A partial discharge signal detection method and system based on ultrasonic sensor array

Abstract

The present invention provides a partial discharge signal detection method based on an ultrasonic sensor array, comprising: S1: collecting sound data including partial discharge and excluding partial discharge, and training to obtain a partial discharge signal detection model after annotation; S2: using the partial discharge signal detection model for preliminary detection; S3: performing DOA estimation on the suspected partial discharge signal in the preliminary detection to determine the partial discharge direction; S4: performing enhancement processing on the sound signal with the partial discharge direction as the enhancement direction; S5: using the partial discharge signal monitoring model to identify the enhanced partial discharge signal to determine whether there is a partial discharge signal. The advantages of the present invention are: through the initial feature recognition, the suspected partial discharge signal fragment is intercepted, and then the DOA estimation is performed on the intercepted signal, which improves the signal-to-noise ratio, thereby improving the positioning accuracy; through the array enhancement technology, non-contact detection can be achieved, and the detection field of view can be expanded, and the equipment cost and operation time can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of partial discharge detection, and in particular to a partial discharge signal detection method and system based on an ultrasonic sensor array. Background Art

[0002] Insulation failure is one of the most common faults in common power system equipment, such as switchgear, GIS (gas insulated switchgear), transformers, etc. Insulation failure can lead to partial discharge, which can further deteriorate the insulation performance of the equipment. Therefore, the detection and location of partial discharge signals are important for power systems.

[0003] At present, partial discharge detection methods are mainly divided into two categories: ultrasonic method and ultra-high frequency method. The ultra-high frequency method detects based on electromagnetic wave signals, has good sensitivity, can locate the discharge position, and is conducive to the realization of real-time online detection. For details, please refer to Kabe, S., T. Yamagiwa and H. Okubo. Detection of Harmful Metallic Particles Inside Gas Insulated Switchgear using UHF Sensor [J]. IEEE Transactions on Dielectrics and Electrical Insulation, 2008, 15 (3): 701-709; but the disadvantage is that it is easily affected by electromagnetic interference and the equipment cost is relatively high. The ultrasonic method detects by measuring the ultrasonic signal generated by partial discharge. The advantage of this method is that the measurement is not related to the electrical circuit inside the electrical equipment, is not affected by the change of electrical quantity, and has strong resistance to electromagnetic interference. The disadvantage is that the measurement sensitivity is easily affected by the propagation path, the vibration of the measuring equipment and the environmental noise, resulting in large errors. The two methods have their own advantages and disadvantages and are suitable for different scenarios.

[0004] The invention patent application with publication number CN113092972A discloses a method and device for identifying partial discharge types based on a non-contact ultrasonic sensor, which uses single-channel data and performs detection by comparing partial discharge acoustic features. This type of method cannot accurately locate the partial discharge position, cannot form spatial filtering, and has limited acoustic anti-interference performance. The invention patent application with publication number CN105425128A discloses a transformer partial discharge ultrasonic detection and precise positioning device and method, which uses a multi-channel microphone to locate the partial discharge signal. However, this method only uses signal energy as detection, and does not use more partial discharge signal features. The detection accuracy is low, and the microphone needs to be arranged around the device being detected. The detection range is small and can only detect a single device. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for accurately detecting partial discharge signals based on an array of ultrasonic sensors.

[0006] The present invention solves the above technical problems through the following technical solutions: a partial discharge signal detection method based on an ultrasonic sensor array, comprising:

[0007] S1: Collect sound data including and excluding partial discharge, annotate and train to obtain a partial discharge signal detection model;

[0008] S2: construct a sensor array, continuously intercept the sound signal to be identified through a sliding window, and use the partial discharge signal detection model for preliminary detection;

[0009] S3: Estimating the DOA of the suspected PD signal in the preliminary detection and determining the PD direction;

[0010] S4: Enhance the sound signal using the local discharge direction as the enhancement direction;

[0011] S5: Use the partial discharge signal monitoring model to identify the enhanced partial discharge signal to determine whether there is a partial discharge signal.

[0012] The present invention intercepts suspected partial discharge signal fragments through initial feature recognition, and then estimates the DOA of the intercepted signal, thereby improving the signal-to-noise ratio and thus improving the positioning accuracy; array enhancement noise reduction is performed based on the result of DOA estimation, and then secondary partial discharge feature recognition and judgment are performed, which can remove most of the interference information and improve the robustness and accuracy of detection; through array enhancement technology, non-contact detection can be achieved, and the detection field of view can be expanded. Multiple devices can be monitored unmanned at the same time, reducing equipment costs and computing time.

[0013] Preferably, in step S1, the partial discharge signal monitoring model is obtained by training a 2D convolutional neural network.

[0014] Preferably, the 2D convolutional neural network includes an input layer, a 2D convolutional layer, a pooling layer, a 2D convolutional layer, a pooling layer, a fully connected layer and a softmax layer connected in sequence, and the softmax layer includes two output nodes, which respectively output the probabilities of containing partial discharge information and not containing partial discharge information.

[0015] Preferably, in step S1, the audio data is subjected to short-time Fourier transform before training to be converted into a spectrogram of equal time length, and the spectrogram is used as input data of the 2D convolutional neural network.

[0016] Preferably, the length of the sliding window in step S2 is the same as the time length of the training sample, the moving length of the sliding window is also the same as the time length of the training sample, and the time domain signal intercepted by the sliding window is subjected to short-time Fourier transform to obtain a spectrogram.

[0017] Preferably, the probability of the presence of partial discharge outputted by the partial discharge signal detection model during preliminary detection is P1, and if P1>PT1, the signal to be detected is considered to be a suspected partial discharge signal, wherein PT1<0.5.

[0018] Preferably, in step S3, the sound signal and sensor array position obtained by each sensor in the time period corresponding to the suspected partial discharge signal are obtained, and all sound signals are used as input for array DOA estimation to perform broadband DOA estimation; if the DOA spectrum peak is greater than a preset threshold T, the DOA corresponding to the spectrum peak position is the partial discharge direction.

[0019] Preferably, the method for enhancing the sound signal is:

[0020] Taking the DOA direction as the enhancement direction, the MVDR enhancement algorithm is used to reduce the noise of the signal intercepted in the sliding window. The formula is:

[0021] X = FFT(x)

[0022]

[0023]

[0024]

[0025]

[0026] Among them, X represents the multi-channel signal matrix intercepted by the sliding window, the matrix size is Nr×N, Nr is the number of channels, N is the time length of the sliding window, FFT() represents fast Fourier transform, IFFT() represents inverse fast Fourier transform, a(f) is the direction vector corresponding to the frequency point f, which is the result of the DOA calculation in the previous step, This is the output result after denoising.

[0027] Preferably, the enhanced and denoised time domain signal is re-short-time Fourier transformed, and the converted spectrogram is input into the partial discharge signal detection model to calculate the secondary probability of the existence of the partial discharge signal as P2. When P2≥PT2, it is considered that the intercepted signal contains a partial discharge signal; wherein PT2 is the probability threshold for determining the secondary partial discharge feature, and PT2>PT1.

[0028] The present invention also provides a partial discharge signal detection system based on an ultrasonic sensor array, comprising:

[0029] A training module collects sound data including and excluding partial discharge, annotates and trains to obtain a partial discharge signal detection model;

[0030] The preliminary detection module constructs a sensor array, continuously intercepts the sound signal to be identified through a sliding window, and uses a partial discharge signal detection model to perform preliminary detection;

[0031] The PD positioning module estimates the DOA of the suspected PD signal in the preliminary detection and determines the PD direction;

[0032] Enhanced noise reduction module, which uses the local discharge direction as the enhancement direction to enhance the sound signal;

[0033] The secondary determination module uses the partial discharge signal monitoring model to identify the enhanced partial discharge signal and determine whether a partial discharge signal exists.

[0034] The advantages of the partial discharge signal detection method and system based on the ultrasonic sensor array provided by the present invention are: through the initial feature recognition, the suspected partial discharge signal fragment is intercepted, and then the DOA estimation is performed on the intercepted signal, thereby improving the signal-to-noise ratio and thus improving the positioning accuracy; array enhancement noise reduction is performed based on the result of the DOA estimation, and then a secondary partial discharge feature recognition and judgment is performed, which can remove most of the interference information and improve the robustness and accuracy of the detection; through the array enhancement technology, non-contact detection can be achieved, and the detection field of view can be expanded, and unmanned monitoring of multiple devices can be performed at the same time, reducing equipment costs and computing time. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flow chart of a partial discharge signal detection method based on an ultrasonic sensor array provided in an embodiment of the present invention;

[0036] Figure 2 A structural diagram of a convolutional neural network for a partial discharge signal detection method based on an ultrasonic sensor array provided in an embodiment of the present invention;

[0037] Figure 3 A schematic diagram of a sliding window intercepting a sound clip in a partial discharge signal detection method based on an ultrasonic sensor array provided in an embodiment of the present invention;

[0038] Figure 4 A schematic diagram of a partial discharge signal detection system based on an ultrasonic sensor array provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the technical solution of the present invention is clearly and completely described below in combination with specific embodiments and with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] like Figure 1 As shown, this embodiment provides a partial discharge signal detection method based on an ultrasonic sensor array, comprising:

[0041] S1: Collect sound data including and excluding partial discharge, annotate and train to obtain a partial discharge signal detection model;

[0042] S2: construct a sensor array, continuously intercept the sound signal to be detected through a sliding window, and use the partial discharge signal detection model to perform preliminary detection;

[0043] S3: Estimating the DOA of the suspected PD signal in the preliminary detection and determining the PD direction;

[0044] S4: Enhance the sound signal using the local discharge direction as the enhancement direction;

[0045] S5: Use the partial discharge signal monitoring model to identify the enhanced partial discharge signal to determine whether there is a partial discharge signal.

[0046] This embodiment uses the initial feature recognition to intercept suspected partial discharge signal fragments, and then performs DOA estimation on the intercepted signal, thereby improving the signal-to-noise ratio and thus improving the positioning accuracy; array enhancement noise reduction is performed based on the results of the DOA estimation, and then a secondary partial discharge feature recognition and judgment is performed, which can remove most of the interference information and improve the robustness and accuracy of the detection; through array enhancement technology, non-contact detection can be achieved, and the detection field of view can be expanded. Unmanned monitoring of multiple devices can be performed at the same time, reducing equipment costs and computing time.

[0047] Specifically, the partial discharge signal detection method based on the ultrasonic sensor array provided in this embodiment includes the following steps:

[0048] S1: Collect sound data including and excluding partial discharge, annotate and train to obtain a partial discharge signal detection model;

[0049] In this embodiment, a 2D convolutional neural network is selected for training to obtain the partial discharge signal detection model. Figure 2The 2D convolutional neural network includes an input layer, a 2D convolutional layer, a pooling layer, a 2D convolutional layer, a pooling layer, a fully connected layer and a softmax layer connected in sequence. The softmax layer includes two output nodes, which are used to output the probabilities of containing partial discharge information and not containing partial discharge information, respectively.

[0050] The duration of the selected sound samples should be equal. Before model training, the sample data needs to be short-time Fourier transformed (STFT) and converted into a spectrogram of equal time length. The converted spectrogram is input into the 2D convolutional neural network for training. The specific training process and termination conditions are conventional means in the field and will not be repeated here.

[0051] S2: construct a sensor array, record the position coordinates of each sensor, continuously intercept the sound signal to be identified through a sliding window, and use the partial discharge signal detection model to perform preliminary detection;

[0052] Since each sensor arranged in the array will collect sound signals, the sound signal of any sensor can be selected as the signal to be identified and processed; the time length of the sliding window is consistent with the time length of the training sample, and the moving length of the sliding window should not be greater than the time length of the training sample to avoid omissions. In this embodiment, the moving length of the sliding window is set to be consistent with the time length of the training sample, and the moving length can also be set to half of the time length of the training sample to ensure that partial discharge information can be detected.

[0053] The time domain signal intercepted by the sliding window is converted into a spectrogram through short-time Fourier transform, and then input into the trained partial discharge signal detection model for judgment.

[0054] The probability of the existence of a partial discharge signal output by the partial discharge signal detection model is P1. If P1>PT1, the signal to be detected intercepted by the sliding window is considered to be a suspected partial discharge signal, where PT1 is a threshold value preset based on experience. In this embodiment, PT1<0.5.

[0055] S3: Estimating the DOA of the suspected PD signal in the preliminary detection and determining the PD direction;

[0056] For the time period corresponding to the suspected partial discharge signal, the sound signal of each sensor in the array and the array position of the sensor are obtained, and all sound signals are used as the input of the array DOA estimation to perform broadband DOA estimation. If the DOA spectrum peak is greater than the preset threshold T, the DOA corresponding to the spectrum peak position is the partial discharge direction, where the DOA estimation method is the existing technology and the threshold value T is an empirical value.

[0057] S4: Enhance the sound signal using the local discharge direction as the enhancement direction;

[0058] Taking the DOA direction as the enhancement direction, the MVDR enhancement algorithm is used to reduce the noise of the signal intercepted in the sliding window. The formula is:

[0059] X = FFT(x)

[0060]

[0061]

[0062]

[0063]

[0064] Among them, X represents the multi-channel signal matrix intercepted by the sliding window, the matrix size is Nr×N, Nr is the number of channels, N is the time length of the sliding window, FFT() represents fast Fourier transform, IFFT() represents inverse fast Fourier transform, a(f) is the direction vector corresponding to the frequency point f, which is the result of the DOA calculation in the previous step, This is the output result after denoising.

[0065] S5: Use the partial discharge signal monitoring model to identify the enhanced partial discharge signal to determine whether there is a partial discharge signal.

[0066] The enhanced and denoised time domain signal is re-short-time Fourier transformed, and the converted spectrogram is input into the partial discharge signal detection model to calculate the secondary probability of the existence of the partial discharge signal as P2. When P2≥PT2, it is considered that the intercepted signal contains a partial discharge signal; PT2 is the probability threshold for determining the secondary partial discharge feature, PT2>PT1, and in this embodiment, PT2=0.5.

[0067] Since the partial discharge signal is a transient signal, the accuracy of array positioning cannot be improved by time accumulation. However, this embodiment creatively identifies the partial discharge signal through two positionings, so that the sound signal where the partial discharge is obviously not present can be quickly filtered out, and then the DOA estimation and signal enhancement of the suspected signal are carried out in a targeted manner, and then the enhanced signal is further identified, thereby realizing the detection and positioning of partial discharge quickly, effectively and accurately, which plays an important role in the abnormal detection of the power system.

[0068] refer to Figure 4 This embodiment also provides a partial discharge signal detection system based on an ultrasonic sensor array, comprising:

[0069] A training module collects sound data including and excluding partial discharge, annotates and trains to obtain a partial discharge signal detection model;

[0070] The preliminary detection module constructs a sensor array, continuously intercepts the sound signal to be identified through a sliding window, and uses a partial discharge signal detection model to perform preliminary detection;

[0071] The PD positioning module estimates the DOA of the suspected PD signal in the preliminary detection and determines the PD direction;

[0072] Enhanced noise reduction module, which uses the local discharge direction as the enhancement direction to enhance the sound signal;

[0073] The secondary determination module uses the partial discharge signal monitoring model to identify the enhanced partial discharge signal and determine whether a partial discharge signal exists.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A partial discharge signal detection method based on an ultrasonic sensor array, characterized in that: include, S1: Collect sound data with and without partial discharge, annotate and train to obtain a partial discharge signal detection model; S2: Construct a sensor array, continuously capture the sound signal to be identified through a sliding window, and use a partial discharge signal detection model for preliminary detection; S3: Estimating the DOA of the suspected PD signal in the preliminary detection and determining the PD direction; S4: Enhance the sound signal using the local discharge direction as the enhancement direction; The method for enhancing the sound signal is as follows: Taking the DOA direction as the enhancement direction, the MVDR enhancement algorithm is used to reduce the noise of the signal intercepted in the sliding window. The formula is: X = FFT(x) Among them, X represents the multi-channel signal matrix intercepted by the sliding window, the matrix size is Nr×N, Nr is the number of channels, N is the time length of the sliding window, FFT() represents fast Fourier transform, IFFT() represents inverse fast Fourier transform, a(f) is the direction vector corresponding to the frequency point f, which is the result of the DOA calculation in the previous step, is the output result after noise reduction; S5: Use the partial discharge signal monitoring model to identify the enhanced partial discharge signal to determine whether there is a partial discharge signal.

2. The method for detecting partial discharge signals based on an ultrasonic sensor array according to claim 1, characterized in that: In step S1, the partial discharge signal monitoring model is obtained by training a 2D convolutional neural network.

3. The method for detecting partial discharge signals based on an ultrasonic sensor array according to claim 2, characterized in that: The 2D convolutional neural network includes an input layer, a 2D convolutional layer, a pooling layer, a 2D convolutional layer, a pooling layer, a fully connected layer and a softmax layer connected in sequence. The softmax layer includes two output nodes, which respectively output the probabilities of containing partial discharge information and not containing partial discharge information.

4. The method for detecting partial discharge signals based on an ultrasonic sensor array according to claim 2, characterized in that: In step S1, the audio data is subjected to short-time Fourier transform before training to be converted into a spectrogram of equal time length, and the spectrogram is used as input data of the 2D convolutional neural network.

5. The method for detecting partial discharge signals based on an ultrasonic sensor array according to claim 4, characterized in that: The length of the sliding window in step S2 is the same as the time length of the training sample, the moving length of the sliding window is also the same as the time length of the training sample, and the time domain signal intercepted by the sliding window is subjected to short-time Fourier transform to obtain a spectrogram.

6. The method for detecting partial discharge signals based on an ultrasonic sensor array according to claim 1, characterized in that: The probability of the presence of partial discharge output by the partial discharge signal detection model is P1. If P1>PT1, the sound signal to be identified is considered to be a suspected partial discharge signal, where PT1 is a threshold value preset based on experience, and PT1<0.

5.

7. The method for detecting partial discharge signals based on an ultrasonic sensor array according to claim 6, characterized in that: In step S3, the sound signal and sensor array position obtained by each sensor in the time period corresponding to the suspected partial discharge signal are obtained, and all sound signals are used as inputs for array DOA estimation to perform broadband DOA estimation; if the DOA spectrum peak is greater than the preset threshold T, the DOA corresponding to the spectrum peak position is the partial discharge direction.

8. The method for detecting partial discharge signals based on an ultrasonic sensor array according to claim 7, characterized in that: The enhanced and denoised time domain signal is re-short-time Fourier transformed, and the converted spectrogram is input into the partial discharge signal detection model to calculate the secondary probability of the existence of the partial discharge signal as P2. When P2≥PT2, it is considered that the intercepted signal contains a partial discharge signal; PT2 is the probability threshold for determining the secondary partial discharge feature, PT2>PT1, and PT1 is a threshold value preset based on experience.

9. A partial discharge signal detection system based on an ultrasonic sensor array, characterized in that: include, A training module collects sound data including and excluding partial discharge, annotates and trains to obtain a partial discharge signal detection model; The preliminary detection module constructs a sensor array, continuously intercepts the sound signal to be identified through a sliding window, and uses a partial discharge signal detection model to perform preliminary detection; The PD positioning module estimates the DOA of the suspected PD signal in the preliminary detection and determines the PD direction; Enhanced noise reduction module, which uses the local discharge direction as the enhancement direction to enhance the sound signal; The method for enhancing the sound signal is as follows: Taking the DOA direction as the enhancement direction, the MVDR enhancement algorithm is used to reduce the noise of the signal intercepted in the sliding window. The formula is: X = FFT(x) Among them, X represents the multi-channel signal matrix intercepted by the sliding window, the matrix size is Nr×N, Nr is the number of channels, N is the time length of the sliding window, FFT() represents fast Fourier transform, IFFT() represents inverse fast Fourier transform, a(f) is the direction vector corresponding to the frequency point f, which is the result of the DOA calculation in the previous step, is the output result after noise reduction; The secondary determination module uses the partial discharge signal monitoring model to identify the enhanced partial discharge signal and determine whether a partial discharge signal exists.

Images