Circuit breaker mechanical state identification method based on improved feature optimization algorithm

Abstract

The invention discloses a circuit breaker mechanical state identification method based on an improved feature optimization algorithm, and relates to the field of noise signal optimization, and the method comprises the steps: S1, collecting circuit breaker closing sound signals in different mechanical states, and obtaining a plurality of signal samples; s2, extracting a Mel-frequency cepstrum coefficient MFCC and a gamma filter cepstrum coefficient GFCC of the plurality of signal samples by using a similar auditory effect filter bank, and mixing the MFCC and the GFCC into a combined cepstrum coefficient to obtain a preliminarily extracted circuit breaker closing sound signal feature vector p; s3, carrying out weighted optimization on the circuit breaker closing sound signal feature vector p by adopting LDA to obtain an optimized combined cepstrum coefficient pl, and carrying out dimensionality reduction on the optimized combined cepstrum coefficient by adopting a KPCA (Kernel Principal Component Analysis) algorithm; and S4, identifying the identification feature vector to complete the identification of the mechanical state of the circuit breaker.

Description

Technical Field

[0001] The invention relates to the field of noise signal optimization, and in particular to a circuit breaker mechanical state identification method based on an improved feature optimization algorithm. Background Art

[0002] The amount of sound signal data collected when a circuit breaker is closed is large. If the unprocessed sound signal is directly used for mechanical state recognition, the signal redundancy will be large, and unnecessary calculations will reduce the response speed of online recognition; and because the closing sound signals of some mechanical states are relatively close, it is difficult to directly complete the mechanical state judgment of the high-voltage circuit breaker based on the unprocessed sound signal. In order to reduce the calculation amount of the classification algorithm and improve the accuracy of the discrimination, it is very important to extract appropriate features from the circuit breaker closing sound signal.

[0003] The Mel Frequency Cepstrum Coefficient (MFCC) and the Gammatone Frequency Cepstrum Coefficient (GFCC) are based on the human auditory system, do not rely on the nature of sound, and do not make assumptions or restrictions on sound signals. They can be used to realize the feature extraction of circuit breaker closing sound signals.

[0004] Studies have shown that the contribution of each dimension of MFCC and GFCC data to the recognition effect is different. In order to highlight the classification information, it is necessary to weight and optimize the data of each dimension. In addition, the joint cepstral coefficient data has a high dimension, which is not conducive to subsequent pattern recognition.

[0005] Therefore, there is an urgent need for a circuit breaker mechanical state recognition method based on an improved feature optimization algorithm to highlight and reduce the dimension of the classification information. Summary of the invention

[0006] In order to solve the above problems, the present application provides a circuit breaker mechanical state recognition method based on an improved feature optimization algorithm, using the circuit breaker closing sound signal feature vector Weighted optimization is performed to obtain the optimized joint cepstral coefficients, and the KPCA algorithm is used to reduce the dimension of the optimized joint cepstral coefficients and highlight the classification information, which is beneficial to subsequent pattern recognition.

[0007] The circuit breaker mechanical state identification method based on the improved feature optimization algorithm provided in this application includes the following steps: S1. Collecting circuit breaker closing sound signals under different mechanical states to obtain a number of signal samples; S2. Using the auditory effect filter bank, extract the Mel frequency cepstral coefficients MFCC and gamma filter cepstral coefficients GFCC of several signal samples, and mix the MFCC and GFCC into joint cepstral coefficients to obtain the initial extracted circuit breaker closing sound signal feature vector, i.e., the initial feature vector ; S3, characteristic vector of circuit breaker closing sound signal Optimize to obtain the recognition feature vector; S4. Identify the identification feature vector to complete the identification of the mechanical state of the circuit breaker.

[0008] Preferably, the method of collecting signal samples under different mechanical states in S1 is: Repeat the closing operation for a preset number of times under the four mechanical conditions: the operating mechanism is normal, the transmission guide rod is stuck, the gap between the iron core and the closing spring is too large, and the opening spring is loose.

[0009] Preferably, the feature vector The expression is as follows: ; in: M 1, ..., M i is the dimensional data of MFCC; G 1 , …, G j is the dimensional data of GFCC, is the dimension of MFCC, is the dimension of GFCC.

[0010] Preferably, the circuit breaker closing sound signal feature vector The specific content of the optimized identification feature vector is: The characteristic vector of the circuit breaker closing sound signal Perform weighted optimization to obtain the optimized joint cepstral coefficients p l ; The KPCA algorithm is used to reduce the dimension of the optimized joint cepstral coefficients.

[0011] Preferably, the circuit breaker closing sound signal feature vector Perform weighted optimization to obtain the optimized joint cepstral coefficients p l The specific content is: The initial feature vector is analyzed by using LDA inter-class divergence and intra-class divergence. The contribution of each dimension of data to identification is judged; It is assumed that there are c types of sound signals in a number of signal samples, and there are c types of feature vectors; Initial eigenvector No. The intra-class divergence S of dimensional data wj as follows: ; In the formula, For the Initial feature vector of closing sound signal in mechanical state Middle Dimensional data, For its traversal, For the Initial feature vector of closing sound signal in mechanical state Middle Dimensional data average; No. Inter-class divergence of dimensional data as follows: ; In the formula, For the The number of frames of the closing signal. For all initial eigenvectors Middle Dimensional data average; definition Contribute weights to the features of the dimension data; Correspond all dimension data to this dimension data L j The weighted result of multiplication is obtained to obtain the initial feature vector after LDA optimization, that is, the optimized feature vector .

[0012] Preferably, the specific content of using the KPCA algorithm to reduce the dimension of the optimized joint cepstral coefficients is: The eigenvector will be optimized Mapping to high-dimensional feature space through kernel function; The PCA algorithm is used to reduce the dimension and obtain the identification feature vector; The kernel function is expressed as , For the matrix In the mapping matrix of high-dimensional feature space, if for The i-th eigenvector of ; Will use Each row vector of is linearly represented to obtain a linear equation, and the linear equation expression is as follows: ; In the formula represent No. column vector, substituting the linear equation into And using the kernel matrix Substituting: ; The kernel matrix expression is as follows: ; The Gaussian kernel function is used as the mapping kernel function, and the expression of the mapping kernel function is as follows: ; Where: , For the , joint cepstral coefficients; is the variance.

[0013] W eigenvectors The matrix composed of is transposed to obtain the dimension reduction matrix G of the KPCA algorithm. For the new test sample , the dimensionality reduction results are as follows: ; Record Joint Cepstral Coefficients After dimensionality reduction, , which is the feature vector for subsequent pattern recognition.

[0014] In summary, the present invention is a circuit breaker mechanical state recognition method based on an improved feature optimization algorithm. Compared with the traditional technology, the present invention uses a quasi-auditory effect filter bank to extract the Mel frequency cepstral coefficients MFCC and the gamma filter cepstral coefficients GFCC of a number of signal samples, and mixes the MFCC and GFCC into a joint cepstral coefficient to obtain a preliminary extracted circuit breaker closing sound signal feature vector, that is, an initial feature vector Then the circuit breaker closing sound signal feature vector Optimization can be performed to reduce the amount of calculation of the classification algorithm while improving the accuracy of discrimination in subsequent recognition modes.

[0015] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a circuit breaker mechanical state identification method based on an improved feature optimization algorithm of the present invention; Figure 2 The combined cepstrum coefficients corresponding to the closing sounds of four mechanical states in the training set of the circuit breaker mechanical state recognition method based on the improved feature optimization algorithm of the present invention; Figure 3 The cepstrum coefficients of the closing sound signal in different mechanical states after LDA optimization of the present invention; Figure 4 It is a trend diagram of the cumulative contribution rate of characteristic values ​​of a circuit breaker mechanical state identification method based on an improved characteristic optimization algorithm of the present invention; Figure 5 This is a visual graph of the three-dimensional data before the feature vector after dimensionality reduction in the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is further described below by means of the accompanying drawings and embodiments. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​described in these embodiments do not limit the scope of the present application.

[0018] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.

[0019] Technologies, systems, and devices known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, systems, and devices should be considered part of the specification.

[0020] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0021] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0022] like Figure 1 As shown, the present invention provides a circuit breaker mechanical state identification method based on an improved feature optimization algorithm, S1, collecting circuit breaker closing sound signals under different mechanical states to obtain a number of signal samples; Preferably, the method of collecting signal samples under different mechanical states in S1 is: The closing operation is repeated for a preset number of times under four mechanical conditions: the operating mechanism is normal, the transmission guide rod is stuck, the clearance between the iron core and the closing spring is too large, and the opening spring is loose. The embodiment repeats the closing operation 120 times.

[0023] S2. Using the auditory effect filter bank, extract the Mel frequency cepstral coefficients MFCC and gamma filter cepstral coefficients GFCC of several signal samples, and mix the MFCC and GFCC into joint cepstral coefficients to obtain the initial extracted circuit breaker closing sound signal feature vector, i.e., the initial feature vector ; Preferably, the feature vector The expression is as follows: ; in: M 1, ..., M i is the dimensional data of MFCC; G 1 , …, G j is the dimensional data of GFCC, is the dimension of MFCC, is the dimension of GFCC.

[0024] In this paper, the number of Mel filter groups and Gammatone filter groups is selected from empirical values. The number of Mel filter groups is 26, the number of Gammatone filter groups is 64, and the order is 4. According to Shannon sampling theorem, the filter bandwidth is set to 22.05kHz, the signal frame length is 1s, the window function is Hamming window, the single cepstral coefficient dimension is the first 31 dimensions, and the joint cepstral coefficient has a total of 62 dimensions. The joint cepstral coefficients corresponding to the closing sound of the four mechanical states in the training set are shown in 2. Figure 2 a is the cepstrum coefficient of the normal closing sound signal. Figure 2 b in the middle is the cepstrum coefficient of the sound signal of the transmission guide rod stuck and closing the switch. Figure 2 c in the middle is the closing sound signal when the gap between the iron core and the closing spring is too large. Figure 2 The d in the middle is the sound signal of the opening spring being loosened and closing the circuit breaker.

[0025] S3, characteristic vector of circuit breaker closing sound signal Optimize to obtain the recognition feature vector; Preferably, the circuit breaker closing sound signal feature vector The specific content of the optimized identification feature vector is: The characteristic vector of the circuit breaker closing sound signal Perform weighted optimization to obtain the optimized joint cepstral coefficients ; Calculate the intra-class divergence of each dimension of the 62 dimensions of the joint cepstral coefficient and the inter-class divergence , and calculate the corresponding L j The value is multiplied with the corresponding dimension data for weighting. After LDA optimization, the corresponding feature vector of the closing sound signal of each mechanical state like Figure 3 As shown, Figure 3 a in the figure is the cepstrum coefficient of the normal closing sound signal after LDA optimization. Figure 3 b in the figure is the cepstrum coefficient of the sound signal of the transmission guide rod stuck and closing after LDA optimization. Figure 3c is the cepstrum coefficient of the closing sound signal when the gap between the iron core and the closing spring is too large after the LDA. Figure 3 d in the figure is the cepstrum coefficient of the sound signal of the opening spring tripping and closing after LDA optimization.

[0026] The KPCA algorithm is used to reduce the dimension of the optimized joint cepstral coefficients.

[0027] according to Figure 3 It can be seen that after LDA optimization, the data of each dimension of the joint cepstral coefficient is more clearly distinguished, which will significantly improve its discrimination ability. p l Perform KPCA dimensionality reduction and arrange the top 20 non-zero eigenvalues ​​of the kernel matrix from large to small. n The sum of the largest eigenvalues ​​is the previous n The cumulative contribution rate of the eigenvalues ​​is shown in the following figure. Figure 4 According to the criterion that the cumulative contribution rate is greater than 85%, the eigenvectors corresponding to the first 7 eigenvalues ​​are selected to form a dimensionality reduction matrix G , the feature vector is reduced from 62 dimensions to 7 dimensions.

[0028] Preferably, the circuit breaker closing sound signal feature vector Perform weighted optimization to obtain the optimized joint cepstral coefficients p l The specific content is: The initial feature vector is analyzed by using LDA inter-class divergence and intra-class divergence. The contribution of each dimension of data to identification is judged; It is assumed that there are c types of sound signals in a number of signal samples, and there are c types of feature vectors; Initial eigenvector No. Intra-class divergence of dimensional data as follows: ; In the formula, For the Initial feature vector of closing sound signal in mechanical state Middle Dimensional data, For its traversal, For the Initial feature vector of closing sound signal in mechanical state Middle Dimensional data average; No. Inter-class divergence of dimensional data as follows: ; In the formula, For the The number of frames of the closing signal. For all initial eigenvectors Middle Dimensional data average; definition Contribute weights to the features of the dimension data; Correspond all dimension data to this dimension data L j The weighted result of multiplication is obtained to obtain the initial feature vector after LDA optimization, that is, the optimized feature vector .

[0029] Intra-class divergence The smaller it is, the more concentrated the data of the same type are in the dimension data; the inter-class scatter The larger the value, the more dispersed the data of different types of samples in this dimension are. and The impact, definition Contribute weight to the feature of this dimension data, and correspond all dimension data to this dimension data L j The weighted multiplication result is the joint cepstral coefficient after LDA optimization. The optimized joint cepstral coefficient is recorded as .

[0030] Preferably, the specific content of using the KPCA algorithm to reduce the dimension of the optimized joint cepstral coefficients is: The eigenvector will be optimized Mapping to high-dimensional feature space through kernel function; The PCA algorithm is used to reduce the dimension and obtain the identification feature vector; The kernel function is expressed as , For the matrix In the mapping matrix of high-dimensional feature space, if for The i-th eigenvector of ; Will use Each row vector of is linearly represented to obtain a linear equation, and the linear equation expression is as follows: ; In the formula represent No. column vector, substituting the linear equation into And using the kernel matrix Substituting: ; The kernel matrix expression is as follows: ; The Gaussian kernel function is used as the mapping kernel function, and the expression of the mapping kernel function is as follows: ; Where: , For the , joint cepstral coefficients; is the variance.

[0031] W eigenvectors The matrix composed of the following is transposed to obtain the dimension reduction matrix of the KPCA algorithm. , for new test samples , the dimensionality reduction results are as follows: ; Let the i-th joint cepstrum coefficient be After dimensionality reduction, , which is the feature vector for subsequent pattern recognition.

[0032] S4. Identify the identification feature vector to complete the identification of the mechanical state of the circuit breaker.

[0033] In order to reflect the linear separability of the feature vector in the feature space after dimensionality reduction, the feature data of the first three dimensions are selected for visualization. The results are as follows: Figure 5 , it can be seen that the feature vector after optimization and dimensionality reduction has good separability in three-dimensional space, which is conducive to subsequent pattern recognition.

[0034] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A circuit breaker mechanical state identification method based on an improved feature optimization algorithm, characterized in that: The following steps are involved: S1. Collecting circuit breaker closing sound signals under different mechanical states to obtain a number of signal samples; S2. Using the auditory effect filter bank, extract the Mel frequency cepstral coefficients MFCC and gamma filter cepstral coefficients GFCC of several signal samples, and mix the MFCC and GFCC into joint cepstral coefficients to obtain the initial extracted circuit breaker closing sound signal feature vector, i.e., the initial feature vector ; S3, characteristic vector of circuit breaker closing sound signal Optimize to obtain the recognition feature vector; S4. Identify the identification feature vector to complete the identification of the mechanical state of the circuit breaker.

2. A circuit breaker mechanical state identification method based on an improved feature optimization algorithm according to claim 1, characterized in that: The method of collecting signal samples under different mechanical states in S1 is: Repeat the closing operation for a preset number of times under the four mechanical conditions: the operating mechanism is normal, the transmission guide rod is stuck, the gap between the iron core and the closing spring is too large, and the opening spring is loose.

3. A circuit breaker mechanical state identification method based on an improved feature optimization algorithm according to claim 2, characterized in that: Eigenvector The expression is as follows: ； in: M 1, ..., M i is the dimensional data of MFCC; G 1 , …, G j is the dimensional data of GFCC, is the dimension of MFCC, is the dimension of GFCC.

4. A circuit breaker mechanical state identification method based on an improved feature optimization algorithm according to claim 3, characterized in that: Characteristic vector of circuit breaker closing sound signal The specific content of the optimized identification feature vector is: Characteristic vector of circuit breaker closing sound signal Perform weighted optimization to obtain the optimized joint cepstral coefficients p l ; The KPCA algorithm is used to reduce the dimension of the optimized joint cepstral coefficients.

5. A circuit breaker mechanical state identification method based on an improved feature optimization algorithm according to claim 4, characterized in that: Characteristic vector of circuit breaker closing sound signal Perform weighted optimization to obtain the optimized joint cepstral coefficients p l The specific content is: The initial feature vector is analyzed by using LDA inter-class divergence and intra-class divergence. The contribution of each dimension of data to identification is judged; It is assumed that there are c types of sound signals in a number of signal samples, and there are c types of feature vectors; Initial eigenvector No. The intra-class divergence S of dimensional data wj as follows: ； In the formula, is the initial feature vector of the closing sound signal of the i-th mechanical state Middle Dimensional data, For its traversal, For the Initial feature vector of closing sound signal in mechanical state Middle Dimensional data average; No. Inter-class divergence of dimensional data as follows: ； In the formula, For the The number of frames of the closing signal. For all initial eigenvectors Middle Dimensional data average; definition Contribute weights to the features of the dimension data; Correspond all dimension data to this dimension data L j The weighted result of multiplication is obtained to obtain the initial feature vector after LDA optimization, that is, the optimized feature vector .

6. A circuit breaker mechanical state identification method based on an improved feature optimization algorithm according to claim 5, characterized in that: The specific content of using KPCA algorithm to reduce the dimension of optimized joint cepstral coefficients is: The eigenvector will be optimized Mapping to high-dimensional feature space through kernel function; The PCA algorithm is used to reduce the dimension and obtain the identification feature vector; The kernel function is expressed as , For the matrix In the mapping matrix of high-dimensional feature space, if for The i-th eigenvector of ; Will use Each row vector of is linearly represented to obtain a linear equation, and the linear equation expression is as follows: ； In the formula represent No. column vector, substituting the linear equation into And using the kernel matrix Substituting: ； The kernel matrix expression is as follows: ； The Gaussian kernel function is used as the mapping kernel function, and the expression of the mapping kernel function is as follows: ； Where: , For the , joint cepstral coefficients; is the variance. 7.W eigenvectors The matrix composed of is transposed to obtain the dimension reduction matrix G of the KPCA algorithm. For the new test sample , the dimensionality reduction results are as follows: ； No. Joint Cepstral Coefficients After dimensionality reduction, , which is the feature vector for subsequent pattern recognition.

Images