Signal processing method for partial discharge signal and related device

Abstract

The invention provides a signal processing method for a partial discharge signal and a related device, and the method comprises the steps: carrying out the frequency band extraction of a first sound pulse signal, and obtaining a second sound pulse signal corresponding to an audible sound frequency band and a third sound pulse signal corresponding to an ultrasonic frequency band; determining a first time domain waveform of the second sound pulse signal, and determining a second time domain waveform of the third sound pulse signal; determining a first frequency domain waveform of the second sound pulse signal, and determining a second frequency domain waveform of the third sound pulse signal; inputting the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector; inputting the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector; splicing the image feature vector and the voiceprint feature vector to obtain a target feature vector; and inputting the target feature vector into a classification network to obtain a target classification result.

Description

Technical Field

[0001] The present application relates to the field of signal processing technology or signal processing technology, and specifically to a signal processing method and related device for partial discharge signals. Background Art

[0002] In practical applications, partial discharge can be understood as the discharge between unpenetrated electrodes caused by defects in the production process of electronic products (such as equipment or components) or other external reasons under the action of high electric field strength, and repeated breakdown and extinction.

[0003] Usually, there are many types of partial discharges, and the recognition accuracy of partial discharge types is often poor. Therefore, the problem of how to improve the recognition accuracy of partial discharge types needs to be solved urgently. Summary of the invention

[0004] The embodiments of the present application provide a signal processing method and related devices for partial discharge signals, which can improve the accuracy of identifying partial discharge categories.

[0005] In a first aspect, an embodiment of the present application provides a signal processing method for a partial discharge signal, the method comprising: Capturing the first acoustic pulse signal generated by partial discharge through a sensor; Performing frequency band extraction on the first acoustic pulse signal to obtain a second acoustic pulse signal corresponding to an audible sound frequency band and a third acoustic pulse signal corresponding to an ultrasonic frequency band; Determine a first time domain waveform of the second acoustic pulse signal, and determine a second time domain waveform of the third acoustic pulse signal; Determine a first frequency domain waveform of the second acoustic pulse signal, and determine a second frequency domain waveform of the third acoustic pulse signal; Inputting the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector; Inputting the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights; Concatenate the image feature vector and the voiceprint feature vector to obtain a target feature vector; The target feature vector is input into the classification network to obtain the target classification result.

[0006] In a second aspect, an embodiment of the present application provides a signal processing device for partial discharge signals, the device comprising: an acquisition unit, an extraction unit, a determination unit, a splicing unit and a classification unit, wherein: The acquisition unit is used to capture the first acoustic pulse signal generated by the partial discharge through a sensor; The extraction unit is used to perform frequency band extraction on the first acoustic pulse signal to obtain a second acoustic pulse signal corresponding to an audible sound frequency band and a third acoustic pulse signal corresponding to an ultrasonic frequency band; The determining unit is used to determine a first time domain waveform of the second acoustic pulse signal and a second time domain waveform of the third acoustic pulse signal; determine a first frequency domain waveform of the second acoustic pulse signal and a second frequency domain waveform of the third acoustic pulse signal; The extraction unit is further used to input the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector; input the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights; The splicing unit is used to splice the image feature vector and the voiceprint feature vector to obtain a target feature vector; The classification unit is used to input the target feature vector into a classification network to obtain a target classification result.

[0007] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the program includes instructions for executing the steps in the first aspect of the embodiment of the present application.

[0008] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps described in the first aspect of the embodiment of the present application.

[0009] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps described in the first aspect of the embodiment of the present application. The computer program product may be a software installation package.

[0010] The implementation of the embodiments of the present application has the following beneficial effects: It can be seen that the signal processing method and related device for local discharge signals described in the embodiments of the present application capture the first acoustic pulse signal generated by local discharge through a sensor, perform frequency band extraction on the first acoustic pulse signal, obtain a second acoustic pulse signal corresponding to the audible sound frequency band and a third acoustic pulse signal corresponding to the ultrasonic frequency band, determine the first time domain waveform of the second acoustic pulse signal, determine the second time domain waveform of the third acoustic pulse signal, determine the first frequency domain waveform of the second acoustic pulse signal, determine the second frequency domain waveform of the third acoustic pulse signal, input the first time domain waveform and the second time domain waveform into the first convolutional neural network model for feature extraction to obtain an image feature vector, input the first frequency domain waveform and the second frequency domain waveform into the second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights, and the image feature vector and the voiceprint feature vector are spliced. , get the target feature vector, input the target feature vector into the classification network, and get the target classification result. First, considering the different sound characteristics of the audible sound band and the ultrasonic frequency band, the local discharge can be directly divided into the corresponding sound pulse signals of the two different frequency bands, the audible sound band and the ultrasonic frequency band, and the corresponding features are extracted from the time domain and frequency domain dimensions for the corresponding sound pulse signals of the two different frequency bands, the audible sound band and the ultrasonic frequency band. Second, in the feature extraction process, considering the homology of the signals in the time domain and the frequency domain, the twin neural network is used for feature extraction, which can reduce the difference between the time domain and the frequency domain, making the feature extraction more in line with the characteristics of the signal itself, and can also ensure the deep extraction of signal features. Third, the features of the time domain and the frequency domain after feature extraction are spliced, and finally input into the classification network for classification, thereby improving the recognition accuracy of the local discharge category. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0012] Figure 1 It is a flow chart of a signal processing method for partial discharge signals provided in an embodiment of the present application; Figure 2 is another flow chart of a signal processing method for partial discharge signals provided by an embodiment of the present application; Figure 3 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application; Figure 4It is a block diagram of the functional units of a signal processing device for partial discharge signals provided in an embodiment of the present application. DETAILED DESCRIPTION

[0013] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but in a possible example also includes steps or units that are not listed, or in a possible example also includes other steps or units inherent to these processes, methods, products or devices.

[0014] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0015] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0016] The electronic device involved in the embodiments of the present application may include a partial discharge detector, or other devices with a discharge detection function.

[0017] In the embodiment of the present application, phase resolved partial discharge (PRPD) specifically refers to displaying each partial discharge pulse with a phase mark according to the phase. In the specific implementation, the discharge information has no time information and is a superposition of PRPS information within a period of time, for example, within 1 second.

[0018] In the embodiments of the present application, partial discharge (PD) may occur in electrical insulating materials, usually due to the local electric field strength exceeding the breakdown voltage of the material itself. Partial discharge generally also occurs in the insulation system of electrical equipment, such as insulated cables, transformers, switchgear, etc., which are not limited here. In actual applications, the frequency of partial discharge will increase significantly for application scenarios such as aging or damaged insulating materials.

[0019] In the embodiment of the present application, the partial discharge category can be based on at least one of the following: surface discharge, internal discharge, corona discharge, floating potential discharge, surface discharge, discharge between conductor and insulator, air gap discharge, end discharge, slot discharge, etc., which are not limited here. Different partial discharges also have different characteristics. For example, different partial discharge sound characteristics will also be different.

[0020] In the embodiment of the present application, surface discharge occurs on the surface of insulating medium, which is common in power cables, motor windings, etc. The voltage waveform, electric field distribution, air quality and other factors of surface discharge will affect its discharge characteristics.

[0021] In the embodiments of the present application, internal discharge can be understood as generally occurring inside a solid insulating medium, which is caused by material and process defects. In specific implementations, internal discharge is usually affected by factors such as dielectric properties and air gap size.

[0022] In the embodiments of the present application, corona discharge can be understood as generally occurring around a high-voltage conductor surrounded by gas, such as a high-voltage transmission line or a high-voltage terminal of a high-voltage transformer. In specific implementations, corona discharge is a self-sustaining discharge form under an extremely non-uniform electric field, which may be affected by factors such as applied voltage, electrode shape, weather, and gas density.

[0023] In the embodiments of the present application, the floating potential discharge can be understood as the discharge caused by the design defects or poor contact of the device itself, usually, for example, between high-potential metal parts or ground-potential metal parts.

[0024] In the embodiments of the present application, surface discharge can be understood as a local discharge phenomenon occurring on the surface of a medium in an electric field, which is usually common in places such as bushing flanges or cable terminals.

[0025] In the embodiments of the present application, the discharge between the conductor and the insulator can be understood as the discharge caused by the air gap between the conductor and the insulator, which is usually common in the stator winding of a high-voltage motor.

[0026] In the embodiments of the present application, air gap discharge can be understood as discharge caused by air gaps or impurities inside the insulating medium, which is usually common in the insulation system of high-voltage electrical equipment.

[0027] In the embodiment of the present application, end discharge can be understood as surface discharge at the connection point of the end of the stator winding of the generator, usually due to vibration and / or friction.

[0028] In the embodiment of the present application, slot discharge can be understood as discharge between the main insulation surface of the coil and the slot wall, which is usually common when the coil insulator expands unevenly due to heat.

[0029] See also Figure 1 , Figure 1 1 is a flow chart of a signal processing method for a partial discharge signal provided in an embodiment of the present application. As shown in the figure, the signal processing method for a partial discharge signal includes: 101. Capture the first acoustic pulse signal generated by local discharge through a sensor.

[0030] In an embodiment of the present application, the sensor may include one or more sensors. In a specific implementation, the first acoustic pulse signal generated by local discharge may be captured by one sensor. For example, the first acoustic pulse signal generated by local discharge may also be captured by multiple sensors.

[0031] For example, the sensor may include a sensor for the audible frequency band and a sensor for the ultrasonic frequency band. The sensor for the audible frequency band may include a microphone array pickup, and the sensor for the ultrasonic frequency band may include an air ultrasonic sensor. The microphone array pickup is used to capture sound signals in the audible frequency band, and the air ultrasonic sensor can be used to capture sound signals in the ultrasonic frequency band.

[0032] In a specific implementation, the first acoustic pulse signal may include an acoustic pulse signal within a period of time, or an acoustic pulse signal at a certain time point.

[0033] In a specific implementation, the first acoustic pulse signal may be a digital signal or an analog signal.

[0034] 102. Perform frequency band extraction on the first acoustic pulse signal to obtain a second acoustic pulse signal corresponding to an audible sound frequency band and a third acoustic pulse signal corresponding to an ultrasonic frequency band.

[0035] In the embodiment of the present application, since the audible sound frequency band and the ultrasonic frequency band correspond to different sound frequency bands, the first sound pulse signal can be subjected to frequency band extraction, that is, a second sound pulse signal corresponding to the audible sound frequency band and a third sound pulse signal corresponding to the ultrasonic frequency band are obtained.

[0036] In a specific implementation, the second acoustic pulse signal and the third acoustic pulse signal may come from the same sensor, or, of course, may come from different sensors.

[0037] 103. Determine a first time domain waveform of the second acoustic pulse signal, and determine a second time domain waveform of the third acoustic pulse signal.

[0038] In a specific implementation, the first time domain waveform of the second acoustic pulse signal and the second time domain waveform of the third acoustic pulse signal can be determined, that is, the first time domain waveform is a waveform corresponding to the audible sound frequency band, and the second time domain waveform is a waveform corresponding to the ultrasonic frequency band.

[0039] 104. Determine a first frequency domain waveform of the second acoustic pulse signal, and determine a second frequency domain waveform of the third acoustic pulse signal.

[0040] In a specific implementation, the first frequency domain waveform of the second acoustic pulse signal and the second frequency domain waveform of the third acoustic pulse signal can also be determined, wherein the first frequency domain waveform is a waveform corresponding to the audible sound frequency band and the second frequency domain waveform is a waveform corresponding to the ultrasonic frequency band.

[0041] 105. Input the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector.

[0042] In a specific implementation, the first convolutional neural network model can be preset or set by system default.

[0043] In an embodiment of the present application, the first time domain waveform and the second time domain waveform can be input into a first convolutional neural network model for feature extraction to obtain an image feature vector.

[0044] Among them, the first convolutional neural network model can be obtained by training the spectral support set of various partial discharge time domain waveforms. The minimum size of the spectral support set samples (such as K-Shot) is 2 spectra, which can be a spectrum corresponding to the audible sound frequency band and a spectrum corresponding to the ultrasonic frequency band. Therefore, the first convolutional neural network model can extract corresponding features based on the partial discharge characteristics.

[0045] Optionally, the above step 105, inputting the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector, may include the following steps: Determine a first PRPD spectrum according to the first time domain waveform; determining a second PRPD spectrum according to the second time domain waveform; Determining a target PRPD spectrum according to the first PRPD spectrum and the second PRPD spectrum; The target PRPD atlas is input into the first convolutional neural network model for feature extraction to obtain the image feature vector.

[0046] In a specific implementation, a first PRPD spectrum can be determined based on a first time domain waveform, a second PRPD spectrum can be determined based on a second time domain waveform, and a target PRPD spectrum can be determined based on the first PRPD spectrum and the second PRPD spectrum, wherein the target PRPD spectrum can be the first PRPD spectrum and the second PRPD spectrum existing independently, or the first PRPD spectrum and the second PRPD spectrum images can be fused together, or the two can be spliced ​​together.

[0047] Next, the target PRPD map can be input into the first convolutional neural network model for feature extraction to obtain an image feature vector.

[0048] Among them, the first time domain waveform and the second time domain waveform can be subjected to noise reduction processing or not. When subjected to noise reduction processing, the quality of the waveform can be improved, noise can be suppressed, and signal quality can be improved. When no noise reduction processing is performed, the characteristics of the original signal can be guaranteed.

[0049] 106. Input the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights.

[0050] In a specific implementation, the second convolutional neural network model can be preset or set by system default.

[0051] In an embodiment of the present application, the first frequency domain waveform and the second frequency domain waveform may be input into a second neural network model for feature extraction to obtain a voiceprint feature vector.

[0052] Among them, the second convolutional neural network model can be obtained by training the spectrogram support set of various partial discharge frequency domain waveforms. The minimum size of the spectrogram support set (such as K-Shot) is 2 spectra, which can be a spectrogram corresponding to the audible sound frequency band and a spectrogram corresponding to the ultrasonic frequency band. Therefore, the second convolutional neural network model can extract corresponding features based on the partial discharge characteristics.

[0053] Among them, the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights. For example, the shared weights of the twin prototype network are obtained through learning and training of the miniImageNet small sample data set. The sampling method of small sample learning is N-Way / K-Shot, which is used to learn sample measurement methods under different types. In the partial discharge recognition scenario, N-way corresponds to the number of categories. For example, it can be usually 8 according to the national grid standard, and K-shot is the number of samples under a single category.

[0054] Among them, a small sample can be understood as a sample number that is smaller than a sample threshold number, wherein the sample threshold number can be preset or system default, for example, the sample threshold number is 10, 5, 3, etc., which is not limited here.

[0055] Among them, the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights, which can reduce the difference between the time domain and the frequency domain, making the feature extraction more in line with the characteristics of the signal itself, and can also ensure the deep extraction of signal features.

[0056] Optionally, the above step 106, inputting the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector, may include the following steps: Performing a first noise reduction process on the first frequency domain waveform to obtain a target first frequency domain wave; Performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform; Determine a first spectrogram according to the target first frequency domain waveform; determining a second spectrogram according to the second frequency domain waveform; Determine a target spectrogram according to the first spectrogram and the second spectrogram; The target spectrogram is input into the second neural network model for feature extraction to obtain the voiceprint feature vector.

[0057] The first noise reduction processing and the second noise reduction processing may be the same or different.

[0058] In a specific implementation, the first frequency domain waveform can be subjected to a first noise reduction process to obtain a target first frequency domain wave, thereby improving the quality of the first frequency domain waveform, suppressing noise, and improving signal quality. Correspondingly, the second frequency domain waveform can be subjected to a second noise reduction process to obtain a target second frequency domain wave, thereby improving the quality of the second frequency domain waveform, suppressing noise, and improving signal quality.

[0059] In a specific implementation, the first spectrogram can be determined according to the target first frequency domain waveform, and the second spectrogram can be determined according to the second frequency domain waveform, and then the target spectrogram can be determined according to the first spectrogram and the second spectrogram, wherein the target spectrogram can be the first spectrogram and the second spectrogram existing independently, or the first spectrogram and the second spectrogram images can be fused together, or the two can be spliced ​​together.

[0060] Next, the target spectrogram can be input into the second neural network model for feature extraction to obtain a voiceprint feature vector.

[0061] Optionally, the above step of performing a first noise reduction process on the first frequency domain waveform to obtain a target first frequency domain waveform may include the following steps: determining a first energy value of the first frequency domain waveform; determining a second energy value of the second time domain waveform; determining a first relative deviation between the first energy value and the second energy value; determining a first signal-to-noise ratio of the first frequency domain waveform; determining a second signal-to-noise ratio of the second frequency domain waveform; determining a second relative deviation between the first signal-to-noise ratio and the second signal-to-noise ratio; determining a first absolute value of a difference between the first relative deviation and the second relative deviation; determining a first noise reduction processing algorithm corresponding to the first absolute value; Acquire a noise reduction algorithm control parameter set corresponding to the first noise reduction processing algorithm, wherein the noise reduction algorithm control parameter set includes a plurality of noise reduction algorithm control parameters, and each noise reduction algorithm control parameter corresponds to a signal-to-noise ratio; Acquire a first noise reduction algorithm control parameter corresponding to the first signal-to-noise ratio from the noise reduction algorithm control parameter set; The first frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the first noise reduction algorithm control parameters to obtain the target first frequency domain waveform.

[0062] In a specific implementation, the first energy value of the first frequency domain waveform, that is, the energy value corresponding to the audible sound frequency band, can be determined. Then, the second energy value of the second time domain waveform, that is, the energy value corresponding to the ultrasonic frequency band, can be determined. Next, the first relative deviation between the first energy value and the second energy value can be determined. The first relative deviation = (first energy value - second energy value) / (first energy value + second energy value). The first relative deviation reflects the energy difference between the audible sound frequency band and the ultrasonic frequency band to a certain extent. The first relative deviation not only represents the deviation size, but also represents the deviation direction.

[0063] Next, the first signal-to-noise ratio of the first frequency domain waveform and the second signal-to-noise ratio of the second frequency domain waveform can be determined, and the second relative deviation between the first signal-to-noise ratio and the second signal-to-noise ratio can be determined. The second relative deviation = (first signal-to-noise ratio - second signal-to-noise ratio) / (first signal-to-noise ratio + second signal-to-noise ratio), which reflects the difference in signal quality between the audible sound band and the ultrasonic frequency band to a certain extent. The second relative deviation not only represents the deviation size, but also represents the deviation direction.

[0064] Next, the first absolute value of the difference between the first relative deviation and the second relative deviation can also be determined. The first absolute value = |first relative deviation-second relative deviation|. The first absolute value reflects to a certain extent the degree of difference in signal characteristics between the audible sound frequency band and the ultrasonic frequency band. In a specific implementation, the degree of difference in signal characteristics can be determined based on a joint decision of signal energy difference and signal quality difference (such as signal-to-noise ratio difference).

[0065] Furthermore, a mapping relationship between a preset absolute value and a noise reduction processing algorithm can be pre-stored, and a first noise reduction processing algorithm corresponding to the first absolute value can be determined based on the mapping relationship. A noise reduction algorithm control parameter set corresponding to the first noise reduction processing algorithm can also be obtained. The noise reduction algorithm control parameter set includes multiple noise reduction algorithm control parameters, and each noise reduction algorithm control parameter corresponds to a signal-to-noise ratio, that is, different signal-to-noise ratios are adapted to different noise reduction algorithm control parameters, thereby ensuring the noise reduction processing effect and preventing both over-noise reduction and under-noise reduction.

[0066] The noise reduction algorithm control parameter is used for at least one of the following functions: noise reduction processing effect, noise reduction processing speed, noise reduction processing position, etc., which are not limited here.

[0067] In a specific implementation, a first noise reduction algorithm control parameter corresponding to the first signal-to-noise ratio can be obtained from a noise reduction algorithm control parameter set, and then a noise reduction algorithm control parameter suitable for the signal-to-noise ratio can be obtained, and then the first frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the first noise reduction algorithm control parameter to obtain a target first frequency domain waveform. In this way, on the one hand, the difference in signal quality between the audible sound frequency band and the ultrasonic frequency band and the difference in signal quality between the audible sound frequency band and the ultrasonic frequency band are considered to determine the degree of difference in signal characteristics, thereby associating the audible sound frequency band and the ultrasonic frequency band, that is, considering both the individuality of the two and the correlation between the two, and thereby determining a corresponding noise reduction processing algorithm based on the degree of difference in signal characteristics, so that the noise reduction processing effect is consistent with both the individuality of the two and the correlation between the two. On the other hand, the noise reduction algorithm control parameter of the appropriate noise reduction processing algorithm can be adapted based on the signal-to-noise ratio, so that the final noise reduction effect depth is consistent with the actual signal quality, which can prevent both over-noise reduction and noise reduction.

[0068] Optionally, the above step of performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform may include the following steps: Determine a third signal-to-noise ratio corresponding to the target first frequency domain wave; determining a target deviation between the third signal-to-noise ratio and the first signal-to-noise ratio; Determining a target feedback adjustment parameter according to the target deviation and the second relative deviation; Acquire a second noise reduction algorithm control parameter corresponding to the second signal-to-noise ratio from the noise reduction algorithm control parameter set; Performing feedback adjustment on the second noise reduction algorithm control parameter according to the target feedback adjustment parameter to obtain the target second noise reduction algorithm control parameter; The second frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the target second noise reduction algorithm control parameters to obtain the target second frequency domain waveform.

[0069] In a specific implementation, the third signal-to-noise ratio corresponding to the target first frequency domain wave can be determined, and the target deviation between the third signal-to-noise ratio and the first signal-to-noise ratio can also be determined. The target deviation = (third signal-to-noise ratio - first signal-to-noise ratio) / first signal-to-noise ratio. The target deviation reflects the noise reduction effect to a certain extent. The second relative deviation reflects the difference in signal quality between the audible sound band and the ultrasonic frequency band to a certain extent. Then, the target feedback adjustment parameter can be determined according to the target deviation and the second relative deviation. Then, the corresponding feedback adjustment parameter can be dynamically determined based on the noise reduction effect of the audible sound band and the difference in signal quality between the audible sound band and the ultrasonic frequency band, thereby effectively preventing over-noise reduction and noise reduction, and optimizing the noise reduction effect of the ultrasonic frequency band to a certain extent.

[0070] Next, a second noise reduction algorithm control parameter corresponding to the second signal-to-noise ratio can be obtained from the noise reduction algorithm control parameter set, and then the second noise reduction algorithm control parameter can be feedback-adjusted according to the target feedback adjustment parameter to obtain the target second noise reduction algorithm control parameter, and the second frequency domain waveform can be noise-reduced according to the first noise reduction processing algorithm and the target second noise reduction algorithm control parameter to obtain the target second frequency domain waveform. Not only can the noise reduction algorithm control parameter of the appropriate noise reduction processing algorithm be adapted based on the signal-to-noise ratio so that the final noise reduction effect is deeply consistent with the actual signal quality, which can prevent both over-noise reduction and noise reduction, but also the corresponding feedback adjustment parameter can be dynamically determined based on the noise reduction effect in the audible sound band and the signal quality difference between the audible sound band and the ultrasonic band, thereby further deeply optimizing the noise reduction effect in the ultrasonic band.

[0071] Optionally, the above step of determining the target feedback adjustment parameter according to the target deviation and the second relative deviation may include the following steps: determining a target signal-to-noise ratio level corresponding to the first signal-to-noise ratio; Acquire a historical enhancement effect parameter set corresponding to the first noise reduction processing algorithm according to the target signal-to-noise ratio level, wherein the historical enhancement effect parameter set includes a plurality of historical enhancement effect parameters, and each historical enhancement effect parameter is used to represent a signal-to-noise ratio improvement ratio; selecting a maximum value and a minimum value among the plurality of historical enhancement effect parameters, dividing the plurality of historical enhancement effect parameters into a plurality of interval segments according to the maximum value and the minimum value, and counting the number of historical enhancement effect parameters in each of the plurality of interval segments to obtain a plurality of numbers; Determine a target interval segment among the multiple interval segments corresponding to the target deviation, and obtain a target quantity and a target interval position corresponding to the target interval segment; determining a first ratio between the target quantity and the sum of the plurality of quantities; determining a second ratio between the target interval position and the plurality of quantities; determining a first feedback adjustment parameter corresponding to the second relative deviation; determining a first standard deviation of the plurality of historical enhancement effect parameters; determining a first optimization parameter corresponding to the first standard deviation; determining a third relative deviation between the first ratio and the second ratio; determining a second optimization parameter corresponding to the third relative deviation; The first feedback adjustment parameter is optimized according to the first optimization parameter and the second optimization parameter to obtain the target feedback adjustment parameter.

[0072] In a specific implementation, a mapping relationship between preset signal-to-noise ratios may be pre-stored, and a target signal-to-noise ratio level corresponding to the first signal-to-noise ratio may be determined based on the mapping relationship. Of course, big data may also be used to obtain a historical enhancement effect parameter set corresponding to the first noise reduction processing algorithm according to the target signal-to-noise ratio level. The historical enhancement effect parameter set includes a plurality of historical enhancement effect parameters, and each historical enhancement effect parameter is used to represent a signal-to-noise ratio improvement ratio. That is, a historical enhancement effect parameter set corresponding to the target signal-to-noise ratio level and the first noise reduction processing algorithm may be obtained from the process of different users using the first noise reduction processing algorithm.

[0073] Of course, you can also select the maximum and minimum values ​​of multiple historical enhancement effect parameters, and divide the multiple historical enhancement effect parameters into multiple intervals according to the maximum and minimum values. The number of intervals in the multiple intervals can be determined based on the difference between the maximum and minimum values, or it can be set by the user. Then, the number of historical enhancement effect parameters of each interval in the multiple intervals can be counted to obtain multiple numbers, and then the target interval in the multiple intervals corresponding to the target deviation is determined, and the target number and target interval position corresponding to the target interval are obtained, and the first ratio between the target number and the sum of the multiple numbers is determined, that is, the first ratio = target number / (sum of multiple numbers), and the second ratio between the target interval position and the multiple numbers is determined. For example, there are 6 intervals, and the serial numbers of the intervals are 1, 2, 3, 4, 5, and 6 respectively. Then the second ratio can be the ratio between the serial number corresponding to the target interval position and the total serial number.

[0074] In a specific implementation, the mapping relationship between the preset relative deviation and the feedback adjustment parameter can be pre-stored, and then, the first feedback adjustment parameter corresponding to the second relative deviation can be determined based on the mapping relationship. The first standard deviation of multiple historical enhancement effect parameters can also be determined, that is, the standard deviation operation of multiple historical enhancement effect parameters can be performed to obtain the first standard deviation, which reflects the stability of the noise reduction effect.

[0075] In a specific implementation, a mapping relationship between a preset standard deviation and an optimization parameter may be pre-stored, and a first optimization parameter corresponding to the first standard deviation may be determined based on the mapping relationship. Then, a third relative deviation between the first ratio and the second ratio may be determined, where the third relative deviation = (first ratio - second ratio) / (first ratio + second ratio). A mapping relationship between a preset relative deviation and an optimization parameter may be pre-stored, and a second optimization parameter corresponding to the third relative deviation may be determined based on the mapping relationship. Then, the first feedback adjustment parameter may be optimized according to the first optimization parameter and the second optimization parameter to obtain a target feedback adjustment parameter, that is, the target feedback adjustment parameter = (1 + first optimization parameter) * (1 + second optimization parameter) * first feedback adjustment parameter. First, the second relative deviation reflects, to a certain extent, the difference in signal quality between the audible sound frequency band and the ultrasonic frequency band, and the corresponding first feedback may be dynamically determined based on the difference in signal quality between the audible sound frequency band and the ultrasonic frequency band. Secondly, by using big data, a historical enhancement effect parameter set corresponding to the first noise reduction processing algorithm is obtained according to the target signal-to-noise ratio level, and then the stability of the noise reduction effect is determined based on the big data, and the corresponding first optimization parameter is determined based on the stability of the noise reduction effect. Thirdly, since the third relative deviation reflects the consistency of noise reduction to a certain extent, the smaller the absolute value of the third relative deviation is, the better the noise reduction consistency is, and vice versa, the larger the absolute value of the third relative deviation is, the worse the noise reduction consistency is. The third relative deviation is also directional, and then the noise reduction effect is dynamically optimized to a certain extent. In this way, the corresponding feedback adjustment parameters are dynamically determined based on the noise reduction effect of the audible sound frequency band and the signal quality difference between the audible sound frequency band and the ultrasonic frequency band, thereby further optimizing the noise reduction effect of the ultrasonic frequency band.

[0076] 107. Concatenate the image feature vector and the voiceprint feature vector to obtain a target feature vector.

[0077] In a specific implementation, the image feature vector and the voiceprint feature vector may be spliced ​​to obtain a target feature vector. Specifically, the image feature vector in the time domain and the voiceprint feature vector in the frequency domain are fused through a fully connected layer to complete the splicing and obtain the target feature vector.

[0078] 108. Input the target feature vector into a classification network to obtain a target classification result.

[0079] The classification network may be preset or set by system default. In a specific implementation, the target feature vector may be input into the classification network to obtain a target classification result. Different classification results may correspond to different categories. The classification network may identify at least one partial discharge category, each partial discharge category may have a probability value, and the target classification result may be a partial discharge category corresponding to a probability value having a probability value greater than a specified probability value.

[0080] Each partial discharge category may also correspond to a category label. The partial discharge category may be determined based on at least one of the following classification criteria, which may include at least one of the following: partial discharge cause, partial discharge location, partial discharge environment, etc., which are not limited here.

[0081] In the embodiment of the present application, the partial discharge category can be based on at least one of the following: surface discharge, internal discharge, corona discharge, floating potential discharge, surface discharge, discharge between conductor and insulator, air gap discharge, end discharge, slot discharge, etc., which are not limited here. Different partial discharges also have different characteristics. For example, different partial discharge sound characteristics will also be different.

[0082] Next, a corresponding maintenance strategy can also be determined based on the target classification result. For example, a mapping relationship between a preset classification result and a maintenance strategy can be pre-stored, and then a corresponding maintenance strategy corresponding to the target classification result can be determined based on the mapping relationship. The maintenance strategy can include at least one of the following: replacing a component of the equipment, inspecting the partial discharge position, etc., which are not limited here.

[0083] Next, a corresponding early warning strategy can also be determined based on the target classification result. For example, a mapping relationship between a preset classification result and an early warning strategy can be pre-stored, and then, a corresponding early warning strategy corresponding to the target classification result can be determined based on the mapping relationship. The early warning strategy can include at least one of the following: voice prompt, SMS prompt, light prompt, alarm prompt, etc., which are not limited here.

[0084] In the specific implementation, the local discharge process will generate acoustic pulses. The width of this acoustic wave frequency band is very large (in the order of magnitude range of 10Hz-10^7Hz). Affected by the test environment, the state of discharge and the medium during the propagation process, the measured acoustic wave signal frequency range is 10Hz-10^5Hz. Of course, in addition to this, the acoustic wave spectrum formed by the discharge may also be affected by the specific type of discharge. For example, the acoustic spectra generated by air gap discharge and tip discharge are different. For discharge in liquid, depending on the form of the electrode, the frequency range of its acoustic spectrum can also be from tens of Hz to 2MHz. For example, the discharge phenomenon generated between flat plates has an acoustic wave frequency of no more than 300 kHz; for example, the peak of the acoustic spectrum of discharge occurring in solids is periodic.

[0085] For example, Figure 2 As shown, a first acoustic pulse signal can be obtained, and a time domain PRPD spectrum and a frequency domain spectrogram can be determined based on the first acoustic pulse signal. The time domain PRPD spectrum is then input into a first convolutional neural network, and the frequency domain spectrogram is input into a second convolutional neural network. The first convolutional neural network and the second convolutional neural network are used as a twin convolutional neural network with shared weights. The time domain features and the frequency domain are then spliced ​​through a fully connected layer, and the spliced ​​features are finally input into a classification network for classification.

[0086] It can be seen that the signal processing method for partial discharge signals described in the embodiment of the present application captures the first acoustic pulse signal generated by partial discharge through a sensor, performs frequency band extraction on the first acoustic pulse signal, obtains a second acoustic pulse signal corresponding to the audible sound frequency band and a third acoustic pulse signal corresponding to the ultrasonic frequency band, determines the first time domain waveform of the second acoustic pulse signal, determines the second time domain waveform of the third acoustic pulse signal, determines the first frequency domain waveform of the second acoustic pulse signal, determines the second frequency domain waveform of the third acoustic pulse signal, inputs the first time domain waveform and the second time domain waveform into the first convolutional neural network model for feature extraction to obtain an image feature vector, inputs the first frequency domain waveform and the second frequency domain waveform into the second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks that share weights, and the image feature vector and the voiceprint feature vector are spliced ​​to obtain To the target feature vector, the target feature vector is input into the classification network to obtain the target classification result. First, considering that the sound characteristics of the audible sound band and the ultrasonic frequency band are different, the local discharge can be directly divided into the corresponding acoustic pulse signals of the two different frequency bands, the audible sound band and the ultrasonic frequency band, and the corresponding features are extracted from the time domain and frequency domain dimensions for the corresponding acoustic pulse signals of the two different frequency bands, respectively. Second, in the feature extraction process, considering the homology of the signals in the time domain and the frequency domain, the twin neural network is used for feature extraction, which can reduce the difference between the time domain and the frequency domain, making the feature extraction more in line with the characteristics of the signal itself, and can also ensure the deep extraction of signal features. Third, the features of the time domain and the frequency domain after feature extraction are spliced, and finally input into the classification network for classification, thereby improving the recognition accuracy of the local discharge category.

[0087] In accordance with the above embodiment, please refer to Figure 3 , Figure 31 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. As shown in the figure, the electronic device includes a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor. In the embodiment of the present application, the program includes instructions for executing the following steps: Capturing the first acoustic pulse signal generated by partial discharge through a sensor; Performing frequency band extraction on the first acoustic pulse signal to obtain a second acoustic pulse signal corresponding to an audible sound frequency band and a third acoustic pulse signal corresponding to an ultrasonic frequency band; Determine a first time domain waveform of the second acoustic pulse signal, and determine a second time domain waveform of the third acoustic pulse signal; Determine a first frequency domain waveform of the second acoustic pulse signal, and determine a second frequency domain waveform of the third acoustic pulse signal; Inputting the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector; Inputting the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights; Concatenate the image feature vector and the voiceprint feature vector to obtain a target feature vector; The target feature vector is input into the classification network to obtain the target classification result.

[0088] Optionally, in the aspect of inputting the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector, the program includes instructions for executing the following steps: Determine a first PRPD spectrum according to the first time domain waveform; determining a second PRPD spectrum according to the second time domain waveform; Determining a target PRPD spectrum according to the first PRPD spectrum and the second PRPD spectrum; The target PRPD atlas is input into the first convolutional neural network model for feature extraction to obtain the image feature vector.

[0089] Optionally, in the aspect of inputting the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector, the program includes instructions for executing the following steps: Performing a first noise reduction process on the first frequency domain waveform to obtain a target first frequency domain wave; Performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform; Determine a first spectrogram according to the target first frequency domain waveform; determining a second spectrogram according to the second frequency domain waveform; Determine a target spectrogram according to the first spectrogram and the second spectrogram; The target spectrogram is input into the second neural network model for feature extraction to obtain the voiceprint feature vector.

[0090] Optionally, in the aspect of performing a first noise reduction process on the first frequency domain waveform to obtain a target first frequency domain waveform, the program includes instructions for executing the following steps: determining a first energy value of the first frequency domain waveform; determining a second energy value of the second time domain waveform; determining a first relative deviation between the first energy value and the second energy value; determining a first signal-to-noise ratio of the first frequency domain waveform; determining a second signal-to-noise ratio of the second frequency domain waveform; determining a second relative deviation between the first signal-to-noise ratio and the second signal-to-noise ratio; determining a first absolute value of a difference between the first relative deviation and the second relative deviation; determining a first noise reduction processing algorithm corresponding to the first absolute value; Acquire a noise reduction algorithm control parameter set corresponding to the first noise reduction processing algorithm, wherein the noise reduction algorithm control parameter set includes a plurality of noise reduction algorithm control parameters, and each noise reduction algorithm control parameter corresponds to a signal-to-noise ratio; Acquire a first noise reduction algorithm control parameter corresponding to the first signal-to-noise ratio from the noise reduction algorithm control parameter set; The first frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the first noise reduction algorithm control parameters to obtain the target first frequency domain waveform.

[0091] Optionally, in the aspect of performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform, the program includes instructions for executing the following steps: Determine a third signal-to-noise ratio corresponding to the target first frequency domain wave; determining a target deviation between the third signal-to-noise ratio and the first signal-to-noise ratio; Determining a target feedback adjustment parameter according to the target deviation and the second relative deviation; Acquire a second noise reduction algorithm control parameter corresponding to the second signal-to-noise ratio from the noise reduction algorithm control parameter set; Performing feedback adjustment on the second noise reduction algorithm control parameter according to the target feedback adjustment parameter to obtain the target second noise reduction algorithm control parameter; The second frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the target second noise reduction algorithm control parameters to obtain the target second frequency domain waveform.

[0092] Optionally, in the aspect of determining the target feedback adjustment parameter according to the third relative deviation and the second relative deviation, the program includes instructions for executing the following steps: determining a target signal-to-noise ratio level corresponding to the first signal-to-noise ratio; Acquire a historical enhancement effect parameter set corresponding to the first noise reduction processing algorithm according to the target signal-to-noise ratio level, wherein the historical enhancement effect parameter set includes a plurality of historical enhancement effect parameters, and each historical enhancement effect parameter is used to represent a signal-to-noise ratio improvement ratio; selecting a maximum value and a minimum value among the plurality of historical enhancement effect parameters, dividing the plurality of historical enhancement effect parameters into a plurality of interval segments according to the maximum value and the minimum value, and counting the number of historical enhancement effect parameters in each of the plurality of interval segments to obtain a plurality of numbers; Determine a target interval segment among the multiple interval segments corresponding to the target deviation, and obtain a target quantity and a target interval position corresponding to the target interval segment; determining a first ratio between the target quantity and the sum of the plurality of quantities; determining a second ratio between the target interval position and the plurality of quantities; determining a first feedback adjustment parameter corresponding to the second relative deviation; determining a first standard deviation of the plurality of historical enhancement effect parameters; determining a first optimization parameter corresponding to the first standard deviation; determining a third relative deviation between the first ratio and the second ratio; determining a second optimization parameter corresponding to the third relative deviation; The first feedback adjustment parameter is optimized according to the first optimization parameter and the second optimization parameter to obtain the target feedback adjustment parameter.

[0093] It can be seen that the electronic device described in the embodiment of the present application captures the first acoustic pulse signal generated by partial discharge through a sensor, performs frequency band extraction on the first acoustic pulse signal, obtains a second acoustic pulse signal corresponding to the audible sound frequency band and a third acoustic pulse signal corresponding to the ultrasonic frequency band, determines the first time domain waveform of the second acoustic pulse signal, determines the second time domain waveform of the third acoustic pulse signal, determines the first frequency domain waveform of the second acoustic pulse signal, determines the second frequency domain waveform of the third acoustic pulse signal, inputs the first time domain waveform and the second time domain waveform into the first convolutional neural network model for feature extraction to obtain an image feature vector, inputs the first frequency domain waveform and the second frequency domain waveform into the second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks that share weights, and the image feature vector and the voiceprint feature vector are spliced ​​to obtain a target feature vector. The target feature vector is input into the classification network to obtain the target classification result. First, considering the different sound characteristics of the audible sound band and the ultrasonic frequency band, the local discharge can be directly divided into the corresponding acoustic pulse signals of the audible sound band and the ultrasonic frequency band, and the corresponding features are extracted from the time domain and frequency domain dimensions for the corresponding acoustic pulse signals of the audible sound band and the ultrasonic frequency band. Second, in the feature extraction process, considering the homology of the signals in the time domain and the frequency domain, the twin neural network is used for feature extraction, which can reduce the difference between the time domain and the frequency domain, making the feature extraction more in line with the characteristics of the signal itself, and can also ensure the deep extraction of signal features. Third, the features of the time domain and the frequency domain after feature extraction are spliced, and finally input into the classification network for classification, thereby improving the recognition accuracy of the local discharge category.

[0094] Figure 4 4 is a functional unit block diagram of a signal processing device 400 for partial discharge signals involved in an embodiment of the present application. The signal processing device 400 for partial discharge signals includes: an acquisition unit 401, an extraction unit 402, a determination unit 403, a splicing unit 404 and a classification unit 405, wherein: The acquisition unit 401 is used to capture the first acoustic pulse signal generated by partial discharge through a sensor; The extraction unit 402 is used to perform frequency band extraction on the first acoustic pulse signal to obtain a second acoustic pulse signal corresponding to an audible sound frequency band and a third acoustic pulse signal corresponding to an ultrasonic frequency band; The determining unit 403 is configured to determine a first time domain waveform of the second acoustic pulse signal and a second time domain waveform of the third acoustic pulse signal; determine a first frequency domain waveform of the second acoustic pulse signal and a second frequency domain waveform of the third acoustic pulse signal; The extraction unit 402 is further configured to input the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector; input the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks that share weights; The concatenation unit 404 is configured to concatenate the image feature vector and the voiceprint feature vector to obtain a target feature vector; The classification unit 405 is used to input the target feature vector into a classification network to obtain a target classification result.

[0095] Optionally, in the aspect of inputting the first time domain waveform and the second time domain waveform into the first convolutional neural network model for feature extraction to obtain the image feature vector, the extraction unit 402 is specifically used for: Determine a first PRPD spectrum according to the first time domain waveform; determining a second PRPD spectrum according to the second time domain waveform; Determining a target PRPD spectrum according to the first PRPD spectrum and the second PRPD spectrum; The target PRPD atlas is input into the first convolutional neural network model for feature extraction to obtain the image feature vector.

[0096] Optionally, in the aspect of inputting the first frequency domain waveform and the second frequency domain waveform into the second neural network model for feature extraction to obtain the voiceprint feature vector, the extraction unit 402 is specifically used for: Performing a first noise reduction process on the first frequency domain waveform to obtain a target first frequency domain wave; Performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform; Determine a first spectrogram according to the target first frequency domain waveform; determining a second spectrogram according to the second frequency domain waveform; Determine a target spectrogram according to the first spectrogram and the second spectrogram; The target spectrogram is input into the second neural network model for feature extraction to obtain the voiceprint feature vector.

[0097] Optionally, in the aspect of performing the first noise reduction process on the first frequency domain waveform to obtain the target first frequency domain waveform, the extraction unit 402 is specifically used to: determining a first energy value of the first frequency domain waveform; determining a second energy value of the second time domain waveform; determining a first relative deviation between the first energy value and the second energy value; determining a first signal-to-noise ratio of the first frequency domain waveform; determining a second signal-to-noise ratio of the second frequency domain waveform; determining a second relative deviation between the first signal-to-noise ratio and the second signal-to-noise ratio; determining a first absolute value of a difference between the first relative deviation and the second relative deviation; determining a first noise reduction processing algorithm corresponding to the first absolute value; Acquire a noise reduction algorithm control parameter set corresponding to the first noise reduction processing algorithm, wherein the noise reduction algorithm control parameter set includes a plurality of noise reduction algorithm control parameters, and each noise reduction algorithm control parameter corresponds to a signal-to-noise ratio; Acquire a first noise reduction algorithm control parameter corresponding to the first signal-to-noise ratio from the noise reduction algorithm control parameter set; The first frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the first noise reduction algorithm control parameters to obtain the target first frequency domain waveform.

[0098] Optionally, in the aspect of performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform, the extraction unit 402 is specifically used to: Determine a third signal-to-noise ratio corresponding to the target first frequency domain wave; determining a target deviation between the third signal-to-noise ratio and the first signal-to-noise ratio; Determining a target feedback adjustment parameter according to the target deviation and the second relative deviation; Acquire a second noise reduction algorithm control parameter corresponding to the second signal-to-noise ratio from the noise reduction algorithm control parameter set; Performing feedback adjustment on the second noise reduction algorithm control parameter according to the target feedback adjustment parameter to obtain the target second noise reduction algorithm control parameter; The second frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the target second noise reduction algorithm control parameters to obtain the target second frequency domain waveform.

[0099] Optionally, in determining the target feedback adjustment parameter according to the third relative deviation and the second relative deviation, the extraction unit 402 is specifically configured to: determining a target signal-to-noise ratio level corresponding to the first signal-to-noise ratio; Acquire a historical enhancement effect parameter set corresponding to the first noise reduction processing algorithm according to the target signal-to-noise ratio level, wherein the historical enhancement effect parameter set includes a plurality of historical enhancement effect parameters, and each historical enhancement effect parameter is used to represent a signal-to-noise ratio improvement ratio; selecting a maximum value and a minimum value among the plurality of historical enhancement effect parameters, dividing the plurality of historical enhancement effect parameters into a plurality of interval segments according to the maximum value and the minimum value, and counting the number of historical enhancement effect parameters in each of the plurality of interval segments to obtain a plurality of numbers; Determine a target interval segment among the multiple interval segments corresponding to the target deviation, and obtain a target quantity and a target interval position corresponding to the target interval segment; determining a first ratio between the target quantity and the sum of the plurality of quantities; determining a second ratio between the target interval position and the plurality of quantities; determining a first feedback adjustment parameter corresponding to the second relative deviation; determining a first standard deviation of the plurality of historical enhancement effect parameters; determining a first optimization parameter corresponding to the first standard deviation; determining a third relative deviation between the first ratio and the second ratio; determining a second optimization parameter corresponding to the third relative deviation; The first feedback adjustment parameter is optimized according to the first optimization parameter and the second optimization parameter to obtain the target feedback adjustment parameter.

[0100] It can be seen that the signal processing device for partial discharge signals described in the embodiment of the present application captures the first acoustic pulse signal generated by partial discharge through a sensor, performs frequency band extraction on the first acoustic pulse signal, obtains a second acoustic pulse signal corresponding to the audible sound frequency band and a third acoustic pulse signal corresponding to the ultrasonic frequency band, determines the first time domain waveform of the second acoustic pulse signal, determines the second time domain waveform of the third acoustic pulse signal, determines the first frequency domain waveform of the second acoustic pulse signal, determines the second frequency domain waveform of the third acoustic pulse signal, inputs the first time domain waveform and the second time domain waveform into the first convolutional neural network model for feature extraction to obtain an image feature vector, inputs the first frequency domain waveform and the second frequency domain waveform into the second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights, and the image feature vector and the voiceprint feature vector are spliced ​​to obtain To the target feature vector, the target feature vector is input into the classification network to obtain the target classification result. First, considering that the sound characteristics of the audible sound band and the ultrasonic frequency band are different, the local discharge can be directly divided into the corresponding acoustic pulse signals of the two different frequency bands, the audible sound band and the ultrasonic frequency band, and the corresponding features are extracted from the time domain and frequency domain dimensions for the corresponding acoustic pulse signals of the two different frequency bands, respectively. Second, in the feature extraction process, considering the homology of the signals in the time domain and the frequency domain, the twin neural network is used for feature extraction, which can reduce the difference between the time domain and the frequency domain, making the feature extraction more in line with the characteristics of the signal itself, and can also ensure the deep extraction of signal features. Third, the features of the time domain and the frequency domain after feature extraction are spliced, and finally input into the classification network for classification, thereby improving the recognition accuracy of the local discharge category.

[0101] It can be understood that the functions of each program module of the signal processing device for partial discharge signals in this embodiment can be specifically implemented according to the method in the above method embodiment. The specific implementation process can refer to the relevant description of the above method embodiment, which will not be repeated here.

[0102] An embodiment of the present application also provides a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps of any method described in the above method embodiments, and the above computer includes an electronic device.

[0103] The embodiment of the present application also provides a computer program product, the computer program product includes a non-transitory computer-readable storage medium storing a computer program, the computer program is operable to cause a computer to execute some or all of the steps of any method described in the method embodiment. The computer program product may be a software installation package, and the computer includes an electronic device.

[0104] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0105] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0106] In the several embodiments provided in the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the above-mentioned units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0107] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0108] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0109] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the above-mentioned methods in each embodiment of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0110] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable memory, which may include a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0111] The embodiments of the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A signal processing method for partial discharge signals, characterized in that: The method comprises: Capturing the first acoustic pulse signal generated by partial discharge through a sensor; Performing frequency band extraction on the first acoustic pulse signal to obtain a second acoustic pulse signal corresponding to an audible sound frequency band and a third acoustic pulse signal corresponding to an ultrasonic frequency band; Determine a first time domain waveform of the second acoustic pulse signal, and determine a second time domain waveform of the third acoustic pulse signal; Determine a first frequency domain waveform of the second acoustic pulse signal, and determine a second frequency domain waveform of the third acoustic pulse signal; Inputting the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector; Inputting the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights; Concatenate the image feature vector and the voiceprint feature vector to obtain a target feature vector; The target feature vector is input into the classification network to obtain the target classification result.

2. The method according to claim 1, characterized in that The step of inputting the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector comprises: Determine a first PRPD spectrum according to the first time domain waveform; determining a second PRPD spectrum according to the second time domain waveform; Determining a target PRPD spectrum according to the first PRPD spectrum and the second PRPD spectrum; The target PRPD atlas is input into the first convolutional neural network model for feature extraction to obtain the image feature vector.

3. The method according to claim 2, characterized in that The step of inputting the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector includes: Performing a first noise reduction process on the first frequency domain waveform to obtain a target first frequency domain wave; Performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform; Determine a first spectrogram according to the target first frequency domain waveform; determining a second spectrogram according to the second frequency domain waveform; Determine a target spectrogram according to the first spectrogram and the second spectrogram; The target spectrogram is input into the second neural network model for feature extraction to obtain the voiceprint feature vector.

4. The method according to claim 3, characterized in that The step of performing a first noise reduction process on the first frequency domain waveform to obtain a target first frequency domain waveform includes: determining a first energy value of the first frequency domain waveform; determining a second energy value of the second time domain waveform; determining a first relative deviation between the first energy value and the second energy value; determining a first signal-to-noise ratio of the first frequency domain waveform; determining a second signal-to-noise ratio of the second frequency domain waveform; determining a second relative deviation between the first signal-to-noise ratio and the second signal-to-noise ratio; determining a first absolute value of a difference between the first relative deviation and the second relative deviation; determining a first noise reduction processing algorithm corresponding to the first absolute value; Acquire a noise reduction algorithm control parameter set corresponding to the first noise reduction processing algorithm, wherein the noise reduction algorithm control parameter set includes a plurality of noise reduction algorithm control parameters, and each noise reduction algorithm control parameter corresponds to a signal-to-noise ratio; Acquire a first noise reduction algorithm control parameter corresponding to the first signal-to-noise ratio from the noise reduction algorithm control parameter set; The first frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the first noise reduction algorithm control parameters to obtain the target first frequency domain waveform.

5. The method according to claim 4, characterized in that The step of performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform includes: Determine a third signal-to-noise ratio corresponding to the target first frequency domain wave; determining a target deviation between the third signal-to-noise ratio and the first signal-to-noise ratio; Determining a target feedback adjustment parameter according to the target deviation and the second relative deviation; Acquire a second noise reduction algorithm control parameter corresponding to the second signal-to-noise ratio from the noise reduction algorithm control parameter set; Performing feedback adjustment on the second noise reduction algorithm control parameter according to the target feedback adjustment parameter to obtain the target second noise reduction algorithm control parameter; The second frequency domain waveform is subjected to noise reduction processing according to the first noise reduction processing algorithm and the target second noise reduction algorithm control parameters to obtain the target second frequency domain waveform.

6. A signal processing device for partial discharge signals, characterized in that: The device comprises: an acquisition unit, an extraction unit, a determination unit, a splicing unit and a classification unit, wherein: The acquisition unit is used to capture the first acoustic pulse signal generated by the partial discharge through a sensor; The extraction unit is used to perform frequency band extraction on the first acoustic pulse signal to obtain a second acoustic pulse signal corresponding to an audible sound frequency band and a third acoustic pulse signal corresponding to an ultrasonic frequency band; The determining unit is used to determine a first time domain waveform of the second acoustic pulse signal and a second time domain waveform of the third acoustic pulse signal; determine a first frequency domain waveform of the second acoustic pulse signal and a second frequency domain waveform of the third acoustic pulse signal; The extraction unit is further used to input the first time domain waveform and the second time domain waveform into a first convolutional neural network model for feature extraction to obtain an image feature vector; input the first frequency domain waveform and the second frequency domain waveform into a second neural network model for feature extraction to obtain a voiceprint feature vector; the first convolutional neural network and the second convolutional neural network are twin neural networks with shared weights; The splicing unit is used to splice the image feature vector and the voiceprint feature vector to obtain a target feature vector; The classification unit is used to input the target feature vector into a classification network to obtain a target classification result.

7. The device according to claim 6, characterized in that In the aspect of inputting the first time domain waveform and the second time domain waveform into the first convolutional neural network model for feature extraction to obtain the image feature vector, the extraction unit is specifically used for: Determine a first PRPD spectrum according to the first time domain waveform; determining a second PRPD spectrum according to the second time domain waveform; Determining a target PRPD spectrum according to the first PRPD spectrum and the second PRPD spectrum; The target PRPD atlas is input into the first convolutional neural network model for feature extraction to obtain the image feature vector.

8. The device according to claim 7, characterized in that In the aspect of inputting the first frequency domain waveform and the second frequency domain waveform into the second neural network model for feature extraction to obtain the voiceprint feature vector, the extraction unit is specifically used for: Performing a first noise reduction process on the first frequency domain waveform to obtain a target first frequency domain wave; Performing a second noise reduction process on the second frequency domain waveform to obtain a target second frequency domain waveform; Determine a first spectrogram according to the target first frequency domain waveform; determining a second spectrogram according to the second frequency domain waveform; Determine a target spectrogram according to the first spectrogram and the second spectrogram; The target spectrogram is input into the second neural network model for feature extraction to obtain the voiceprint feature vector.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store one or more programs and is configured to be executed by the processor, wherein the program comprises instructions for executing the steps in the method according to any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that: A computer program for electronic data exchange is stored, wherein the computer program enables a computer to execute the method according to any one of claims 1 to 5.

Images