Converter valve partial discharge voiceprint signal extraction method and system

Abstract

The invention discloses a converter valve partial discharge voiceprint signal extraction method and system, and the method comprises the steps: collecting a mixed voiceprint signal of a converter valve, carrying out the noise reduction of the mixed voiceprint signal, obtaining a noise reduction voiceprint signal, extracting an initial partial discharge voiceprint signal from the noise reduction voiceprint signal, carrying out the sparse representation of the initial partial discharge voiceprint signal, and obtaining a sparse representation result. A single source point is identified according to a sparse representation result, a hybrid matrix is estimated based on the single source point, a hybrid voiceprint signal is reconstructed by using the hybrid matrix, a final partial discharge voiceprint signal is obtained, and noise reduction processing can effectively improve the quality of the voiceprint signal and ensure the accuracy of subsequent extraction. A series of processing such as initial partial discharge voiceprint signal extraction, sparse representation, single source point recognition and hybrid matrix estimation can effectively prevent partial discharge voiceprint signals from being submerged in noise signals, so that the partial discharge voiceprint signals are extracted quickly and accurately.

Description

Technical Field

[0001] This invention relates to the field of partial discharge monitoring technology, and in particular to a method and system for extracting partial discharge acoustic signature signals from converter valves. Background Technology

[0002] Currently, partial discharge acoustic signature monitoring technology for large power equipment is mainly applied to equipment such as transformers and GIS (Geographic Information System), and rarely used for partial discharge monitoring of converter equipment. The converter valve hall has a complex noise environment, including continuous broadband noise from reactors and fans, as well as transient noise such as voices, footsteps, and switchgear noise during personnel inspections. These sounds overlap with the partial discharge sounds of the converter valve in the frequency domain, easily leading to misjudgments. Furthermore, the reflection and overlap of acoustic signature signals in the enclosed space are significant, making it easier for discharge sounds to be submerged in the complex noise environment. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and system for extracting partial discharge acoustic signals of converter valves, which can extract partial discharge acoustic signals quickly and accurately.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for extracting acoustic signature signals from partial discharge of a converter valve includes the following steps: Acquire mixed acoustic signals from the converter valve; The hybrid voiceprint signal is subjected to noise reduction processing to obtain a noise-reduced voiceprint signal; Extract the initial partial discharge acoustic waveform signal from the noise reduction waveform signal; The initial partial discharge acoustic signature signal is sparsely represented to obtain the sparse representation result of the initial partial discharge acoustic signature signal, and a single source point is identified based on the sparse representation result; Estimate the mixing matrix based on the single source point; The hybrid acoustic signature signal is reconstructed using the hybrid matrix to obtain the final partial discharge acoustic signature signal.

[0005] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A converter valve partial discharge acoustic fingerprint signal extraction system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the converter valve partial discharge acoustic fingerprint signal extraction method described above.

[0006] The beneficial effects of this invention are as follows: It acquires the mixed acoustic signature signal of the converter valve, performs noise reduction processing on the mixed acoustic signature signal to obtain a noise-reduced acoustic signature signal, extracts the initial partial discharge acoustic signature signal from the noise-reduced acoustic signature signal, performs sparse representation on the initial partial discharge acoustic signature signal to obtain the sparse representation result, identifies single source points based on the sparse representation result, estimates the mixing matrix based on the single source points, and reconstructs the mixed acoustic signature signal using the mixing matrix to obtain the final partial discharge acoustic signature signal. The noise reduction processing can effectively improve the quality of the acoustic signature signal and ensure the accuracy of subsequent extraction. The series of processes, including initial partial discharge acoustic signature signal extraction, sparse representation, single source point identification, and mixing matrix estimation, can effectively prevent the partial discharge acoustic signature signal from being submerged in noise signals, thereby enabling fast and accurate extraction of the partial discharge acoustic signature signal. Attached Figure Description

[0007] Figure 1 This is a flowchart of a method for extracting partial discharge acoustic signatures from a converter valve according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a converter valve partial discharge acoustic signature extraction system according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the ambient noise in the converter valve hall during a method for extracting partial discharge acoustic signatures of a converter valve according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the mixed sound of a converter valve after adding simulated discharge sound in a method for extracting partial discharge acoustic patterns of a converter valve according to an embodiment of the present invention. Detailed Implementation

[0008] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0009] Before detailing the embodiments of this application, some related concepts will first be explained: Single-source point: A time-frequency point dominated by a single source; Fast Fourier Transform (FFT): One of the most fundamental methods in time-domain to frequency-domain transform analysis.

[0010] In existing technologies, the converter valve hall presents a complex noise environment, including continuous broadband noise from reactors and fans, as well as transient noises such as voices, footsteps, and switch cabinet noises during personnel inspections. These sounds overlap with the partial discharge sounds of the converter valve in the frequency domain, easily leading to misjudgment. Furthermore, the reflection and overlap of acoustic signals in the enclosed space are significant, making it easier for discharge sounds to be submerged in the complex noise environment.

[0011] To at least solve the above problems, please refer to Figure 1 This invention provides a method for extracting partial discharge acoustic signature signals from converter valves, comprising the following steps: Acquire mixed acoustic signals from the converter valve; The hybrid voiceprint signal is subjected to noise reduction processing to obtain a noise-reduced voiceprint signal; Extract the initial partial discharge acoustic waveform signal from the noise reduction waveform signal; The initial partial discharge acoustic signature signal is sparsely represented to obtain the sparse representation result of the initial partial discharge acoustic signature signal, and a single source point is identified based on the sparse representation result; Estimate the mixing matrix based on the single source point; The hybrid acoustic signature signal is reconstructed using the hybrid matrix to obtain the final partial discharge acoustic signature signal.

[0012] As can be seen from the above description, the beneficial effects of the present invention are as follows: the mixed acoustic fingerprint signal of the converter valve is collected, the mixed acoustic fingerprint signal is denoised to obtain a denoised acoustic fingerprint signal, the initial partial discharge acoustic fingerprint signal is extracted from the denoised acoustic fingerprint signal, the initial partial discharge acoustic fingerprint signal is sparsely represented to obtain the sparse representation result, and single source points are identified according to the sparse representation result. The mixing matrix is ​​estimated based on the single source points, and the mixed acoustic fingerprint signal is reconstructed using the mixing matrix to obtain the final partial discharge acoustic fingerprint signal. The denoising process can effectively improve the quality of the acoustic fingerprint signal and ensure the accuracy of subsequent extraction. The series of processes such as initial partial discharge acoustic fingerprint signal extraction, sparse representation, single source point identification, and mixing matrix estimation can effectively prevent the partial discharge acoustic fingerprint signal from being submerged in the noise signal, thereby quickly and accurately extracting the partial discharge acoustic fingerprint signal.

[0013] Further, the mixed voiceprint signal is subjected to noise reduction processing to obtain a noise-reduced voiceprint signal, including: The noise-reduced ripple signal is windowed and framed to obtain a frame sequence; Perform a Fast Fourier Transform on the frame sequence to obtain the frequency domain signal; Calculate the average amplitude spectrum of the frequency domain signal; Multiple frames of noise signals are selected from the frequency domain signal, and the average power spectrum of the noise signals is calculated. Calculate the adaptive gain factor based on the average power spectrum; Calculate the amplitude spectrum after spectral subtraction based on the adaptive gain factor and the average power spectrum; Based on the amplitude spectrum after spectral subtraction, an inverse fast Fourier transform is performed to obtain a noise-reduced ripple signal.

[0014] Further, the average amplitude spectrum of the frequency domain signal is calculated, specifically as follows: ; In the formula, Represents the frequency domain signal. i Frame in k The average amplitude spectrum at each frequency point Z Indicates the number of frames before and after. Indicates the first i + j Frame in k The original amplitude spectrum at each frequency point; The average power spectrum of the noise signal is calculated as follows: ; In the formula, Indicates the first k The noise average power spectrum at each frequency point N The number of frames representing the noise signal. Indicates the first i Frame noise signal in the first k Power spectrum at each frequency point; The adaptive gain factor is calculated based on the average power spectrum, specifically as follows: ; ; In the formula, Indicates the adaptive gain factor. Indicates the frequency adaptive over-subtraction factor. Indicates the gain compensation factor. K This represents the total number of frequency points; The amplitude spectrum after spectral subtraction is calculated based on the adaptive gain factor and the average power spectrum, specifically as follows: ; In the formula, Indicates the first i The amplitude spectrum of the frame after spectral subtraction; Based on the amplitude spectrum after spectral subtraction, an inverse fast Fourier transform is performed to obtain the noise-reduced ripple signal, specifically: ; In the formula, Indicates the first i Frame noise reduction ripple signal, z Indicates the index of the sample point within the frame. () represents the inverse fast Fourier transform. Indicates the first i Frame number k Phase at each frequency point j It represents the imaginary unit.

[0015] As described above, noise can be accurately estimated by the average power spectrum of the noise signal. Combined with the adaptive gain factor, frequency adaptive spectral reduction is achieved, which effectively suppresses background noise and preserves the core features of the target acoustic pattern. Windowing and framing, as well as the calculation of the average amplitude spectrum, can reduce signal fluctuations, make the frequency domain features more stable, and reduce the interference of noise on subsequent analysis.

[0016] Further, extracting the initial partial discharge acoustic waveform signal from the noise reduction waveform signal includes: The noise-reduced ripple signal is processed by frame segmentation, and the energy of each frame is calculated; Calculate the improved energy based on the stated energy; Perform a Fast Fourier Transform on the noise-reduced ripple signal to obtain the Fourier Transform result, and calculate the power spectrum of the noise-reduced ripple signal based on the Fourier Transform result; The power spectrum is normalized to obtain the normalized power spectrum; The spectral entropy is calculated based on the normalized power spectrum. Calculate the energy entropy ratio based on the spectral entropy and the improved energy; If the energy entropy ratio is greater than the first preset threshold, then the voiceprint signal corresponding to the energy entropy ratio is determined to be a voice segment; All the audio segments are combined to obtain the initial partial discharge acoustic waveform signal.

[0017] Furthermore, the energy of each frame is calculated, specifically as follows: ; In the formula, Indicates the first i Frame energy, Indicates the first i The first frame m One sample point, M Indicates the number of sample points; The improved energy is calculated based on the aforementioned energy, specifically as follows: ; In the formula, Indicates the first i Improved frame energy a This represents a constant that controls logarithmic scaling; The spectral entropy is calculated based on the normalized power spectrum, specifically as follows: ; In the formula, Indicates the first i Spectral entropy of a frame Indicates the first i The normalized power spectrum of the frame; The energy-entropy ratio is calculated based on the spectral entropy and the improved energy, specifically as follows: ; In the formula, Indicates the first i The energy entropy ratio of the frames, This represents a constant that prevents division by zero.

[0018] As described above, energy reflects signal strength, spectral entropy reflects frequency distribution characteristics, and the energy-entropy ratio combines the advantages of both, making it highly targeted for the identification of partial discharge signals. It effectively distinguishes partial discharge sound from background noise, accurately extracts sound segments, avoids omission or misjudgment of partial discharge signals, and reduces signal fluctuations through framing and normalization operations, making feature extraction more stable and adaptable to partial discharge soundprint processing under different noise environments.

[0019] Furthermore, identifying single-source points based on the sparse representation results includes: The single-source determination score at time-frequency points is calculated based on the sparse representation results. If the single-source determination score is greater than the second preset threshold, and the amplitude value of the sparse representation result is greater than the third preset threshold, then the time-frequency point is determined to be a single-source point.

[0020] As described above, the single-source determination score quantifies the degree of single-source dominance, and the amplitude threshold ensures the effective signal strength. By combining the single-source determination score and the amplitude value as dual thresholds, the single-source dominance at the time and frequency points is guaranteed, while weak signal interference is eliminated, reducing misjudgments and missed judgments.

[0021] Further, estimating the mixing matrix based on the single source point includes: Each single source point is considered as a cluster, and the distance between clusters is calculated; Repeatedly merge the clusters with the smallest inter-cluster distance until all single source points are merged into one cluster or the current number of clusters reaches the preset number of clusters, and the clustering is completed; The center direction of each cluster after clustering is used as a column of the mixing matrix.

[0022] As described above, by clustering single source points and fitting the column vector of the mixing matrix using the cluster center direction, the linear mixing relationship between the source signal and the observed signal can be accurately captured, which is suitable for multi-source mixing scenarios. The clustering process can reduce the noise interference of single source points by merging the distance between clusters, making the mixing matrix estimation more stable.

[0023] Further, reconstructing the hybrid acoustic signature signal using the hybrid matrix to obtain the final partial discharge acoustic signature signal includes: While ensuring the sparsity and temporal continuity of the final partial discharge acoustic signature signal, an objective function is established based on the final partial discharge acoustic signature signal with the goal of minimizing the reconstruction error. Based on the initial partial discharge acoustic waveform signal and the mixing matrix, establish the constraint conditions corresponding to the objective function; The objective function is solved based on the constraints to obtain the final partial discharge acoustic signature signal.

[0024] Furthermore, while ensuring the sparsity and temporal continuity of the final partial discharge acoustic signature signal, an objective function is established based on the final partial discharge acoustic signature signal with the goal of minimizing the reconstruction error. Specifically: ; In the formula, This indicates that the final broadcast of the voiceprint signal is complete. T Indicates the total length of the time series. t Indicates a time index. express L 1-norm, express L 2-norm number terms, Indicates the weighting coefficient; Based on the initial partial discharge acoustic signature signal and the mixing matrix, the constraint conditions corresponding to the objective function are established as follows: ; In the formula, This represents the initial partial discharge acoustic signature signal. This represents a mixture matrix.

[0025] As described above, by combining sparsity and temporal continuity constraints, residual noise and interference components are effectively eliminated, and a high-quality final partial discharge acoustic signature signal is restored. With the goal of minimizing reconstruction error, the constraints of the initial partial discharge acoustic signature signal and the mixing matrix are combined to ensure that the reconstructed signal is highly consistent with the original source signal.

[0026] Please refer to Figure 2 Another embodiment of the present invention provides a converter valve partial discharge acoustic fingerprint signal extraction system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the converter valve partial discharge acoustic fingerprint signal extraction method described above.

[0027] The present invention provides a method and system for extracting partial discharge acoustic signatures from converter valves, applicable to scenarios such as partial discharge detection in high-voltage power transmission and transformation equipment and fault diagnosis of electrical equipment. The specific implementation methods described below illustrate this approach. Please refer to Figure 1 One embodiment of the present invention is as follows: A method for extracting acoustic signature signals from partial discharge of a converter valve includes the following steps: S1. Acquire the mixed acoustic signature signal of the converter valve.

[0028] In one alternative implementation, during the research phase, in cases where fault samples (i.e., partial discharge sounds) are lacking, simulated discharge sounds can be added to the ambient noise of the valve hall. For example... Figure 3As shown, real noise was collected from inside the valve hall. Due to the lack of actual discharge fault samples, simulated discharge fault sounds were merged into the background noise to construct a hybrid sample, such as... Figure 4 As shown.

[0029] S2. Perform noise reduction processing on the hybrid voiceprint signal to obtain a noise-reduced voiceprint signal, specifically including S21-S27: S21. Window and frame the noise-reduced ripple signal to obtain a frame sequence. .

[0030] S22. Perform a Fast Fourier Transform on the frame sequence to obtain the frequency domain signal. .

[0031] S23. Calculate the average amplitude spectrum of the frequency domain signal, specifically as follows: ; In the formula, Represents the frequency domain signal. i Frame in k The average amplitude spectrum at each frequency point Z Indicates the number of frames before and after, used for smoothing. Indicates the first i + j Frame in k The original amplitude spectrum at each frequency point.

[0032] S24. Select multiple frames of noise signals from the frequency domain signal and calculate the average power spectrum of the noise signals, specifically as follows: ; In the formula, Indicates the first k The noise average power spectrum at each frequency point N The number of frames representing the noise signal. Indicates the first i Frame noise signal in the first k The power spectrum at each frequency point.

[0033] S25. Calculate the adaptive gain factor based on the average power spectrum, specifically as follows: ; ; In the formula, Indicates the adaptive gain factor. Indicates the frequency adaptive over-subtraction factor. Indicates the gain compensation factor. K This represents the total number of frequency points.

[0034] S26. Calculate the amplitude spectrum after spectral subtraction based on the adaptive gain factor and the average power spectrum, specifically as follows: ; In the formula, Indicates the first i The amplitude spectrum of the frame after spectral subtraction.

[0035] S27. Perform an inverse fast Fourier transform on the amplitude spectrum after spectral subtraction to obtain the noise-reduced ripple signal, specifically: ; In the formula, Indicates the first i Frame noise reduction ripple signal, z Indicates the index of the sample point within the frame. () represents the inverse fast Fourier transform. Indicates the first i Frame number k Phase at each frequency point j It represents the imaginary unit.

[0036] After merging all frames, the noise-reduced ripple signal is obtained. .

[0037] S3. Extract the initial partial discharge acoustic waveform signal from the noise reduction waveform signal, specifically including S31-S38: S31. Perform frame segmentation processing on the noise reduction ripple signal and calculate the energy of each frame, specifically as follows: ; In the formula, Indicates the first i Frame energy, Indicates the first i The first frame m One sample point, M Indicates the number of sample points.

[0038] S32. Calculate the improved energy based on the stated energy, specifically as follows: ; In the formula, Indicates the first i Improved frame energy a This represents a constant that controls logarithmic scaling.

[0039] S33. Perform a Fast Fourier Transform on the noise-reduced ripple signal to obtain the Fourier Transform result, and calculate the power spectrum of the noise-reduced ripple signal based on the Fourier Transform result.

[0040] Specifically, the power spectrum of the noise-reduced ripple signal is calculated based on the Fourier transform result, as follows: ; In the formula, Indicates the noise reduction ripple signal. i Power spectrum of the frame.

[0041] S34. Normalize the power spectrum to obtain the normalized power spectrum, specifically as follows: .

[0042] S35. Calculate the spectral entropy based on the normalized power spectrum, specifically as follows: ; In the formula, Indicates the first i Spectral entropy of a frame Indicates the first i The normalized power spectrum of the frame.

[0043] S36. Calculate the energy entropy ratio based on the spectral entropy and the improved energy, specifically as follows: ; In the formula, Indicates the first i The energy entropy ratio of the frames, This represents a constant that prevents division by zero. .

[0044] S37. If the energy entropy ratio is greater than the first preset threshold, then the voiceprint signal corresponding to the energy entropy ratio is determined to be a voice segment.

[0045] S38. Combine all the audio segments to obtain the initial partial discharge acoustic signal.

[0046] The starting and ending points are determined by consecutive audio segments, and all audio segments are recombined into a new initial partial discharge acoustic pattern signal. .

[0047] S4. Perform sparse representation on the initial partial discharge acoustic signature signal to obtain the sparse representation result of the initial partial discharge acoustic signature signal, and identify single source points based on the sparse representation result, specifically including S41-S43: S41. Perform sparse representation on the initial partial discharge acoustic signature signal to obtain the sparse representation result of the initial partial discharge acoustic signature signal, specifically as follows: ; In the formula, The sparse representation result of the initial partial discharge acoustic signature signal is shown. Indicates a point in time m The initial partial discharge acoustic signature signal, w () represents the Hamming window function. Represents a complex exponential function. f Indicates frequency.

[0048] S42. Calculate the single-source determination score for time-frequency points based on the sparse representation result, specifically as follows: ; In the formula, Indicates time t ,frequency f The single-source determination score at the time-frequency point is given by Re[], where Re[] represents the operation of taking the real part of the complex number and Im[] represents the operation of taking the imaginary part of the complex number. , This represents the sparse representation results of the initial partial discharge acoustic signature signals acquired by different sensors. This represents a constant that prevents division by zero. .

[0049] S43. If the single-source determination score is greater than the second preset threshold, and the amplitude value of the sparse representation result is... If the value is greater than the third preset threshold, then the time-frequency point is determined to be a single-source point.

[0050] S5. Estimate the mixing matrix based on the single source point, specifically including S51-S53: S51. Treat each single source point as a cluster and calculate the inter-cluster distance, specifically as follows: ; In the formula, Cluster with cluster The distance between them p Cluster The point in the middle, q Cluster The point in the middle.

[0051] S52. Repeatedly merge the clusters with the smallest inter-cluster distance until all single source points are merged into one cluster or the current number of clusters reaches the preset number of clusters, and the clustering is completed.

[0052] The preset cluster number refers to the number of discharge sound sources.

[0053] S53. After clustering, the center direction of each cluster is used as a column of the mixing matrix. .

[0054] S6. Reconstruct the hybrid acoustic signature signal using the hybrid matrix to obtain the final partial discharge acoustic signature signal, specifically including S61-S63: S61. Under the premise of ensuring the sparsity and temporal continuity of the final partial discharge acoustic signature signal, and with the goal of minimizing the reconstruction error, an objective function is established based on the final partial discharge acoustic signature signal, specifically as follows: ; In the formula, This indicates that the final broadcast of the voiceprint signal is complete. T Indicates the total length of the time series. t Indicates a time index. express L 1-norm, express L 2-norm number terms, This represents the weighting coefficient.

[0055] S62. Based on the initial partial discharge acoustic signature signal and the mixing matrix, establish the constraint conditions corresponding to the objective function, specifically as follows: ; In the formula, This represents the initial partial discharge acoustic signature signal. This represents a mixture matrix.

[0056] S63. Solve the objective function based on the constraints to obtain the final partial discharge acoustic signature signal.

[0057] The method described above in this invention has a fast analysis speed and low computational load, which can quickly complete the separation and extraction of discharge signals. At the same time, it can effectively extract the discharge sound signal from complex noise backgrounds and reduce misjudgments.

[0058] In summary, the method for extracting partial discharge acoustic signature signals from a converter valve, as described above, involves acquiring a mixed acoustic signature signal from the converter valve, performing noise reduction processing on the mixed acoustic signature signal to obtain a noise-reduced acoustic signature signal, extracting an initial partial discharge acoustic signature signal from the noise-reduced acoustic signature signal, performing sparse representation on the initial partial discharge acoustic signature signal to obtain a sparse representation result, identifying single source points based on the sparse representation result, estimating the mixing matrix based on the single source points, and reconstructing the mixed acoustic signature signal using the mixing matrix to obtain the final partial discharge acoustic signature signal. The noise reduction processing effectively improves the acoustic signature signal quality and ensures the accuracy of subsequent extraction. The series of processes, including initial partial discharge acoustic signature signal extraction, sparse representation, single source point identification, and mixing matrix estimation, effectively prevent the partial discharge acoustic signature signal from being submerged in noise signals, thereby enabling fast and accurate extraction of the partial discharge acoustic signature signal. Furthermore, by combining sparsity and temporal continuity constraints, residual noise and interference components are effectively removed, restoring a high-quality final partial discharge acoustic signature signal. With the goal of minimizing reconstruction error, the constraints of the initial partial discharge acoustic signature signal and the mixing matrix ensure a high degree of consistency between the reconstructed signal and the original source signal.

[0059] According to another aspect of the invention, Figure 2This is a schematic diagram illustrating a converter valve partial discharge acoustic signature extraction system according to an embodiment of the present invention. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the converter valve partial discharge acoustic signature extraction method described above.

[0060] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for extracting acoustic signature signals from partial discharge of a converter valve, characterized in that, Including the following steps: Acquire mixed acoustic signals from the converter valve; The hybrid voiceprint signal is subjected to noise reduction processing to obtain a noise-reduced voiceprint signal; Extract the initial partial discharge acoustic waveform signal from the noise reduction waveform signal; The initial partial discharge acoustic signature signal is sparsely represented to obtain the sparse representation result of the initial partial discharge acoustic signature signal, and a single source point is identified based on the sparse representation result; Estimate the mixing matrix based on the single source point; The hybrid acoustic signature signal is reconstructed using the hybrid matrix to obtain the final partial discharge acoustic signature signal.

2. The method for extracting partial discharge acoustic signature signals from a converter valve according to claim 1, characterized in that, The mixed voiceprint signal is denoised to obtain a denoised voiceprint signal, which includes: The noise-reduced ripple signal is windowed and framed to obtain a frame sequence; Perform a Fast Fourier Transform on the frame sequence to obtain the frequency domain signal; Calculate the average amplitude spectrum of the frequency domain signal; Multiple frames of noise signals are selected from the frequency domain signal, and the average power spectrum of the noise signals is calculated. Calculate the adaptive gain factor based on the average power spectrum; Calculate the amplitude spectrum after spectral subtraction based on the adaptive gain factor and the average power spectrum; An inverse fast Fourier transform is performed on the amplitude spectrum after spectral subtraction to obtain a noise-reduced ripple signal.

3. The method for extracting partial discharge acoustic signature signals from a converter valve according to claim 2, characterized in that, The average amplitude spectrum of the frequency domain signal is calculated as follows: ； In the formula, Represents the frequency domain signal. i Frame in k The average amplitude spectrum at each frequency point Z Indicates the number of frames before and after. Indicates the first i + j Frame in k The original amplitude spectrum at each frequency point; The average power spectrum of the noise signal is calculated as follows: ； In the formula, Indicates the first k The noise average power spectrum at each frequency point N Indicates the number of frames in the noise signal. Indicates the first i Frame noise signal in the first k Power spectrum at each frequency point; The adaptive gain factor is calculated based on the average power spectrum, specifically as follows: ； ； In the formula, Indicates the adaptive gain factor. Indicates the frequency adaptive over-subtraction factor. Indicates the gain compensation factor. K This represents the total number of frequency points; The amplitude spectrum after spectral subtraction is calculated based on the adaptive gain factor and the average power spectrum, specifically as follows: ； In the formula, Indicates the first i The amplitude spectrum of the frame after spectral subtraction; Based on the amplitude spectrum after spectral subtraction, an inverse fast Fourier transform is performed to obtain the noise-reduced ripple signal, specifically: ； In the formula, Indicates the first i Frame noise reduction ripple signal, z Indicates the index of the sample point within the frame. () represents the inverse fast Fourier transform. Indicates the first i Frame number k Phase at each frequency point j It represents the imaginary unit.

4. The method for extracting partial discharge acoustic signature signals from a converter valve according to claim 1, characterized in that, Extracting the initial partial discharge acoustic waveform signal from the noise reduction waveform signal includes: The noise-reduced ripple signal is processed by frame segmentation, and the energy of each frame is calculated; Calculate the improved energy based on the stated energy; Perform a Fast Fourier Transform on the noise-reduced ripple signal to obtain the Fourier Transform result, and calculate the power spectrum of the noise-reduced ripple signal based on the Fourier Transform result; The power spectrum is normalized to obtain the normalized power spectrum; The spectral entropy is calculated based on the normalized power spectrum. Calculate the energy entropy ratio based on the spectral entropy and the improved energy; If the energy entropy ratio is greater than the first preset threshold, then the voiceprint signal corresponding to the energy entropy ratio is determined to be a voice segment; All the audio segments are combined to obtain the initial partial discharge acoustic waveform signal.

5. The method for extracting partial discharge acoustic signature signals from a converter valve according to claim 4, characterized in that, The energy of each frame is calculated as follows: ； In the formula, Indicates the first i Frame energy, Indicates the first i The first frame m One sample point, M Indicates the number of sample points; The improved energy is calculated based on the aforementioned energy, specifically as follows: ； In the formula, Indicates the first i Improved frame energy a This represents a constant that controls logarithmic scaling; The spectral entropy is calculated based on the normalized power spectrum, specifically as follows: ； In the formula, Indicates the first i Spectral entropy of a frame Indicates the first i The normalized power spectrum of the frame; The energy-entropy ratio is calculated based on the spectral entropy and the improved energy, specifically as follows: ； In the formula, Indicates the first i The energy entropy ratio of the frames, This represents a constant that prevents division by zero.

6. The method for extracting partial discharge acoustic signature signals from a converter valve according to claim 1, characterized in that, Identifying single-source points based on the sparse representation results includes: The single-source determination score at time-frequency points is calculated based on the sparse representation results. If the single-source determination score is greater than the second preset threshold, and the amplitude value of the sparse representation result is greater than the third preset threshold, then the time-frequency point is determined to be a single-source point.

7. The method for extracting partial discharge acoustic signature signals from a converter valve according to claim 1, characterized in that, The mixture matrix estimated based on the single-source point includes: Each single source point is considered as a cluster, and the distance between clusters is calculated; Repeatedly merge the clusters with the smallest inter-cluster distance until all single source points are merged into one cluster or the current number of clusters reaches the preset number of clusters, and the clustering is completed; The center direction of each cluster after clustering is used as a column of the mixing matrix.

8. The method for extracting partial discharge acoustic signature signals from a converter valve according to claim 1, characterized in that, The hybrid acoustic signature signal is reconstructed using the hybrid matrix to obtain the final partial discharge acoustic signature signal, including: While ensuring the sparsity and temporal continuity of the final partial discharge acoustic signature signal, an objective function is established based on the final partial discharge acoustic signature signal with the goal of minimizing the reconstruction error. Based on the initial partial discharge acoustic waveform signal and the mixing matrix, establish the constraint conditions corresponding to the objective function; The objective function is solved based on the constraints to obtain the final partial discharge acoustic signature signal.

9. The method for extracting partial discharge acoustic signature signals from a converter valve according to claim 8, characterized in that, To ensure the sparsity and temporal continuity of the final partial discharge acoustic signature signal, and with the goal of minimizing the reconstruction error, an objective function is established based on the final partial discharge acoustic signature signal, as follows: ； In the formula, This indicates that the final broadcast of the voiceprint signal is complete. T Indicates the total length of the time series. t Indicates a time index. express L 1-norm, express L 2-norm number terms, Indicates the weighting coefficient; Based on the initial partial discharge acoustic signature signal and the mixing matrix, the constraint conditions corresponding to the objective function are established as follows: ； In the formula, This represents the initial partial discharge acoustic signature signal. This represents a mixture matrix.

10. A converter valve partial discharge acoustic signature extraction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step of the method for extracting partial discharge acoustic fingerprint signals of a converter valve according to any one of claims 1 to 9.

Images