A sound-based capacitor fault detection method and related apparatus

Abstract

The application discloses a sound-based capacitor fault detection method and related device, the method comprising: using a VAD method to perform endpoint detection on a preset sound signal to obtain a noise-reduced sound signal; performing feature extraction on the noise-reduced sound signal based on a mel-frequency cepstral coefficient to obtain a sound feature vector; inputting the sound feature vector into a preset SVM classifier to perform fault detection to obtain a detection result, wherein the detection result comprises abnormal sound and normal sound. The application can alleviate the technical problem that the existing capacitor fault detection is prone to missed detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical equipment, and particularly relates to a capacitor fault detection method based on sound and a related device. BACKGROUND

[0002] Power capacitors are used as reactive power compensation devices in power systems to balance the reactive power demand in the power system. However, long-term discharge of the capacitors can cause a decrease in the insulation performance of the capacitors, and the capacitors can have external or internal partial discharge phenomena, at which time the capacitors are in the premonition of internal insulation collapse, and the capacitors should be removed from the power system in time to avoid greater losses. Therefore, a timely and accurate monitoring model is particularly important.

[0003] Most of the existing capacitor fault online monitoring technologies are to measure external discharge to achieve the detection purpose. Because the internal current cannot be monitored, the fault missed detection can easily occur. SUMMARY

[0004] The present application provides a capacitor fault detection method based on sound and a related device, which is used to alleviate the technical problem that the capacitor fault detection in the prior art is prone to missed detection.

[0005] Therefore, the first aspect of the present application provides a capacitor fault detection method based on sound, comprising:

[0006] performing endpoint detection on the preset sound signal by using a VAD method to obtain a noise-reduced sound signal;

[0007] performing feature extraction on the noise-reduced sound signal based on a mel-frequency cepstral coefficient to obtain a sound feature vector;

[0008] inputting the sound feature vector into a preset SVM classifier for fault detection to obtain a detection result, wherein the detection result comprises abnormal sound and normal sound.

[0009] Preferably, before performing endpoint detection on the preset sound signal by using the VAD method to obtain the noise-reduced sound signal, the method further comprises:

[0010] performing a preprocessing operation on an original sound signal to obtain the preset sound signal, wherein the preprocessing operation comprises noise reduction, framing, and windowing.

[0011] Preferably, before performing endpoint detection on the preset sound signal by using the VAD method to obtain the noise-reduced sound signal, the method further comprises:

[0012] configuring an adaptive threshold for threshold judgment for the VAD method, wherein the adaptive threshold changes according to the running time of the capacitor.

[0013] Preferably, the sound feature vector is input into a preset SVM classifier for fault detection, to obtain a detection result, the detection result including abnormal sound and normal sound, and then including:

[0014] The abnormal sound is screened for fault according to a preset transformer algorithm, to obtain capacitor fault sound.

[0015] The second aspect of the application provides a sound-based capacitor fault detection device, comprising:

[0016] An endpoint detection module is configured to perform endpoint detection on a preset sound signal by using a VAD method, to obtain a noise-reduced sound signal.

[0017] A feature extraction module is configured to perform feature extraction on the noise-reduced sound signal based on mel-frequency cepstral coefficients, to obtain a sound feature vector.

[0018] A fault detection module is configured to input the sound feature vector into a preset SVM classifier for fault detection, to obtain a detection result, the detection result including abnormal sound and normal sound.

[0019] Preferably, the device further comprises:

[0020] A preprocessing module is configured to perform a preprocessing operation on an original sound signal, to obtain a preset sound signal, the preprocessing operation including noise reduction, framing, and windowing.

[0021] Preferably, the device further comprises:

[0022] A parameter configuration module is configured to configure an adaptive threshold for threshold judgment for the VAD method, the adaptive threshold varying according to capacitor running time.

[0023] Preferably, the device further comprises:

[0024] A fault screening module is configured to screen the abnormal sound for fault according to a preset transformer algorithm, to obtain capacitor fault sound.

[0025] The third aspect of the application provides a sound-based capacitor fault detection device, the device comprising a processor and a memory;

[0026] The memory is configured to store program code and transmit the program code to the processor.

[0027] The processor is configured to execute the sound-based capacitor fault detection method according to the instructions in the program code.

[0028] The fourth aspect of the present application provides a computer readable storage medium for storing program codes for executing the sound-based capacitor fault detection method of the first aspect.

[0029] From the above technical solutions, the embodiments of the present application have the following advantages:

[0030] In the present application, a sound-based capacitor fault detection method is provided, which includes: performing endpoint detection on a preset sound signal by using a VAD method to obtain a noise-reduced sound signal; performing feature extraction on the noise-reduced sound signal based on a mel-frequency cepstral coefficient to obtain a sound feature vector; inputting the sound feature vector into a preset SVM classifier for fault detection to obtain a detection result, which includes abnormal sound and normal sound.

[0031] The sound-based capacitor fault detection method provided by the present application is aimed at the situation that the capacitor generates some sound during operation. The sound during the operation of the capacitor is collected for sound analysis, and the SVM classifier is used for fault analysis of the sound to obtain abnormal sound during the operation. The fault detection is directly performed through the running sound instead of the discharge condition, so there is no missed detection of the discharge phenomenon. Therefore, the present application can alleviate the technical problem of easy missed detection in the prior art capacitor fault detection. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 A flowchart of a sound-based capacitor fault detection method provided by an embodiment of the present application is shown in the figure.

[0033] Figure 2 A structural diagram of a sound-based capacitor fault detection device provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0034] In order to enable personnel in the technical field to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0035] For better understanding, please refer to Figure 1 An embodiment of a sound-based capacitor fault detection method provided by the present application includes:

[0036] Step 101: performing endpoint detection on a preset sound signal by using a VAD method to obtain a noise-reduced sound signal.

[0037] The VAD (silence suppression) method is to identify and eliminate long silence periods from the sound signal stream to save the speech path resources without reducing the service quality. The silence suppression can save the valuable bandwidth resources and can be beneficial to reduce the end-to-end delay perceived by the user. The embodiment adopts the VAD method to eliminate the long silence part in the preset sound signal based on the sound frequency and volume to obtain the noise-reduced sound signal.

[0038] Further, the step 101 further includes:

[0039] The pre-processing operation is performed on the original sound signal to obtain the preset sound signal, and the pre-processing operation includes noise reduction, frame division and windowing.

[0040] The pre-processing operation is to improve the quality of the original sound signal to facilitate the subsequent sound feature extraction and detection, so the pre-processing setting can be performed according to the actual situation, and is not limited to the noise reduction, frame division and windowing processing included in the embodiment.

[0041] The frame division is to collect N sampling points of the original sound signal as an observation unit, and the time covered by each frame is set to 25 ms. In order to avoid too abrupt changes between two adjacent frames, the two adjacent frames are overlapped by a preset area, which can be 15 ms. Every 10 ms, a frame is taken. These times are not limited to this, and can be adjusted as needed.

[0042] The windowing is to perform windowing processing on each frame of the divided frame to avoid abrupt changes when the frame signal is subjected to FFT transformation. The embodiment gives a windowing method, i.e. the Hamming method:

[0043] w[n] = 0.54 - 0.46cos(2πN / M)

[0044] Wherein, M is the frame length, which is 400 in the embodiment, N is the number of sampling points, and n is the sampling point.

[0045] Then the noise reduction processing can be further performed. In the embodiment, the spectral subtraction method is used for noise reduction. Before that, each frame is supplemented to 512 bits, and then 512-point FFT transformation is performed to obtain the corresponding frame spectrum. Then the power spectrum of the signal is obtained by taking the modulus square of the frame spectrum. The power spectrum can be plotted into a spectrum graph, and it contains the sound frequencies corresponding to different time points. The spectral subtraction noise reduction process is:

[0046]

[0047] Wherein, ω is the frequency, S(·) is the noise-reduced information, X(·) is the input signal, and D(·) is the noise. The first 5 frames of the window are defined as noise, and the sound signal after that can be obtained by subtracting the average value of the 5 frames.

[0048] Further, step 101 further comprises:

[0049] The adaptive threshold value is configured for the VAD method for threshold judgment, and the adaptive threshold value varies according to the capacitor running time.

[0050] The process of endpoint monitoring by the VAD method can be expressed as:

[0051]

[0052]

[0053] H1:

[0054] H0:

[0055] Wherein, H0 represents the silent background sound, H1 is the capacitor abnormal sound, p(Y|H0) is the silent background sound probability, p(Y|H1) is the capacitor abnormal sound probability, Q is the frequency point, λ Q (t,k), λ X (t,k) respectively represent the background sound variance and the abnormal sound variance, X t,k is the Fourier coefficient amplitude value of the tth frame input signal, the kth frequency point, log∧ t,k represents the log-likelihood ratio of the tth frame signal at the kth frequency point, and δ is the adaptive threshold value. The adaptive threshold value can screen out the silent background sound and the capacitor abnormality, and the continuous updating of the adaptive threshold value can be realized according to the previous frames of sound, so as to improve the accuracy of judgment.

[0056] The specific VAD method implementation process is:

[0057] First, noise classification: the noise classification process uses wavelet time-frequency analysis means to extract features of the signal, and uses a support vector machine to classify the noise environment. The wavelet transform is used to highlight the abnormal sound features, and the time (space) frequency localization analysis is used, the signal (function) is gradually refined by stretching and shifting operation, and finally the time is subdivided at high frequency, the frequency is subdivided at low frequency, and the requirements of automatic adaptation of time-frequency signal analysis are met. The formula is as follows:

[0058] F(t) = s(t) + e(t)

[0059]

[0060] Wherein, s(t) represents the capacitor abnormal sound, e(t) represents the silent background sound, is a wavelet function, F(t) represents the collected substation sound, W e(a, t) represents a wavelet transform process, a, u, e(u) represent scale, translation and translated noise respectively. The subsequent can be used to support vector machine to segment out background noise and abnormal sound, get the maximum interval plane, for detection. Then is according to the historical decision result to the corresponding threshold value is automatically updated; Finally, according to the adaptive threshold value threshold judgment operation, get the noise reduction sound signal.

[0061] Step 102, based on the mel frequency cepstral coefficient, the feature of the noise reduction sound signal is extracted, and the sound feature vector is obtained.

[0062] The mel frequency cepstral coefficient is the coefficient of the mel frequency cepstral, and is the linear transformation of the logarithmic energy spectrum based on the nonlinear mel scale of the sound frequency. The obtained noise reduction sound signal is filtered by a set of mel filter banks in this embodiment. The logarithmic energy of the mel filter bank output signal is obtained, that is, the sound feature vector. Specifically, it can be expressed as:

[0063]

[0064]

[0065] Wherein, X(k) is the signal spectrum distribution information obtained by discrete Fourier transform, E(k) is the energy spectrum obtained by taking modulus square, S(m) is the sound feature vector processed by mel filter; k, n, j are frequency points, sampling points and phase shift respectively, x(n) is the sound information of the nth sampling point, H m (k) represents the conversion coefficient at the kth frequency point.

[0066] Step 103, the sound feature vector is input into the preset SVM classifier for fault detection, and the detection result is obtained. The detection result includes abnormal sound and normal sound.

[0067] The preset SVM classifier is a model that has been trained and can directly perform fault detection task; the training process before this is:

[0068] The obtained historical sound signal is preprocessed and divided into training data set and test data set; the mel frequency cepstral coefficient is used for feature extraction of the training data and input into the initial SVM classifier for model training; the test set is used to test the trained SVM classifier, and the classifier with better test result is reserved as the preset SVM classifier. The preset SVM classifier is a binary classification model, so the detection result is abnormal sound and normal sound.

[0069] In addition, in order to improve the accuracy of the preset SVM classifier, the AUC is used to measure the performance of the classifier, which helps to improve the training efficiency of the classifier, and the AUC related formula can be expressed as:

[0070]

[0071]

[0072] Wherein, AUC is the area under the ROC curve, pAUC is the partial AUC, which is the AUC calculated from a part of the ROC curve in a predetermined target range, p is the specified target range, and the value is 0.2, N - , N + are the number of normal test samples and abnormal test samples, A θ (*) is the positive and negative sample probability, are the normal and abnormal test samples, respectively, |·| is the lower limit function, and H(*) is the hard threshold function, which returns 1 when "*" > 0, otherwise 0.

[0073] Further, step 103 further comprises:

[0074] According to the preset transformer algorithm, the abnormal sound is screened for fault, and the capacitor fault sound is obtained.

[0075] In order to detect more accurate capacitor fault conditions, the transformer algorithm can also be used to screen the abnormal sound for fault, and further determine the capacitor fault sound. Because not all abnormal sounds are caused by capacitor failure, and the capacitor fault sound is not single, the transformer algorithm can be used for more accurate fault screening operation.

[0076] The transformer algorithm in the embodiment adopts an encoding and decoding structure, each encoder contains a self-attention layer and a fully connected feedforward neural network layer; and the decoder contains a masked multi-head self-attention layer, a multi-head self-attention layer and a fully connected feedforward neural network layer. The sound recognition is completed through the cooperation between the encoder and the decoder.

[0077] The sound-based capacitor fault detection method provided by the embodiment of the application is aimed at the situation that the capacitor generates some sound during operation, collects the sound during the operation of the capacitor for sound analysis, and uses the SVM classifier to analyze the fault of the sound to obtain the abnormal sound during the operation. The fault detection is directly performed through the running sound instead of the discharge condition, so there is no case of missing discharge phenomenon. Therefore, the embodiment of the application can alleviate the technical problem of easy missing detection in the prior art for capacitor fault detection.

[0078] For ease of understanding, please refer to Figure 2 The application provides an embodiment of a sound-based capacitor fault detection device, comprising:

[0079] An endpoint detection module 201 is configured to perform endpoint detection on a preset sound signal by using a VAD method to obtain a noise-reduced sound signal.

[0080] A feature extraction module 202 is configured to perform feature extraction on the noise-reduced sound signal based on a mel-frequency cepstral coefficient to obtain a sound feature vector.

[0081] A fault detection module 203 is configured to input the sound feature vector into a preset SVM classifier to perform fault detection and obtain a detection result, wherein the detection result comprises abnormal sound and normal sound.

[0082] Further, the application further comprises:

[0083] A preprocessing module 204 is configured to perform a preprocessing operation on an original sound signal to obtain a preset sound signal, wherein the preprocessing operation comprises noise reduction, framing, and windowing.

[0084] Further, the application further comprises:

[0085] A parameter configuration module 205 is configured to configure an adaptive threshold value for threshold judgment for the VAD method, wherein the adaptive threshold value changes according to a capacitor running time.

[0086] Further, the application further comprises:

[0087] A fault screening module 206 is configured to perform fault screening on the abnormal sound according to a preset transformer algorithm to obtain a capacitor fault sound.

[0088] The application also provides a sound-based capacitor fault detection device, which comprises a processor and a memory.

[0089] The memory is configured to store program code and transmit the program code to the processor.

[0090] The processor is configured to execute the sound-based capacitor fault detection method in the above method embodiment according to instructions in the program code.

[0091] The application also provides a computer-readable storage medium for storing program code, wherein the program code is used to execute the sound-based capacitor fault detection method in the above method embodiment.

[0092] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented by other manners. For example, the apparatus embodiments described above are merely illustrative, for example, the division of the units is merely a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, apparatuses or units, and can be electrical, mechanical or other forms.

[0093] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0094] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0095] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for executing all or part of the steps of the method described in each embodiment of the present application by a computer device (which can be a personal computer, a server, or a network device, etc.). The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (English full name: Read-Only Memory, English abbreviation: ROM), a random access memory (English full name: Random Access Memory, English abbreviation: RAM), a magnetic disk or an optical disk, and various program code storage media.

[0096] The above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A sound-based capacitor fault detection method, characterized by, The method comprises the following steps: The VAD method is used to perform endpoint detection on the preset sound signal to obtain a noise-reduced sound signal, including: configuring an adaptive threshold for threshold judgment for the VAD method, the adaptive threshold changes according to the running time of the capacitor, and is updated according to the first few frames of the sound signal, to filter out the silent background sound and capacitor discharge sound in the sound signal; The VAD method includes feature extraction of the signal by using wavelet time-frequency analysis means, and classification of the noise environment by using a support vector machine, which is represented as: ； wherein, represents a capacitor abnormal sound, represents a silent background sound, is a wavelet function, represents a collected substation sound, represents a wavelet transform process, , , respectively represent a scale, a translation amount, and a translated noise; The noise-reduced sound signal is subjected to feature extraction based on a mel-frequency cepstral coefficient to obtain a sound feature vector; The sound feature vector is input into a preset SVM classifier for fault detection to obtain a detection result, the detection result including abnormal sound and normal sound.

2. The sound-based capacitor fault detection method of claim 1, wherein, The method further comprises the following steps before the VAD method is used to perform endpoint detection on the preset sound signal to obtain a noise-reduced sound signal: The original sound signal is subjected to a preprocessing operation to obtain a preset sound signal, the preprocessing operation including noise reduction, framing and windowing.

3. The sound-based capacitor fault detection method of claim 1, wherein, The VAD method is used to perform endpoint detection on the preset sound signal to obtain a noise-reduced sound signal, wherein: The process of endpoint monitoring by the VAD method can be represented as: ； wherein, represents a silent background sound, represents a capacitor abnormal sound, represents a silent background sound probability, represents a capacitor abnormal sound probability, Q represents a frequency point, , respectively represent a background sound variance and an abnormal sound variance, represents a Fourier coefficient amplitude value of the tth frame input signal at the kth frequency point, represents a log-likelihood ratio of the tth frame signal at the kth frequency point, represents an adaptive threshold value.

4. The sound-based capacitor fault detection method of claim 1, wherein, The sound feature vector is input into a preset SVM classifier for fault detection to obtain a detection result, the detection result including abnormal sound and normal sound, and the method further comprises the following steps: The abnormal sound is subjected to fault screening according to a preset transformer algorithm to obtain capacitor fault sound.

5. A sound-based capacitor fault detection apparatus, characterized by, The method comprises the following steps: An endpoint detection module is configured to use a VAD method to perform endpoint detection on a preset sound signal to obtain a noise-reduced sound signal, including configuring an adaptive threshold for threshold judgment for the VAD method, the adaptive threshold changing according to the running time of the capacitor, and being updated according to the first few frames of the sound signal, to filter out the silent background sound and capacitor discharge sound in the sound signal; The VAD method includes feature extraction of the signal by using wavelet time-frequency analysis means, and classification of the noise environment by using a support vector machine, which is represented as: ； wherein, represents a capacitor rattling sound, represents a silent background sound, is a wavelet function, represents a collected substation sound, represents a wavelet transform process, , , respectively represent a scale, a translation amount, and a translated noise. A feature extraction module is configured to perform feature extraction on the noise-reduced sound signal based on a mel-frequency cepstral coefficient to obtain a sound feature vector; A fault detection module is configured to input the sound feature vector into a preset SVM classifier for fault detection to obtain a detection result, the detection result including abnormal sound and normal sound.

6. The sound-based capacitor fault detection device of claim 5, wherein, The method further comprises the following steps: A preprocessing module is configured to perform a preprocessing operation on an original sound signal to obtain a preset sound signal, the preprocessing operation including noise reduction, framing and windowing.

7. The sound-based capacitor fault detection device of claim 5, wherein, The method further comprises the following steps: In the endpoint detection module, the process of endpoint monitoring by the VAD method can be represented as: ； wherein, represents a silent background sound, represents a capacitor abnormal sound, represents a silent background sound probability, represents a capacitor abnormal sound probability, Q represents a frequency point, , respectively represent a background sound variance and an abnormal sound variance, represents a Fourier coefficient amplitude value of the tth frame input signal at the kth frequency point, represents a log-likelihood ratio of the tth frame signal at the kth frequency point, represents an adaptive threshold value.

8. The sound-based capacitor fault detection device of claim 5, wherein, The method further comprises the following steps: A fault screening module is configured to perform fault screening on the abnormal sound according to a preset transformer algorithm to obtain capacitor fault sound.

9. A sound-based capacitor fault detection apparatus, comprising: The device comprises a processor and a memory; The memory is configured to store program code and transmit the program code to the processor; The processor is configured to execute the sound-based capacitor fault detection method according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium is configured to store program code for executing the sound-based capacitor fault detection method according to any one of claims 1-4.

Images