Transformer background noise removal method based on blind source separation

Abstract

The invention discloses a transformer background noise removal method based on blind source separation, and the method comprises the following steps: collecting a voiceprint signal of a transformer, and carrying out the preprocessing of the collected voiceprint signal through an RLS algorithm; using an improved fishing optimization algorithm to replace a joint diagonalization operation in a JADE algorithm to obtain an optimal parameter of the JADE algorithm, and forming an improved JADE algorithm; performing blind source separation on the pre-processed voiceprint signal by using an improved JADE algorithm; the improved fishing optimization algorithm is characterized in that an adaptive factor is introduced to improve an independent search behavior, and an escape formula is introduced to improve a development stage; according to the invention, blind source separation is carried out by adopting the JADE algorithm optimized by the improved fishing optimization algorithm, the problem that parameters of the JADE algorithm are difficult to accurately select is solved, noise in a complex environment of the transformer can be separated, and blind source separation of the JADE algorithm optimized by the improved fishing optimization algorithm is more accurate.

Description

Technical Field

[0001] The invention relates to the field of technical noise removal, and in particular to a method for removing transformer background noise based on blind source separation. Background Art

[0002] At present, it is increasingly important and urgent to carry out operating status detection of power equipment with ultra-high voltage and above voltage levels, especially the detection of abnormal conditions. As an important part of the power system, the power transformer is one of the most important electrical equipment in the substation. Generally speaking, the abnormal state of the transformer can be divided into core abnormality and winding abnormality. The core abnormality is mainly manifested as core saturation, and the winding abnormality usually includes winding deformation, winding looseness, etc.

[0003] The basic principle of transformer abnormal state detection is to extract various characteristic quantities in transformer operation, analyze, identify and track the characteristic quantities to monitor the abnormal operation state of the transformer. The detection methods can be divided into invasive detection and non-invasive detection according to the degree of contact; they can be divided into live detection and power-off detection according to whether the shutdown detection is required; they can be divided into electrical quantity method and non-electrical quantity method according to the type of detection quantity. In comparison, non-invasive detection has strong portability and is more convenient to install; live detection does not affect the operation of the transformer; non-electrical quantity method has no electrical connection with the power system and is safer. The commonly used detection methods for the current transformer operation state include pulse current method and ultrasonic detection method for detecting partial discharge, frequency response method for detecting winding deformation, and voiceprint detection method for detecting mechanical and electrical faults. These detection methods mainly detect the insulation condition and mechanical structure condition of the transformer. The latest transformer voiceprint signal detection is more comprehensive in comparison, and can respond to most transformer faults and abnormal conditions. Since the voiceprint detection method uses the voiceprint signal emitted by the transformer, it is easily affected by background noise. Therefore, how to effectively remove the background noise is the key to the success of this method.

[0004] As a core component of power transmission, the operating stability of the transformer directly affects the safety and efficiency of the power grid. Transformer voiceprint aliasing is a challenging problem because the collected voiceprint signals are mixed with background noise, so it is difficult to train a high-precision and generalized voiceprint recognition model. This reduces the accuracy and reliability of transformer operating status monitoring through voiceprints. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a method for removing background noise of a transformer based on blind source separation, which aims to solve the problems in the background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for removing transformer background noise based on blind source separation, comprising the following steps:

[0007] Step S1: collecting the voiceprint signal of the transformer, and preprocessing the collected voiceprint signal through the RLS algorithm;

[0008] Step S2: Using the improved fishing optimization algorithm to replace the joint diagonalization operation in the JADE algorithm, obtaining the optimal parameters of the JADE algorithm, and forming a JADE algorithm based on the improved fishing optimization;

[0009] Step S3: using the JADE algorithm based on improved fishing optimization to perform blind source separation on the pre-processed voiceprint signal;

[0010] In step S2, the improved fishing optimization algorithm is used to replace the joint diagonalization operation in the JADE algorithm. The specific process of obtaining the optimal parameters of the JADE algorithm is as follows:

[0011] Step S2.1: setting the initial population and maximum number of iterations of the improved fishing optimization algorithm;

[0012] Step S2.2: Initialize the improved fishing optimization algorithm, use the parameters of the JADE algorithm as the initial fisher population of the improved fishing optimization algorithm, and set the parameters of each individual fisherman in the initial fisher population;

[0013] Step S2.3: Evaluate the fitness of each individual fisherman and calculate the fitness function value of the fisher population position;

[0014] Step S2.4: simulating the search phase of the improved fishing optimization algorithm and introducing an adaptive factor to improve the independent search behavior; the search phase of the improved fishing optimization algorithm includes independent search behavior and group capture behavior;

[0015] According to the capture rate α, the fishermen choose independent search behavior or group capture behavior; when p < α, the fishermen choose independent search behavior, and when p ≥ α, the fishermen choose group capture behavior, where p is a random number between (0, 1);

[0016] Step S2.5: simulate the development phase of the improved fishing optimization algorithm and introduce the escape formula for improvement;

[0017] Step S2.6: Determine whether the current number of iterations reaches the preset maximum number of iterations. If it reaches the preset maximum number of iterations, stop iterating. If it does not reach the preset maximum number of iterations, return to step S2.3 and continue iterating and updating.

[0018] Step S2.7: Output the optimal individual fisherman in the current fisher population, that is, the optimal parameters of the JADE algorithm.

[0019] Furthermore, the improved fishing optimization algorithm is initialized, the parameters of the JADE algorithm are used as the initial fisher population of the improved fishing optimization algorithm, and the parameters of each individual fisherman in the initial fisher population are set, which is expressed as:

[0020]

[0021] x i,j =rand×(UB j -LB j )+LB j ;

[0022] In the formula, X represents the fisher population; i,j represents the value of the i-th fisherman in the j-th dimension, i∈1,2,...,n, j∈1,2,...,d; n represents the size of the fisherman population, d represents the dimension of the problem; rand represents a random number in the range [0,1]; LB j UB j They represent the upper and lower bounds of the j-dimensional search space respectively; each fisherman individual represents a set of parameters of the JADE algorithm.

[0023] Furthermore, the fitness function value Fitness of the fishermen population position is calculated:

[0024]

[0025] In the formula, K(x i,j ) represents the fourth-order cumulative value of the voiceprint signal.

[0026] Further, the capture rate α is expressed as:

[0027]

[0028] Where α is the capture rate; t is the current number of iterations; and T represents the maximum number of iterations.

[0029] Furthermore, an adaptive factor is introduced to improve the independent search behavior, which is expressed as:

[0030]

[0031] In the formula, fit i represents the fitness value of the i-th fisherman individual; fit pos represents the fitness value of any individual fisherman in the fisher population; fit min and fit max Represent the worst fitness value and the best fitness respectively; R represents the exploration range; represents the position of a random fisherman individual after t+1 iterations in the jth dimension; and represents the position of the i-th fisherman individual after t and t+1 iterations in the j-th dimension; r s represents a random number between [0, 1]; Dis represents the Euclidean distance between the i-th fisher individual and the reference object; s represents a random unit vector.

[0032] Furthermore, the group capture behavior is expressed as:

[0033]

[0034] In the formula, Centre c It indicates the target point surrounded by group c, which is composed of 3-4 fisher individuals in the fisher population; represents the average value of individual fishermen in group c; and They represent the position of the i-th fisherman in group c after the t and t+1 iterations in the j dimension respectively; r2 represents the speed at which the fisherman approaches the center, with a value range of (0,1); r3 is the offset of the movement, with a value range of (-1,1).

[0035] Furthermore, the development phase of the improved fishing optimization algorithm is simulated, and the escape formula is introduced for improvement, which is expressed as:

[0036]

[0037] In the formula, σ represents a constant that changes with the number of iterations; GD represents the Gaussian distribution function; X best represents the position of the global optimal solution; mean(X) represents the matrix of the mean values ​​of each dimension of the fisher population center; r4 represents a random number with a value of {1,2,3}; represents the fisherman individual closest to the i-th fisherman individual; r5 represents a random number between 0 and 1.

[0038] Furthermore, the voiceprint signal of the transformer is collected, and the collected voiceprint signal is preprocessed by the RLS algorithm, which is expressed as:

[0039] w(k)=w(k-1)+K k ζ k ;

[0040] ζ k =d(k)-u T′ (k)w(k-1);

[0041]

[0042] Where w(k) is the filter coefficient vector at time k; kis the expected difference at time k; d(k) is the expected response vector at time k; u(k) is the input vector at time k; K k is the gain vector at time k; P k is the inverse matrix of the autocorrelation matrix of the input voiceprint signal; w(k-1) represents the filter coefficient vector at time k-1; u T′ (k) represents the transpose of the input vector at time k; P k-1 It represents the inverse matrix of the autocorrelation matrix of the input voiceprint signal at time k-1; λ is the forgetting factor.

[0043] Furthermore, the specific process of step S3 is as follows:

[0044] Step S3.1: input the preprocessed voiceprint signal into the JADE algorithm based on improved fishing optimization, and in the JADE algorithm based on improved fishing optimization, the preprocessed voiceprint signal is combined with an unknown mixing matrix to obtain an observation signal;

[0045] Step S3.2: performing spheroidization preprocessing on the observed signal through the spheroidization matrix to obtain the fourth-order cumulative amount of the spheroidization process of the observed signal;

[0046] Step S3.3: Perform a joint diagonalized orthogonal matrix estimation on the fourth-order cumulants of the spheroidization process of the observed signal, and finally calculate the estimated value to achieve blind source separation.

[0047] Compared with the existing technology, the present invention has the following beneficial effects: the present invention solves the problem that it is difficult to accurately select parameters of the JADE algorithm by adopting the JADE algorithm optimized by the improved fishing optimization algorithm for blind source separation, and can separate the noise in the complex environment of the transformer. Moreover, the blind source separation of the JADE algorithm optimized by the improved fishing optimization algorithm will be more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the steps of the present invention.

[0049] Figure 2 This is a comparison chart of the convergence curves before and after the improvement of the fishing optimization algorithm. DETAILED DESCRIPTION

[0050] like Figure 1 As shown, the present invention provides a technical solution: a method for removing transformer background noise based on blind source separation, comprising the following steps:

[0051] Step S1: Use a sound collection device to collect the voiceprint signal of the transformer, and pre-process the collected voiceprint signal through the RLS algorithm to make the signal clearer, reduce interference components, improve the quality and reliability of the signal, and form a voiceprint signal data set with the pre-processed voiceprint signal.

[0052] Step S2: Use the improved fishing optimization algorithm (ICFOA) to replace the joint diagonalization operation in the JADE algorithm to obtain the optimal parameters of the JADE algorithm, and construct a JADE algorithm based on improved fishing optimization (ICFOA-JADE).

[0053] Step S3: Use the improved fishing optimization based JADE algorithm (ICFOA-JADE) to perform blind source separation on the preprocessed voiceprint signal.

[0054] Among them, a sound collection device is used to collect the voiceprint signal of the transformer, and the collected voiceprint signal is preprocessed by the RLS algorithm to make the signal clearer, reduce interference components, and improve the quality and reliability of the signal. The preprocessed voiceprint signal constitutes a voiceprint signal data set, which is expressed as:

[0055] w(k)=w(k-1)+K k ζ k (1);

[0056] ζ k =d(k)-u T′ (k)w(k-1)(2);

[0057]

[0058] Where w(k) is the filter coefficient vector at time k; k is the expected difference at time k; d(k) is the expected response vector at time k; u(k) is the input vector at time k; K k is the gain vector at time k; P k is the inverse matrix of the autocorrelation matrix of the input voiceprint signal; w(k-1) represents the filter coefficient vector at time k-1; u T′ (k) represents the transpose of the input vector at time k; P k-1 represents the inverse matrix of the autocorrelation matrix of the input voiceprint signal at time k-1; λ is the forgetting factor, which is 0.98 in this implementation.

[0059] Among them, in step S2, the improved fishing optimization algorithm (ICFOA) is used to replace the joint diagonalization operation in the JADE algorithm, and the specific process of obtaining the optimal parameters of the JADE algorithm is:

[0060] Step S2.1: Set the initial population and maximum number of iterations of the improved fishing optimization algorithm (ICFOA).

[0061] Step S2.2: Initialize the improved fishing optimization algorithm, use the parameters of the JADE algorithm as the initial fisher population of the improved fishing optimization algorithm, and set the parameters of each individual fisherman in the initial fisher population; expressed as:

[0062]

[0063] x i,j =rand×(UB j -LB j )+LB j (6);

[0064] In the formula, X represents the fisher population; i,j represents the value of the i-th fisherman in the j-th dimension, i∈1,2,...,n, j∈1,2,...,d; n represents the size of the fisherman population, d represents the dimension of the problem; each fisherman represents a set of parameters of the JADE algorithm; rand represents a random number in the range [0,1]; LB j UB j They represent the upper and lower bounds of the j-dimensional search space respectively.

[0065] Step S2.3: Evaluate the fitness of each individual fisherman. The fitness of each fisherman is the performance of each set of parameters of the JADE algorithm. Calculate the fitness function value Fitness of the fisherman population position:

[0066]

[0067] In the formula, K(x i,j ) represents the fourth-order cumulative value (also known as kurtosis) of the voiceprint signal.

[0068] Step S2.4: Simulate the search phase of the improved fishing optimization algorithm, including independent search behavior and group capture behavior; select independent search behavior or group capture behavior according to the capture rate α; the capture rate α is expressed as:

[0069]

[0070] Where α is the capture rate; t is the current iteration number; T max Indicates the maximum number of iterations.

[0071] Individual fishermen can freely choose between independent search behavior or group capture behavior, but when the capture rate α is high, the capture strategy of individual fishermen tends to be independent search behavior. As α decreases, individual fishermen will switch to group capture behavior. When p<α, individual fishermen choose independent search behavior, and when p≥α, individual fishermen choose group capture behavior, where p is a random number between (0,1).

[0072] Step S2.41: Introduce an adaptive factor to improve the independent search behavior, expressed as:

[0073]

[0074] In the formula, fit i represents the fitness value of the i-th fisherman individual; fit pos represents the fitness value of any individual fisherman in the fisher population; fit min and fit max Represent the worst fitness value and the best fitness respectively; R represents the exploration range; represents the position of a random fisherman individual after t+1 iterations in the jth dimension; and represents the position of the i-th fisherman individual after t and t+1 iterations in the j-th dimension; r s represents a random number between [0, 1]; Dis represents the Euclidean distance between the i-th fisher individual and the reference object; s represents a random unit vector.

[0075] Step S2.42: Group capture behavior, expressed as:

[0076]

[0077] In the formula, Centre c It indicates the target point surrounded by group c, which is composed of 3-4 fisher individuals in the fisher population; represents the average value of individual fishermen in group c; and They represent the position of the i-th fisherman in group c after the t and t+1 iterations in the j dimension respectively; r2 represents the speed at which the fisherman approaches the center, with a value range of (0,1); r3 is the offset of the movement, with a value range of (-1,1).

[0078] Step S2.5: Simulate the development phase of the improved fishing optimization algorithm. In view of the fact that the late group capture behavior of the fisher population is prone to fall into the local optimal value, the escape formula is introduced for improvement, which is expressed as:

[0079]

[0080] In the formula, σ represents a constant that changes with the number of iterations; GD represents the Gaussian distribution function; X best represents the position of the global optimal solution; mean(X) represents the matrix of the mean values ​​of each dimension of the fisher population center; r4 represents a random number with a value of {1,2,3}; represents the fisherman individual closest to the i-th fisherman individual; r5 represents a random number between 0 and 1.

[0081] Step S2.6: Determine whether the current number of iterations has reached the preset maximum number of iterations. If so, stop iterating. If not, return to step S2.3 and continue iterating and updating.

[0082] Step S2.7: Output the optimal individual fisherman in the current fisher population, that is, the optimal parameters of the JADE algorithm.

[0083] The specific process of step S3 is:

[0084] Step S3.1: Input the preprocessed voiceprint signal into the JADE algorithm based on improved fishing optimization (ICFOA-JADE). In the JADE algorithm based on improved fishing optimization (ICFOA-JADE), the preprocessed voiceprint signal is combined with an unknown mixing matrix to obtain an observation signal.

[0085] Step S3.2: Perform spheroidization preprocessing on the observed signal through the spheroidization matrix to obtain the fourth-order cumulant of the spheroidization process of the observed signal.

[0086] Step S3.3: Perform a joint diagonalized orthogonal matrix estimation on the fourth-order cumulants of the spheroidization process of the observed signal, and finally calculate the estimated value to achieve blind source separation.

[0087] like Figure 2 As shown in the figure, the improved fishing optimization algorithm (ICFOA) has better optimization effect and is faster than the original fishing optimization algorithm (CFOA). It can find the parameters of the JADE algorithm faster and better, making the blind source separation effect of the JADE algorithm better.

[0088] 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 method for removing transformer background noise based on blind source separation, characterized in that: The steps include: Step S1: collecting the voiceprint signal of the transformer, and preprocessing the collected voiceprint signal through the RLS algorithm; Step S2: Using the improved fishing optimization algorithm to replace the joint diagonalization operation in the JADE algorithm, obtaining the optimal parameters of the JADE algorithm, and forming a JADE algorithm based on the improved fishing optimization; Step S3: using the JADE algorithm based on improved fishing optimization to perform blind source separation on the pre-processed voiceprint signal; In step S2, the improved fishing optimization algorithm is used to replace the joint diagonalization operation in the JADE algorithm. The specific process of obtaining the optimal parameters of the JADE algorithm is as follows: Step S2.1: setting the initial population and maximum number of iterations of the improved fishing optimization algorithm; Step S2.2: Initialize the improved fishing optimization algorithm, use the parameters of the JADE algorithm as the initial fisher population of the improved fishing optimization algorithm, and set the parameters of each individual fisherman in the initial fisher population; Step S2.3: Evaluate the fitness of each individual fisherman and calculate the fitness function value of the fisher population position; Step S2.4: simulate the search phase of the improved fishing optimization algorithm and introduce an adaptive factor to improve the independent search behavior; The search phase of the improved fishing optimization algorithm includes independent search behavior and group capture behavior; Choose independent search behavior or group capture behavior according to the capture rate α; When p<α, individual fishermen choose independent search behavior, and when p≥α, individual fishermen choose group capture behavior, where p is a random number between (0,1); Step S2.5: simulate the development phase of the improved fishing optimization algorithm and introduce the escape formula for improvement; Step S2.6: Determine whether the current number of iterations reaches the preset maximum number of iterations. If it reaches the preset maximum number of iterations, stop iterating. If it does not reach the preset maximum number of iterations, return to step S2.3 and continue iterating and updating. Step S2.7: Output the optimal individual fisherman in the current fisher population, that is, the optimal parameters of the JADE algorithm.

2. The method for removing transformer background noise based on blind source separation according to claim 1 is characterized in that: The improved fishing optimization algorithm is initialized, and the parameters of the JADE algorithm are used as the initial fisher population of the improved fishing optimization algorithm, and the parameters of each individual fisherman in the initial fisher population are set, which is expressed as: x i,j =rand×(UB j -LB j )+LB j ; In the formula, X represents the fisher population; i,j represents the value of the i-th fisherman in the j-th dimension, i∈1,2,...,n, j∈1,2,...,d; n represents the size of the fisherman population, d represents the dimension of the problem; rand represents a random number in the range [0,1]; LB j UB j They represent the upper and lower bounds of the j-dimensional search space respectively; each fisherman individual represents a set of parameters of the JADE algorithm.

3. The method for removing transformer background noise based on blind source separation according to claim 2 is characterized in that: Calculate the fitness function value Fitness of the fisher population location: In the formula, K(x i,j ) represents the fourth-order cumulative value of the voiceprint signal.

4. The method for removing transformer background noise based on blind source separation according to claim 3 is characterized in that: The capture rate α is expressed as: Where α is the capture rate; t is the current number of iterations; and T represents the maximum number of iterations.

5. The method for removing transformer background noise based on blind source separation according to claim 4 is characterized in that: The adaptive factor is introduced to improve the independent search behavior, which is expressed as: In the formula, fit i represents the fitness value of the i-th fisherman individual; fit pos represents the fitness value of any individual fisherman in the fisher population; fit min and fit max Represent the worst fitness value and the best fitness respectively; R represents the exploration range; represents the position of a random fisherman individual after t+1 iterations in the jth dimension; and represents the position of the i-th fisherman individual after t and t+1 iterations in the j-th dimension; r s represents a random number between [0, 1]; Dis represents the Euclidean distance between the i-th fisherman individual and the reference object; s represents a random unit vector.

6. The method for removing transformer background noise based on blind source separation according to claim 5 is characterized in that: The group captures the behavior, expressed as: In the formula, Centre c It indicates the target point surrounded by group c, which is composed of 3-4 fisher individuals in the fisher population; represents the average value of individual fishermen in group c; and They represent the position of the i-th fisherman in group c after the t and t+1 iterations in the j dimension respectively; r2 represents the speed at which the fisherman approaches the center, with a value range of (0,1); r3 is the offset of the movement, with a value range of (-1,1).

7. The method for removing transformer background noise based on blind source separation according to claim 6 is characterized in that: The development phase of the simulation-improved fishing optimization algorithm is presented, and the escape formula is introduced for improvement, which is expressed as: In the formula, σ represents a constant that changes with the number of iterations; GD represents the Gaussian distribution function; X best represents the position of the global optimal solution; mean(X) represents the matrix of the mean values ​​of each dimension of the fisher population center; r4 represents a random number with a value of {1,2,3}; represents the fisherman individual closest to the i-th fisherman individual; r5 represents a random number between 0 and 1.

8. The method for removing transformer background noise based on blind source separation according to claim 1 is characterized in that: The voiceprint signal of the transformer is collected and preprocessed by the RLS algorithm, which is expressed as: w(k)=w(k-1)+K k ζ k ; ζ k =d(k)-u T′ (k)in(k-1); Where w(k) is the filter coefficient vector at time k; ζ k is the expected difference at time k; d(k) is the expected response vector at time k; u(k) is the input vector at time k; K k is the gain vector at time k; P k is the inverse matrix of the autocorrelation matrix of the input voiceprint signal; w(k-1) represents the filter coefficient vector at time k-1; u T′ (k) represents the transpose of the input vector at time k; P k-1 It represents the inverse matrix of the autocorrelation matrix of the input voiceprint signal at time k-1; λ is the forgetting factor.

9. The method for removing transformer background noise based on blind source separation according to claim 8 is characterized in that: The specific process of step S3 is: Step S3.1: input the preprocessed voiceprint signal into the JADE algorithm based on improved fishing optimization, and in the JADE algorithm based on improved fishing optimization, the preprocessed voiceprint signal is combined with an unknown mixing matrix to obtain an observation signal; Step S3.2: performing spheroidization preprocessing on the observed signal through the spheroidization matrix to obtain the fourth-order cumulative amount of the spheroidization process of the observed signal; Step S3.3: Perform a joint diagonalized orthogonal matrix estimation on the fourth-order cumulants of the spheroidization process of the observed signal, and finally calculate the estimated value to achieve blind source separation.

Images