A method for denoising transformer voiceprint signals

Abstract

The invention discloses a transformer voiceprint signal denoising method, comprising the following steps: using a sound collection device to collect transformer voiceprint signals; using an improved bobcat optimization algorithm to optimize the bandwidth threshold and B-spline order of time-varying filter empirical mode decomposition, and then using the optimized time-varying filter empirical mode decomposition to decompose the collected transformer voiceprint signal into an eigenmode function; using a sliding window to segmentally intercept the eigenmode function, and constructing the intercepted eigenmode function into a third-order signal tensor; inputting the constructed third-order signal tensor into a Bayesian Gaussian tensor decomposition model for decomposition and reconstruction, and then restoring the reconstructed third-order signal tensor into a one-dimensional vector according to the tensor construction inverse process; the invention uses the improved bobcat optimization algorithm to optimize the parameters of the time-varying filter empirical mode decomposition, thereby avoiding aliasing during decomposition.

Description

Technical Field

[0001] The present invention relates to the technical field of voiceprint signal denoising, and in particular to a transformer voiceprint signal denoising method. Background Art

[0002] As one of the most important electrical equipment in the power system, power transformers need to be kept in a reliable operating state. Currently, transformer troubleshooting mainly relies on regular inspections and visual inspections by specialized inspectors. This inspection method consumes a lot of manpower and is highly subjective.

[0003] When a transformer has partial discharge, it often emits an acoustic signal. Therefore, the transformer's vibration and acoustic signal contains a wealth of information about the transformer's operating status, which can be used as a basis for judging its operating status. However, in addition to collecting the transformer's operating status soundprint signal, the acoustic sensor also has environmental noise, which makes the effective operating status information submerged in various interferences and difficult to process later. Therefore, how to effectively process the transformer's acoustic signal and remove the noise in the soundprint signal becomes the key to subsequent transformer fault diagnosis. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method for denoising a transformer voiceprint signal, which aims to solve the problems mentioned in the background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a transformer voiceprint signal denoising method, comprising the following steps:

[0006] Step S1: using a sound collection device to collect transformer voiceprint signals;

[0007] Step S2: using an improved Bobcat optimization algorithm to optimize the bandwidth threshold and B-spline order of the time-varying filter empirical mode decomposition, and then using the optimized time-varying filter empirical mode decomposition to decompose the collected transformer voiceprint signal into an eigenmode function;

[0008] The specific process of optimizing the bandwidth threshold and B-spline order of time-varying filter empirical mode decomposition using the improved Bobcat optimization algorithm is as follows:

[0009] Step S2.11: setting the population size and maximum number of iterations of the improved Bobcat optimization algorithm;

[0010] Step S2.12: Introduce the sobol sequence to initialize the population of the improved lynx optimization algorithm;

[0011] Step S2.13: simulating attacking prey;

[0012] Step S2.14: simulating the prey chasing behavior and introducing a variable spiral flight strategy to improve the prey chasing behavior;

[0013] Step S2.15: Determine whether the maximum number of iterations has been reached. If not, continue iterating. If reached, stop iterating and output the optimal individual, i.e., the bandwidth threshold of the time-varying filter empirical mode decomposition and the optimal value of the B-spline order;

[0014] Step S3: using a sliding window to segmentally intercept the intrinsic mode function, and constructing the intercepted intrinsic mode function into a third-order signal tensor;

[0015] Step S4: input the constructed third-order signal tensor into the Bayesian Gaussian tensor decomposition model for decomposition and reconstruction, and then restore the reconstructed third-order signal tensor into a one-dimensional vector according to the inverse process of tensor construction.

[0016] Furthermore, the population of the improved Bobcat optimization algorithm is initialized by introducing the sobol sequence, which is expressed as:

[0017] ;

[0018] ;

[0019] In the formula, Represents the value of the nth individual in the d-1th dimension; represents the population of the improved Bobcat optimization algorithm; Indicates The individual, the bobcat, is The values ​​on the dimensions, i∈1,2,...,n; j∈1,2,...,d; n represents the size of the population; d represents the dimension of the initial population uneven distribution problem; and They represent the upper and lower bounds of the problem of uneven initial population distribution; Represents a random number uniformly distributed within a preset range obtained by using the sobol sequence.

[0020] Furthermore, the simulated attack on prey behavior is expressed as:

[0021] ;

[0022] ;

[0023] In the formula, Indicates the first The updated position of each individual; is the current iteration number; Represents a random number between 0 and 1; represents a number randomly selected from the set {1, 2}; Indicates the location of prey; Indicates the first The fitness value of each individual after the position is updated; It represents the fitness value of the best individual in the population that attacks prey; represents the location of the best individual in the population to attack prey; Indicates the first The current location of an individual.

[0024] Furthermore, the simulation of the prey chasing behavior and the introduction of a variable spiral flight strategy to improve the prey chasing behavior are expressed as:

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] In the formula, Indicates the first The updated position of each individual; Indicates the first The fitness value of each individual after the position is updated; Represents a random number uniformly distributed between [-1,1]; Indicates the maximum number of iterations; represents a constant that changes with the number of iterations; represents the best individual in the population chasing prey and the first individual in the population chasing prey The distance between individuals in the iteration; represents the coefficient of variation; Indicates the first The current location of each individual, Represents the natural logarithm.

[0030] Furthermore, the specific process of using the optimized time-varying filtering empirical mode decomposition to decompose the collected transformer voiceprint signal into the eigenmode function is as follows:

[0031] Step S2.21: First, use Hilbert transform to find the transformer voiceprint signal The instantaneous frequency and instantaneous amplitude a(t), and then calculate the instantaneous frequency and the local maximum and local minimum of the instantaneous amplitude a(t), for the instantaneous frequency The difference between the local maximum and local minimum of the instantaneous amplitude a(t) is calculated. The difference between the local maximum and the local minimum The difference between the local maximum and local minimum of a(t) Finally, a time-varying filter is used to process the transformer voiceprint signal to obtain the instantaneous mean and the instantaneous envelope ,according to , , and Calculate the local cutoff frequency :

[0032] ;

[0033] In the formula, Indicates time;

[0034] Step S2.22: Based on the local cutoff frequency Reconstruct transformer voiceprint signal:

[0035] ;

[0036] In the formula, represents the reconstructed transformer voiceprint signal; represents differential;

[0037] Step S2.23: According to Loughlin instantaneous bandwidth and the weighted mean instantaneous frequency Calculate the cutoff frequency , and determine the cutoff frequency Whether the stopping criteria are met:

[0038] ;

[0039] ;

[0040] ;

[0041] ;

[0042] ;

[0043] In the formula, represents the weighted mean instantaneous frequency; Indicates The instantaneous frequency of the order component; represents the Loughlin instantaneous bandwidth; represents the instantaneous amplitude of the first-order component; represents the instantaneous amplitude of the second-order component; represents the instantaneous frequency of the first-order component; The instantaneous frequency of the second-order component;

[0044] Assume that Loughlin instantaneous bandwidth threshold is ξ, when ≤ξ, then judge Corresponding transformer voiceprint signal is an intrinsic mode function, otherwise, B-spline interpolation is used Corresponding transformer voiceprint signal Approximate and get the approximation result m(t), that is, , and repeat steps S2.21 to S2.23 until ≤ξ.

[0045] Furthermore, the specific process of using a sliding window to segmentally intercept the intrinsic mode function and constructing the intercepted intrinsic mode function into a third-order signal tensor is:

[0046] Step S3.1: Use a segmented truncation sliding window with a set length of w to segmentally truncate each intrinsic mode function to obtain multiple truncated sub-signals, and use the multiple truncated sub-signals to construct a truncated sub-signal matrix. The expression is:

[0047] ;

[0048] In the formula, represents the number of truncated sub-signals, and N is the length of the truncated sub-signals; represents the wth value in the eigenmode function; Intrinsic mode function IMF Values;

[0049] When the length of the last truncated sub-signal is less than the set length w, the right end of the last truncated sub-signal is filled with the average of all truncated sub-signals that have been truncated;

[0050] Step S3.2: Truncated sub-signal matrix Constructed into a third-order signal tensor.

[0051] Furthermore, the specific process of inputting the constructed third-order signal tensor into the Bayesian Gaussian tensor decomposition model for decomposition and reconstruction, and then restoring the reconstructed third-order signal tensor into a one-dimensional vector according to the tensor construction inverse process is:

[0052] Step S4.1: Decompose and reconstruct the third-order signal tensor using the Bayesian Gaussian tensor decomposition model. The Bayesian Gaussian tensor decomposition model is expressed as:

[0053] ;

[0054] In the formula, , , represents the factor matrix; represents the vector outer product; is the CP rank of the third-order signal tensor; Represents the constructed third-order signal tensor;

[0055] Step S4.2: According to the inverse process of the third-order signal tensor reconstruction, the reconstructed third-order signal tensor is restored to a one-dimensional vector to achieve denoising of the transformer voiceprint signal.

[0056] Compared with the existing technology, the present invention has the following beneficial effects: the present invention uses the improved Bobcat optimization algorithm to optimize the parameters of the time-varying filter empirical mode decomposition, avoids the aliasing phenomenon during decomposition, improves the signal-to-noise ratio of the voiceprint signal, and uses the Bayesian Gaussian tensor decomposition model to decompose and reconstruct the constructed tensor, making the signal clearer, reducing interference components, and improving the quality and reliability of the signal, thereby achieving the purpose of improving the transformer fault diagnosis rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a flow chart of the steps of the present invention. DETAILED DESCRIPTION

[0058] like Figure 1 As shown, the present invention provides a technical solution: a transformer voiceprint signal denoising method, comprising the following steps:

[0059] Step S1: using a sound collection device to collect transformer voiceprint signals.

[0060] Step S2: using an improved bobcat optimization algorithm to optimize the bandwidth threshold and B-spline order of a time-varying filter empirical mode decomposition (TVF-EMD), and then using the optimized time-varying filter empirical mode decomposition (ISOA-TVF-EMD) to decompose the collected transformer soundprint signal into an eigenmode function (IMF); the improved bobcat optimization algorithm is an improvement of the original bobcat optimization algorithm (Serval Optimization) by introducing a variable spiral flight strategy into the prey chasing behavior.

[0061] Step S3: Using a sliding window to segmentally intercept the intrinsic mode function (IMF), and construct the intercepted intrinsic mode function (IMF) into a third-order signal tensor.

[0062] Step S4: input the constructed third-order signal tensor into the Bayesian Gaussian tensor decomposition model for decomposition and reconstruction, and then restore the reconstructed third-order signal tensor into a one-dimensional vector according to the inverse process of tensor construction.

[0063] Among them, the specific process of optimizing the bandwidth threshold and B-spline order of time-varying filter empirical mode decomposition (TVF-EMD) using the improved Bobcat optimization algorithm is as follows:

[0064] Step S2.11: Set the population size and maximum number of iterations of the improved Bobcat optimization algorithm.

[0065] Step S2.12: Introduce the sobol sequence to initialize the population of the improved Bobcat optimization algorithm to solve the problem of uneven distribution of the initial population caused by the random generation of the initial population of the original Bobcat optimization algorithm, which is expressed as:

[0066] (1);

[0067] (2);

[0068] In the formula, Represents the value of the nth individual in the d-1th dimension; represents the population of the improved Bobcat optimization algorithm; Indicates Individuals (lynx) in the The values ​​on the dimensions, i∈1,2,...,n; j∈1,2,...,d; n represents the size of the population; d represents the dimension of the initial population uneven distribution problem; and They represent the upper and lower bounds of the problem of uneven initial population distribution; Represents a random number uniformly distributed within a preset range obtained by using the sobol sequence.

[0069] Step S2.13: Simulate the attack behavior of prey, expressed as:

[0070] (3);

[0071] (4);

[0072] In the formula, Indicates the first The updated position of each individual; is the current iteration number; Represents a random number between 0 and 1; represents a number randomly selected from the set {1, 2}; Indicates the location of prey; Indicates the first The fitness value of each individual after the position is updated; It represents the fitness value of the best individual in the population that attacks prey; represents the location of the best individual in the population to attack prey; Indicates the first The current location of an individual.

[0073] Step S2.14: simulate the behavior of chasing prey, and introduce a variable spiral flight strategy to improve the behavior of chasing prey to solve the problem that when the algorithm falls into the local optimum, the algorithm cannot jump out of the local optimum, which is expressed as:

[0074] (5);

[0075] (6);

[0076] (7);

[0077] (8);

[0078] In the formula, Indicates the first The updated position of each individual; Indicates the first The fitness value of each individual after the position is updated; Represents a random number uniformly distributed between [-1,1]; Indicates the maximum number of iterations; represents a constant that changes with the number of iterations; represents the best individual in the population chasing prey and the first individual in the population chasing prey The distance between individuals in the iteration; represents the coefficient of variation, which is 5 in this embodiment; Indicates the first The current location of each individual; Represents the natural logarithm.

[0079] Step S2.15: Determine whether the maximum number of iterations has been reached. If not, continue iterating. If reached, stop iterating and output the optimal individual, i.e., the bandwidth threshold of the time-varying filter empirical mode decomposition (TVF-EMD) and the optimal value of the B-spline order.

[0080] Among them, the specific process of using the optimized time-varying filter empirical mode decomposition (ISOA-TVF-EMD) to decompose the collected transformer voiceprint signal into the eigenmode function (IMF) is as follows:

[0081] Step S2.21: First, use Hilbert transform to find the transformer voiceprint signal The instantaneous frequency and instantaneous amplitude a(t), and then calculate the instantaneous frequency and the local maximum and local minimum of the instantaneous amplitude a(t), for the instantaneous frequency The difference between the local maximum and local minimum of the instantaneous amplitude a(t) is calculated. The difference between the local maximum and the local minimum The difference between the local maximum and local minimum of a(t) Finally, a time-varying filter is used to process the transformer voiceprint signal to obtain the instantaneous mean and the instantaneous envelope ,according to , , and Calculate the local cutoff frequency :

[0082] (9);

[0083] In the formula, Indicates time.

[0084] Step S2.22: Based on the local cutoff frequency Reconstruct transformer voiceprint signal:

[0085] (10);

[0086] In the formula, represents the reconstructed transformer voiceprint signal; Represents differential.

[0087] Step S2.23: According to Loughlin instantaneous bandwidth and the weighted mean instantaneous frequency Calculate the cutoff frequency , and determine the cutoff frequency Whether the stopping criteria are met:

[0088] (11);

[0089] (12);

[0090] (13);

[0091] (14);

[0092] (15);

[0093] In the formula, represents the weighted mean instantaneous frequency; Indicates The instantaneous frequency of the order component; represents the Loughlin instantaneous bandwidth; represents the instantaneous amplitude of the first-order component; represents the instantaneous amplitude of the second-order component; represents the instantaneous frequency of the first-order component; The instantaneous frequency of the 2nd order component.

[0094] Assume that Loughlin instantaneous bandwidth threshold is ξ, when ≤ξ, then judge Corresponding transformer voiceprint signal is an intrinsic mode function (IMF), otherwise, B-spline interpolation is used Corresponding transformer voiceprint signal Approximate and get the approximation result m(t), that is, , and repeat steps S2.21 to S2.23 until ≤ξ.

[0095] The specific process of step S3 is as follows:

[0096] Step S3.1: Use a segmented truncation sliding window with a set length of w to segmentally truncate each intrinsic mode function (IMF) to obtain multiple truncated sub-signals, and use the multiple truncated sub-signals to construct a truncated sub-signal matrix. The expression is:

[0097] (16);

[0098] In the formula, represents the number of truncated sub-signals, and N is the length of the truncated sub-signals; represents the wth value in the intrinsic mode function (IMF); represents the first intrinsic mode function (IMF) value.

[0099] When the length of the last truncated sub-signal is less than the set length w, the right end of the last truncated sub-signal is filled with the average value of all truncated sub-signals that have been truncated.

[0100] Step S3.2: Truncated sub-signal matrix Constructed into a third-order signal tensor.

[0101] The specific process of step S4 is as follows:

[0102] Step S4.1: Decompose and reconstruct the third-order signal tensor using the Bayesian Gaussian tensor decomposition model to achieve the purpose of denoising. The Bayesian Gaussian tensor decomposition model is expressed as:

[0103] (17);

[0104] In the formula, , , represents the factor matrix; represents the vector outer product; is the CP rank of the third-order signal tensor; Represents the constructed third-order signal tensor.

[0105] Step S4.2: According to the inverse process of the third-order signal tensor reconstruction, the reconstructed third-order signal tensor is restored to a one-dimensional vector to achieve denoising of the transformer voiceprint signal.

[0106] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A transformer voiceprint signal denoising method, characterized in that: The steps include: Step S1: using a sound collection device to collect transformer voiceprint signals; Step S2: using an improved Bobcat optimization algorithm to optimize the bandwidth threshold and B-spline order of the time-varying filter empirical mode decomposition, and then using the optimized time-varying filter empirical mode decomposition to decompose the collected transformer voiceprint signal into an eigenmode function; The specific process of optimizing the bandwidth threshold and B-spline order of time-varying filter empirical mode decomposition using the improved Bobcat optimization algorithm is as follows: Step S2.11: setting the population size and maximum number of iterations of the improved Bobcat optimization algorithm; Step S2.12: Introduce the sobol sequence to initialize the population of the improved lynx optimization algorithm; Step S2.13: simulating attacking prey; Step S2.14: simulate the prey chasing behavior and introduce a variable spiral flight strategy to improve the prey chasing behavior, which is expressed as: ； ； ； ； In the formula, Indicates the first The updated position of each individual; is the current iteration number; Indicates The individual, the bobcat, is The values ​​on the dimensions, i∈1,2,...,n; j∈1,2,...,d; n represents the size of the population; d represents the dimension of the initial population uneven distribution problem; Represents a random number uniformly distributed between [-1,1]; represents the natural logarithm; represents the coefficient of variation; represents a constant that changes with the number of iterations; represents the best individual in the population chasing prey and the first individual in the population chasing prey The distance between individuals in the iteration; represents the location of the best individual in the population to attack prey; Indicates the maximum number of iterations; Indicates the first The current location of each individual; Indicates the first The fitness value of each individual after the position is updated; It represents the fitness value of the best individual in the population that attacks prey; Step S2.15: Determine whether the maximum number of iterations has been reached. If not, continue iterating. If reached, stop iterating and output the optimal individual, i.e., the bandwidth threshold of the time-varying filter empirical mode decomposition and the optimal value of the B-spline order; Step S3: using a sliding window to segmentally intercept the intrinsic mode function, and constructing the intercepted intrinsic mode function into a third-order signal tensor; Step S4: input the constructed third-order signal tensor into the Bayesian Gaussian tensor decomposition model for decomposition and reconstruction, and then restore the reconstructed third-order signal tensor into a one-dimensional vector according to the inverse process of tensor construction; The specific process of step S3 is: Step S3.1: Use a segmented truncation sliding window with a set length of w to segmentally truncate each intrinsic mode function to obtain multiple truncated sub-signals, and use the multiple truncated sub-signals to construct a truncated sub-signal matrix. The expression is: ； In the formula, represents the number of truncated sub-signals, and N is the length of the truncated sub-signals; represents the wth value in the eigenmode function; Intrinsic mode function IMF values; When the length of the last truncated sub-signal is less than the set length w, the right end of the last truncated sub-signal is filled with the average of all truncated sub-signals that have been truncated; Step S3.2: Truncated sub-signal matrix Constructed into a third-order signal tensor; The specific process of step S4 is: Step S4.1: Decompose and reconstruct the third-order signal tensor using the Bayesian Gaussian tensor decomposition model. The Bayesian Gaussian tensor decomposition model is expressed as: ； In the formula, , , represents the factor matrix; represents the vector outer product; is the CP rank of the third-order signal tensor; Represents the constructed third-order signal tensor; Step S4.2: According to the inverse process of the third-order signal tensor reconstruction, the reconstructed third-order signal tensor is restored to a one-dimensional vector to achieve denoising of the transformer voiceprint signal.

2. A transformer voiceprint signal denoising method according to claim 1, characterized in that: The population of the improved Bobcat optimization algorithm is initialized by introducing the sobol sequence, which is expressed as: ； ； In the formula, Represents the value of the nth individual in the d-1th dimension; represents the population of the improved Bobcat optimization algorithm; and They represent the upper and lower bounds of the problem of uneven initial population distribution; Represents a random number uniformly distributed within a preset range obtained by using the sobol sequence.

3. A transformer voiceprint signal denoising method according to claim 2, characterized in that: The simulated attack on prey behavior is expressed as: ； ； In the formula, Indicates the first The updated position of each individual; is the current iteration number; Represents a random number between 0 and 1; represents a number randomly selected from the set {1, 2}; Indicates the location of prey; Indicates the first The fitness value of each individual after the position is updated; It represents the fitness value of the best individual in the population that attacks prey; represents the location of the best individual in the population to attack prey; Indicates the first The current location of an individual.

4. A transformer voiceprint signal denoising method according to claim 3, characterized in that: The specific process of using the optimized time-varying filter empirical mode decomposition to decompose the collected transformer voiceprint signal into the eigenmode function is as follows: Step S2.21: First, use Hilbert transform to find the transformer voiceprint signal The instantaneous frequency and instantaneous amplitude a(t), and then calculate the instantaneous frequency and the local maximum and local minimum of the instantaneous amplitude a(t), for the instantaneous frequency The difference between the local maximum and local minimum of the instantaneous amplitude a(t) is calculated. The difference between the local maximum and the local minimum The difference between the local maximum and local minimum of a(t) Finally, a time-varying filter is used to process the transformer voiceprint signal to obtain the instantaneous mean and the instantaneous envelope ,according to , , and Calculate the local cutoff frequency : ； In the formula, Indicates time; Step S2.22: Based on the local cutoff frequency Reconstruct transformer voiceprint signal: ； In the formula, represents the reconstructed transformer voiceprint signal; represents differential; Step S2.23: According to Loughlin instantaneous bandwidth and the weighted mean instantaneous frequency Calculate the cutoff frequency , and determine the cutoff frequency Whether the stopping criteria are met: ； ； ； ； ； In the formula, represents the weighted mean instantaneous frequency; Indicates The instantaneous frequency of the order component; represents the Loughlin instantaneous bandwidth; represents the instantaneous amplitude of the first-order component; represents the instantaneous amplitude of the second-order component; represents the instantaneous frequency of the first-order component; The instantaneous frequency of the second-order component; Assume that Loughlin instantaneous bandwidth threshold is ξ, when ≤ξ, then judge Corresponding transformer voiceprint signal is an intrinsic mode function, otherwise, B-spline interpolation is used Corresponding transformer voiceprint signal Approximate and get the approximation result m(t), that is, , and repeat steps S2.21 to S2.23 until ≤ξ.

Images