering of ivy plants; if , executing formula (13), represents the elongation stage of the ivy plant's young leaves growing rapidly upward.

[0126] After each iteration, the new population is merged with the old population, and then sorted from the best individual to the worst individual according to the fitness function value, and the top individuals are retained. The population size is used as the population for the next iteration.

[0127] ⑥ Algorithm end judgment

[0128] Determine whether the algorithm has reached the set maximum number of iterations. If so, terminate the IVYA algorithm and output the optimal solution. Otherwise, return to step ②.

[0129] The specific process of optimizing VMD parameters by IVYA algorithm in this embodiment is as follows:

[0130] Quadratic Penalty Factor and the number of modal decompositions K It is a key parameter that needs to be determined in advance in the VMD denoising algorithm. The traditional method of adjusting parameters by relying on manual experience may not accurately represent the true characteristic information of the signal. Therefore, the present invention adopts IVYA (ivy algorithm) to adjust the VMD parameters. and K The IVYA parameter optimization VMD solution process is as follows: Figure 3 As shown, the specific steps are as follows:

[0131] ① Input signal containing noise ;

[0132] ② Initialize the IVYA algorithm, including the maximum number of iterations, population size, population initial position, etc.; initialize the VMD parameter search range and set the quadratic penalty factor , the number of modal decompositions ;

[0133] ③ Construct the fitness function of the IVYA algorithm based on the maximum information coefficient (MIC). MIC can describe the nonlinear similarity relationship between variables and is defined as follows:

[0134] (15)

[0135] Where: x and y is a random variable; is the joint probability density function; , They are x and y The marginal probability density function of ; a , b Indicated by x and y The rows and columns that make up the two-dimensional grid; U is 0.6 of the total amount of data. The MIC between decomposition modes should be minimized to prevent modal aliasing, while the MIC between decomposition modes and input signals should be maximized to retain valid information. Therefore, the fitness function constructed based on MIC is It can be expressed as:

[0136] (16)

[0137] Where: , Respectively k 、( k +1) IMF components; is the original signal; K is the total number of IMF components;

[0138] ④ Continuously updated through IVYA algorithm K and Optimize towards the minimum fitness value until the maximum number of iterations is reached, and the current parameter combination is considered as the optimal solution and output;

[0139] ⑤ Use the optimal parameter combination result of step ④ to input the noisy signal Perform VMD decomposition.

[0140] Step 3: Signal component classification

[0141] Modal component division criteria based on correlation coefficient curvature spectrum. For the IMF components of each frequency band obtained by IVYA-VMD decomposition, the correlation coefficient curvature spectrum is used to divide the modal components according to the critical value. It is divided into two categories: useful signal dominant component and noise dominant component. The correlation coefficient curvature spectrum is defined as the curvature value calculated for the correlation coefficient sequence. k The correlation coefficient between the IMF component and the original signal The calculation formula is as follows:

[0142] (17)

[0143] Where: E and D represent the mathematical expectation and variance respectively.

[0144] like (constant), indicating that the component has a high similarity with the original signal and is classified as the dominant component of the useful signal; if , indicating that the component has a low similarity with the original signal and is classified as a noise-dominated component. c The value is determined by the position corresponding to the maximum curvature value in the correlation coefficient curve. The curvature calculation formula is as follows:

[0145] (18)

[0146] Where: For the z ( ) positions’ correlation coefficient curvature values; K is the total number of IMF components; , , Respectively , z , the first z +1) correlation coefficient; , Respectively z The first and second order derivatives of the correlation coefficients at each position.

[0147] Step 4: Signal component denoising

[0148] Noise-dominant component denoising method based on improved wavelet threshold. The noise-dominant component screened out by the correlation coefficient is denoised by wavelet threshold denoising technology based on discrete wavelet transform, so as to retain the small amount of useful information in the component as much as possible. The basic idea of ​​wavelet threshold denoising technology is: after the noisy signal is transformed by discrete wavelet transform, the wavelet coefficients corresponding to the useful signal are relatively large, and the noise is converted into smaller coefficients. Signal denoising is achieved by threshold processing of the coefficients. The key to the denoising effect lies in the threshold function and the selection of the threshold. The existing threshold function has certain limitations: ① The discontinuity problem in the hard threshold function will cause signal reconstruction oscillation; ② The constant bias problem in the soft threshold function will cause signal reconstruction errors; ③ The traditional improved threshold function often requires additional empirical parameters to be set in advance, which may affect the actual denoising quality. In view of this, the present invention constructs a novel parameter-free improved threshold function based on the hyperbolic tangent function, and its expression is:

[0149] (19)

[0150] Where: , Respectively represent l Tier d High-frequency wavelet coefficients before and after threshold processing at each position; is the hyperbolic tangent function; is a sign function; Represents the threshold value, which is used to determine the boundary between the useful signal coefficient and the noise coefficient, that is:

[0151] (20)

[0152] Where: N is the length of the input IMF component sequence; is the noise standard deviation; express median of absolute values; j Is an imaginary unit.

[0153] Figure 4Comparison between the improved threshold function and the soft and hard threshold functions. The improved threshold function is continuously differentiable, reduces the degree of oscillation and deviation, and does not require additional parameters to control the threshold function, which helps to produce a good denoising effect.

[0154] The basic process of improved threshold wavelet threshold denoising technology is as follows:

[0155] ① Wavelet decomposition. The wavelet base is selected as Dobesi wavelet, and the number of decomposition layers is set between 3 and 6 layers. Multi-scale decomposition is performed on each noise dominant component to obtain the high-frequency wavelet coefficients of each layer. ;

[0156] ② Threshold processing. Combine the improved threshold function of formula (19) and the threshold pair of formula (20) Quantitative processing is performed to obtain ;

[0157] ③ Wavelet reconstruction. By applying inverse wavelet transform, the IMF components after further noise reduction are obtained.

[0158] Step 5: Signal reconstruction

[0159] Finally, the dominant component of the useful signal and the dominant component of the denoised noise are superimposed and reconstructed to obtain the final denoised signal.

[0160] This embodiment provides a joint denoising method for the acoustic signal of the AC filter reactor of a UHV converter station. The method first adaptively decomposes the original acoustic signal containing complex noise through VMD with optimized IVYA algorithm parameters; then divides the decomposed IMF component into a useful signal-dominated component and a noise-dominated component according to the correlation coefficient; finally, the improved parameter-free wavelet threshold denoising algorithm is used to further denoise the noise-dominated component and reconstruct it by superposition with the useful signal-dominated component to obtain the final denoised signal. The method has excellent denoising performance and high adaptability, and can effectively extract pure AC filter reactor acoustic signals, which helps to reduce the difficulty of subsequent feature extraction and pattern recognition tasks, and thus provides a basis for the status evaluation of the AC filter.

[0161] The design background and principle of the present invention are:

[0162] With the construction and development of new power systems, new challenges have been raised to ensure the safe and stable operation of key power equipment. In view of the relative lack of research work in the field of acoustic signal denoising for AC filter field reactors in UHV converter stations at home and abroad, this paper focuses on applying advanced signal processing methods to denoise acoustic signals, which can provide new ideas for AC filter status assessment;

[0163] Aiming at the problem that a single denoising method has too much residual noise information and is difficult to extract pure signals, a denoising method combining VMD and wavelet threshold is proposed. VMD is used to decompose the original complex low signal-to-noise ratio signal into modal components with simple and clear frequency bands, and then the wavelet threshold is used to mine the hidden small amount of useful information from the components with more noise, giving full play to the advantages of the two denoising methods and combining them into a powerful combination;

[0164] In order to solve the problem that the two key parameters of VMD, the quadratic penalty factor and the number of decomposed modes, are difficult to determine based on experience, a novel IVYA algorithm with strong local and global optimization capabilities is introduced to perform parameter adaptive optimization of VMD, effectively improving the denoising effect and efficiency. In addition, unlike the common envelope entropy as the fitness function, a fitness function is constructed using the maximum information coefficient (MIC), which is the ratio of the MIC between modal components to the MIC between the same modal components of the original signal. The design idea of ​​this fitness function is that the MIC between modal components should be minimized to prevent modal aliasing, while the MIC between modal components and the input signal should be maximized to retain effective information;

[0165] Aiming at the problem of difficulty in identifying effective components of VMD decomposition, a IMF component classification criterion based on correlation coefficient curvature spectrum is proposed. This method does not require prior information of signal and noise.

[0166] Aiming at the problems of existing wavelet threshold denoising technology, such as the discontinuity of hard threshold function, the constant deviation of soft threshold function and the need for additional parameters to perform threshold function, a parameter-free threshold function improved according to the characteristics of hyperbolic tangent function was proposed, which effectively overcomes the shortcomings of the existing threshold function and thus produces a good denoising effect.

[0167] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those 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. However, these 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 invention.

Claims

1. A method for denoising an acoustic signal of an AC filter field reactor, characterized in that: The following steps are involved: Acquire the reactor acoustic signal in the AC filter field of the converter station; The optimized VMD algorithm is used to adaptively and non-recursively decompose the original AC filter reactor noise signal into a series of IMF components. Classify multiple IMF components into useful signal-dominated components and noise-dominated components; The noise-dominated signal is denoised by improving the wavelet threshold using a threshold function, and the threshold function expression is as follows: Where: , Respectively represent l Tier d High-frequency wavelet coefficients before and after threshold processing at each position; is the hyperbolic tangent function; is a sign function; Represents the threshold value, which is used to determine the boundary between the useful signal coefficient and the noise coefficient, that is: Where: N is the length of the input IMF component sequence; is the noise standard deviation; express median of absolute values; j is an imaginary unit; The noise dominant component and the effective signal dominant component after threshold processing are reconstructed by superposition to obtain the final denoised sound signal; The optimization process of the VMD algorithm is as follows: the VMD parameters are optimized using the IVYA algorithm. and K The specific steps for adaptive optimization are as follows: ① Input signal containing noise ; ② Initialize the IVYA algorithm, including the maximum number of iterations, population size, and population initial position; initialize the VMD parameter search range and set the quadratic penalty factor , the number of modal decompositions ; ③ The fitness function of the IVYA algorithm is constructed based on the maximum information coefficient and is defined as follows: Where: x and y is a random variable; is the joint probability density function; , They are x and y The marginal probability density function of ; a , b Represents the rows and columns of the two-dimensional grid composed of x and y; U It is 0.6 times the total amount of data; The MIC between decomposition modes should be minimized to prevent modal aliasing, while the MIC between decomposition modes and input signals should be maximized to retain valid information; therefore, the fitness function constructed based on MIC is It can be expressed as: Where: , Respectively k 、( k +1) IMF components; is the original signal; K is the total number of IMF components; ④ Continuously updated through IVYA algorithm K and Optimize towards the minimum fitness value until the maximum number of iterations is reached, and the current parameter combination is considered as the optimal solution and output; ⑤ Use the optimal parameter combination result of step ④ to input the noisy signal Perform VMD decomposition.

2. The method for denoising an acoustic signal of an AC filter field reactor according to claim 1, characterized in that: The specific process of classifying multiple IMF components is as follows: using the correlation coefficient curvature spectrum according to the critical value Divide it into two categories: useful signal dominant component and noise dominant component; The correlation coefficient curvature spectrum is defined as the curvature value calculated for the correlation coefficient sequence; k The correlation coefficient between the IMF component and the original signal The calculation formula is as follows: Where: E and D denote mathematical expectation and variance respectively; like , is a constant, indicating that the component has a high similarity with the original signal and is classified as the dominant component of the useful signal; if , indicating that the component has a low similarity with the original signal and is classified as a noise-dominated component; c The value is determined by the position corresponding to the maximum curvature value in the correlation coefficient curve; the curvature calculation formula is as follows: Where: For the z ( ) positions’ correlation coefficient curvature values; K is the total number of IMF components; , , Respectively , z , the first z +1) correlation coefficient; , Respectively z The first and second order derivatives of the correlation coefficients at each position.

3. The method for denoising an acoustic signal of an AC filter field reactor according to claim 1 or 2, characterized in that: The noise dominant component denoising method based on improved wavelet threshold is as follows: ① Wavelet decomposition: The wavelet base is selected as Dobesi wavelet, and the number of decomposition layers is set between 3 and 6 layers. Multi-scale decomposition is performed on each noise dominant component to obtain the high-frequency wavelet coefficients of each layer. ; ② Threshold processing: combining improved threshold function and threshold right Quantitative processing is performed to obtain ; ③ Wavelet reconstruction: By applying inverse wavelet transform, the IMF components after further noise reduction are obtained.

4. AC filter field reactor acoustic signal denoising system, characterized in that: The following steps are involved: An acoustic signal acquisition module, used to acquire the acoustic signal of the reactor in the AC filter field of the converter station; An IMF component decomposition module is used to adaptively and non-recursively decompose the original AC filter reactor noise signal into a series of IMF components using an optimized VMD algorithm; An IMF component classification module is used to classify multiple IMF components into useful signal-dominated components and noise-dominated components; The secondary denoising module is used to denoise the noise-dominated signal by improving the wavelet threshold using a threshold function. The threshold function expression is as follows: Where: , Respectively represent l Tier d High-frequency wavelet coefficients before and after threshold processing at each position; is the hyperbolic tangent function; is a sign function; Represents the threshold value, which is used to determine the boundary between the useful signal coefficient and the noise coefficient, that is: Where: N is the length of the input IMF component sequence; is the noise standard deviation; express median of absolute values; j is an imaginary unit; A signal superposition and reconstruction module is used to reconstruct the noise dominant component and the effective signal dominant component after threshold processing by superposition to obtain the final denoised sound signal; The optimization process of the VMD algorithm is as follows: the VMD parameters are optimized using the IVYA algorithm. and K The specific steps for adaptive optimization are as follows: ① Input signal containing noise ; ② Initialize the IVYA algorithm, including the maximum number of iterations, population size, and population initial position; initialize the VMD parameter search range and set the quadratic penalty factor , the number of modal decompositions ; ③ The fitness function of the IVYA algorithm is constructed based on the maximum information coefficient and is defined as follows: Where: x and y is a random variable; is the joint probability density function; , They are x and y The marginal probability density function of ; a , b Indicated by x and y The rows and columns that make up the two-dimensional grid; B U It is 0.6 times the total amount of data; The MIC between decomposition modes should be minimized to prevent modal aliasing, while the MIC between decomposition modes and input signals should be maximized to retain valid information; therefore, the fitness function constructed based on MIC is It can be expressed as: Where: , Respectively k 、( k +1) IMF components; is the original signal; K is the total number of IMF components; ④ Continuously updated through IVYA algorithm K and Optimize towards the minimum fitness value until the maximum number of iterations is reached, and the current parameter combination is considered as the optimal solution and output; ⑤ Use the optimal parameter combination result of step ④ to input the noisy signal Perform VMD decomposition.

5. The AC filter field reactor acoustic signal denoising system according to claim 4, characterized in that: The specific process of classifying multiple IMF components is as follows: using the correlation coefficient curvature spectrum according to the critical value Divide it into two categories: useful signal dominant component and noise dominant component; The correlation coefficient curvature spectrum is defined as the curvature value calculated for the correlation coefficient sequence; k The correlation coefficient between the IMF component and the original signal The calculation formula is as follows: Where: E and D denote mathematical expectation and variance respectively; like , is a constant, indicating that the component has a high similarity with the original signal and is classified as the dominant component of the useful signal; if , indicating that the component has a low similarity with the original signal and is classified as a noise-dominated component; c The value is determined by the position corresponding to the maximum curvature value in the correlation coefficient curve; the curvature calculation formula is as follows: Where: For the z ( ) positions’ correlation coefficient curvature values; K is the total number of IMF components; , , Respectively , z , the first z +1) correlation coefficient; , Respectively z The first and second order derivatives of the correlation coefficients at each position.

6. The AC filter field reactor acoustic signal denoising system according to claim 4 or 5, characterized in that: The noise dominant component denoising method based on improved wavelet threshold is as follows: ① Wavelet decomposition: The wavelet base is selected as Dobesi wavelet, and the number of decomposition layers is set between 3 and 6 layers. Multi-scale decomposition is performed on each noise dominant component to obtain the high-frequency wavelet coefficients of each layer. ; ② Threshold processing: combining improved threshold function and threshold right Quantitative processing is performed to obtain ; ③ Wavelet reconstruction: By applying inverse wavelet transform, the IMF components after further noise reduction are obtained.

7. A processing device, characterized in that: It comprises at least one processor and at least one memory in communication with the processor, wherein the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute any method according to claims 1 to 3.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the method according to any one of claims 1 to 3.

Images