FPGA-based partial discharge voiceprint real-time recognition and classification integrated circuit processing method and system

Abstract

The application provides an FPGA-based partial discharge voiceprint real-time recognition and classification integrated circuit processing method and system, and relates to the technical field of power equipment monitoring.The method comprises the following steps: receiving original voiceprint signals of a high-frequency ultrasonic voiceprint sensor array installed on the side walls of GIS chambers; performing pre-conditioning on the original voiceprint signals in an FPGA to obtain conditioned voiceprint signals; performing adaptive noise reduction processing on the conditioned voiceprint signals to obtain noise-reduced voiceprint signals; extracting time-domain envelopes, spectral energy distributions and pulse waveform parameters from the noise-reduced voiceprint signals to obtain initial voiceprint feature vectors of partial discharge characteristics; and constructing a virtual space reference configuration of GIS chamber geometrical structures and sound wave propagation paths by using a parameterized modeling method of spatial pose gradient projection, in combination with the spatial coordinates of each sensor and the corresponding chamber volume parameters, through the initial feature vectors. The application improves the accuracy and real-time performance of GIS partial discharge monitoring.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, and in particular to an integrated circuit processing method and system for real-time identification and classification of partial discharge acoustic prints based on FPGA. Background Technology

[0002] In power systems, gas-insulated metal-enclosed switchgear (GIS) is a core power transmission and distribution equipment. Partial discharge caused by internal insulation defects is a key factor leading to equipment insulation breakdown and power supply interruption. Real-time identification and classification of partial discharge acoustic patterns has become a core requirement for GIS equipment condition monitoring. However, existing partial discharge acoustic pattern processing technologies mostly adopt an architecture that combines a general-purpose processor with a conventional ultrasonic detection module, which has technical shortcomings in practical engineering applications.

[0003] During online monitoring of the GIS equipment group in a 500kV substation, the traditional processing system received the raw acoustic signature signals from the ultrasonic sensor array on the sidewall of the air chamber. Due to the superposition of mechanical vibration noise and electromagnetic interference from the transformers and reactors on site, the background noise could not be effectively suppressed by a fixed filtering method alone. At the same time, the extracted acoustic signature features were not corrected by combining the spatial coordinates of each sensor, the geometry of the GIS air chamber, and the acoustic attenuation and reflection characteristics of the shell material. This resulted in serious distortion of the initial feature vector. The serial processing mode of the general-purpose processor could not achieve real-time analysis of the acoustic signature signals. Furthermore, the classifier used a fixed discrimination threshold, which misclassified air gap discharge as floating potential discharge during the identification process. The fault level judgment was significantly inaccurate. The staff failed to take timely maintenance measures, which ultimately led to the accelerated deterioration of the insulation of the GIS equipment air chamber and caused a partial power outage in the substation.

[0004] This case exposes the technical defects of traditional partial discharge acoustic signature processing technology, such as weak anti-interference ability, distortion of feature extraction due to lack of integration of spatial and acoustic propagation characteristics of GIS equipment, insufficient real-time signal processing, and lack of dynamic calibration for classification and discrimination. It is no longer suitable for the engineering requirements of power systems for real-time and accurate monitoring of the partial discharge status of GIS equipment. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an integrated circuit processing method and system for real-time identification and classification of partial discharge acoustic prints based on FPGA, so as to improve the accuracy and real-time performance of GIS partial discharge monitoring.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for real-time identification and classification of partial discharge acoustic prints based on FPGA integrated circuit processing, the method comprising: The system receives the raw acoustic signature signal from the high-frequency ultrasonic acoustic signature sensor array installed on the side wall of each air chamber of the GIS; and pre-conditions the raw acoustic signature signal inside the FPGA to obtain the conditioned acoustic signature signal. Adaptive noise reduction is performed on the conditioned voiceprint signal to obtain the noise-reduced voiceprint signal; the time-domain envelope, spectral energy distribution and pulse waveform parameters are extracted from the noise-reduced voiceprint signal to obtain the initial voiceprint feature vector of partial discharge characteristics. Using initial feature vectors, combined with the spatial coordinates of each sensor and the corresponding air chamber volume parameters, a parametric modeling method based on spatial pose gradient projection is used to construct a virtual spatial reference configuration of the GIS air chamber geometry and sound wave propagation path; and the virtual spatial reference configuration is divided into gridded regions to obtain the divided gridded regions. The geometric constraints of the shell material within the gridded area are analyzed, and the attenuation and reflection characteristics of sound waves in the material are integrated to generate a set of spatial propagation compensation coefficients. The initial acoustic feature vector is spatially normalized and corrected using the set of spatial propagation compensation coefficients to obtain the corrected acoustic feature vector. The corrected voiceprint feature vector is input into the parallel matching and discrimination logic embedded in the FPGA, and the discrimination threshold of the classifier is dynamically calibrated according to the statistical voiceprint feature distribution to obtain the classification result. Based on the classification results and the magnitude of the corrected acoustic signature feature vector, the fault level of partial discharge is determined; the classification results and fault level are encapsulated to achieve real-time identification and classification of partial discharge acoustic signatures.

[0007] Furthermore, the system receives the raw acoustic signature signals from the high-frequency ultrasonic acoustic signature sensor array installed on the side walls of each air chamber of the GIS; and pre-conditions the raw acoustic signature signals within the FPGA to obtain conditioned acoustic signature signals, including: The received raw voiceprint signal is amplified by programmable gain control to obtain an amplitude-adjusted analog voiceprint signal; The amplitude-adjusted analog voiceprint signal is subjected to anti-aliasing filtering to remove high-frequency interference, resulting in a filtered analog voiceprint signal. The filtered analog voiceprint signal is converted into a digital voiceprint signal through an analog-to-digital converter, and the digital voiceprint signal is used as the conditioned voiceprint signal.

[0008] Furthermore, adaptive noise reduction processing is performed on the conditioned voiceprint signal to obtain the noise-reduced voiceprint signal; from the noise-reduced voiceprint signal, the time-domain envelope, spectral energy distribution, and pulse waveform parameters are extracted to obtain the initial voiceprint feature vector of partial discharge characteristics, including: The digital voiceprint signal is input into an adaptive filter, and the background noise in the signal is dynamically tracked and suppressed through an iterative algorithm to obtain the noise-reduced voiceprint signal. The noise-reduced voiceprint signal is subjected to Hilbert transform to extract the time-domain envelope parameter that characterizes the signal amplitude variation trend; the noise-reduced voiceprint signal is subjected to fast Fourier transform to obtain the spectral energy distribution parameters of each frequency band; the noise-reduced voiceprint signal is subjected to pulse detection to extract pulse waveform parameters including pulse amplitude, rise time and decay constant. The extracted time-domain envelope parameters, spectral energy distribution parameters, and pulse waveform parameters are fused to form an initial acoustic signature feature vector describing the characteristics of partial discharge.

[0009] Furthermore, using initial feature vectors and combining the spatial coordinates of each sensor with the corresponding air chamber volume parameters, a parametric modeling method based on spatial pose gradient projection is employed to construct a virtual spatial reference configuration of the GIS air chamber geometry and sound wave propagation path. The virtual spatial reference configuration is then divided into gridded regions, resulting in the following gridded regions: Obtain the pre-stored spatial coordinate dataset of each high-frequency ultrasonic acoustic sensor and the corresponding volume parameters of the air chamber; The initial acoustic feature vector is fused with the spatial coordinate dataset and volume parameters. Based on the spatial geometric relationship between the sound source point and each sensor determined after fusion, the three-dimensional geometric structure of the GIS air chamber and the path trajectory of the sound wave from the sound source point to each sensor are constructed by calculating the projection of the spatial gradient, thus obtaining the virtual space reference configuration. The virtual space reference configuration is divided into finite element meshes according to a preset spatial resolution to obtain multiple continuous meshed regions.

[0010] Furthermore, the geometric constraints of the shell material within the meshed region are analyzed, and the attenuation and reflection characteristics of sound waves in the material are integrated to generate a set of spatial propagation compensation coefficients. The initial acoustic signature feature vector is then spatially normalized using this set of coefficients to obtain the corrected acoustic signature feature vector, which includes: The geometric constraints of the shell material in each region of the meshed area are analyzed to obtain the attenuation and reflection parameters of the sound wave when it propagates in the corresponding shell material; By associating and fusing the attenuation parameters and reflection parameters with the spatial location of each gridded region, a set of spatial propagation compensation coefficients corresponding to each gridded region is obtained; Based on the spatial propagation compensation coefficient set, the initial voiceprint feature vector is spatially normalized to eliminate the signal attenuation caused by the difference in propagation path and shell material, thus obtaining the corrected voiceprint feature vector.

[0011] Furthermore, the corrected voiceprint feature vector is input into the parallel matching and discrimination logic embedded in the FPGA, and the discrimination threshold of the classifier is dynamically calibrated according to the statistical voiceprint feature distribution to obtain the classification result, including: The corrected voiceprint feature vector is input into the parallel matching and discrimination logic embedded in the FPGA. The initial matching result is obtained by performing parallel matching calculations with pre-stored templates of various partial discharge types. The distribution characteristics of the historically corrected voiceprint feature vectors are statistically analyzed, and the discrimination thresholds corresponding to each category in the parallel matching and discrimination logic are dynamically adjusted based on the distribution characteristics to obtain the adjusted discrimination thresholds. The adjusted discrimination threshold is applied to the matching calculation process of the parallel matching discrimination logic to calibrate the initial matching results and obtain the classification results.

[0012] Furthermore, based on the classification results and the amplitude of the corrected acoustic signature feature vector, the fault level of partial discharge is determined; the classification results and fault level are encapsulated and uploaded to the station control platform via a communication interface, including: Based on at least one classification result among corona discharge, floating potential discharge, surface discharge, and air gap discharge, combined with the amplitude parameter in the corrected acoustic signature feature vector, and in accordance with the preset fault level judgment rules, the fault level of the current partial discharge is determined. The classification results and the determined fault levels are encapsulated according to a preset data frame format to generate an upload data frame; The uploaded data frames are sent to the station control platform through the communication interface to complete the real-time identification, classification, and reporting of fault level determination results of partial discharge acoustic patterns.

[0013] Secondly, an FPGA-based real-time partial discharge acoustic signature recognition and classification integrated circuit processing system includes: The acquisition module is used to receive the raw acoustic fingerprint signals from the high-frequency ultrasonic acoustic fingerprint sensor array installed on the side wall of each air chamber of the GIS; and to pre-condition the raw acoustic fingerprint signals inside the FPGA to obtain the conditioned acoustic fingerprint signals. The noise reduction module is used to perform adaptive noise reduction processing on the conditioned voiceprint signal to obtain the noise-reduced voiceprint signal; the time domain envelope, spectral energy distribution and pulse waveform parameters are extracted from the noise-reduced voiceprint signal to obtain the initial voiceprint feature vector of partial discharge characteristics. The projection module is used to construct a virtual spatial reference configuration of the GIS air cell geometry and sound wave propagation path by using an initial feature vector, combined with the spatial coordinates of each sensor and the corresponding air cell volume parameters, and adopting a parametric modeling method of spatial pose gradient projection; and to divide the virtual spatial reference configuration into gridded regions to obtain the divided gridded regions. The correction module is used to analyze the geometric constraints of the shell material within the meshed area, integrate the attenuation and reflection characteristics of sound waves in the material, and generate a set of spatial propagation compensation coefficients. The initial soundprint feature vector is spatially normalized and corrected using the set of spatial propagation compensation coefficients to obtain the corrected soundprint feature vector. The calibration module is used to input the corrected voiceprint feature vector into the parallel matching and discrimination logic embedded in the FPGA, and dynamically calibrate the discrimination threshold of the classifier according to the statistical voiceprint feature distribution to obtain the classification result. The processing module is used to determine the fault level of partial discharge based on the classification results and the magnitude of the corrected acoustic signature feature vector; it encapsulates the classification results and fault level into data and uploads them to the station control platform through the communication interface.

[0014] Thirdly, a computing device, comprising: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0015] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0016] The above-described solution of the present invention has at least the following beneficial effects: This invention employs an integrated circuit processing architecture centered on an FPGA, combining programmable gain amplification, anti-aliasing filtering pre-conditioning, adaptive noise reduction algorithms, and multi-dimensional acoustic signature feature extraction. It also incorporates parametric modeling of spatial pose gradient projection, gridded region division, and spatial propagation compensation coefficient set feature correction technology that integrates the attenuation and reflection characteristics of the shell material. Furthermore, it utilizes parallel matching and discrimination logic embedded within the FPGA and a dynamic threshold calibration method for a classifier based on acoustic signature feature distribution. This integrated processing approach, combining partial discharge classification results and feature vector amplitudes to determine fault levels and complete data encapsulation, effectively overcomes the weaknesses of traditional GIS partial discharge acoustic signature processing technologies, such as weak anti-interference capabilities and limited acoustic signature features. This paper addresses technical issues such as feature distortion caused by neglecting the spatial geometry and acoustic wave propagation characteristics of GIS air chambers, insufficient real-time performance of serial signal processing, misjudgment due to fixed classification thresholds, and lack of accurate basis for fault grading. It achieves integrated real-time processing of GIS partial discharge acoustic signature signals from acquisition to identification, classification, and fault grading via integrated circuits. This improves the accuracy of acoustic signature feature extraction and the precision of partial discharge type classification, enabling accurate determination of partial discharge fault levels and standardized encapsulation and reporting of results. It achieves real-time and accurate identification and classification of partial discharge acoustic signatures, providing reliable technical support for partial discharge status monitoring and maintenance decisions of GIS equipment, and effectively ensuring the stable operation of GIS equipment. Attached Figure Description

[0017] Figure 1This is a flowchart illustrating the FPGA-based real-time partial discharge acoustic signature recognition and classification integrated circuit processing method provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of an FPGA-based real-time partial discharge acoustic print recognition and classification integrated circuit processing system provided in an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0020] like Figure 1 As shown, embodiments of the present invention propose a real-time partial discharge acoustic print recognition and classification integrated circuit processing method based on FPGA, the method comprising the following steps: Step 1: Receive the raw acoustic signature signal from the high-frequency ultrasonic acoustic signature sensor array installed on the side wall of each air chamber of the GIS; pre-condition the raw acoustic signature signal inside the FPGA to obtain the conditioned acoustic signature signal. Step 2: Perform adaptive noise reduction on the conditioned voiceprint signal to obtain the noise-reduced voiceprint signal; extract the time-domain envelope, spectral energy distribution and pulse waveform parameters from the noise-reduced voiceprint signal to obtain the initial voiceprint feature vector of partial discharge characteristics. Step 3: Using the initial feature vector, combined with the spatial coordinates of each sensor and the corresponding air chamber volume parameters, a parametric modeling method of spatial pose gradient projection is used to construct a virtual spatial reference configuration of the GIS air chamber geometry and sound wave propagation path; and the virtual spatial reference configuration is divided into gridded regions to obtain the divided gridded regions. Step 4: Analyze the geometric constraints of the shell material within the meshed area, integrate the attenuation and reflection characteristics of sound waves in the material, and generate a set of spatial propagation compensation coefficients; use the set of spatial propagation compensation coefficients to perform spatial normalization correction on the initial voiceprint feature vector to obtain the corrected voiceprint feature vector. Step 5: Input the corrected voiceprint feature vector into the parallel matching and discrimination logic embedded in the FPGA, and dynamically calibrate the discrimination threshold of the classifier according to the statistical voiceprint feature distribution to obtain the classification result; Step 6: Based on the classification results and the magnitude of the corrected acoustic signature feature vector, determine the fault level of partial discharge; encapsulate the classification results and fault level into data to achieve real-time identification and classification of partial discharge acoustic signatures.

[0021] In this embodiment of the invention, the present invention employs an integrated circuit processing architecture with FPGA as its core, sequentially implementing FPGA preconditioning of acoustic signature signals, adaptive noise reduction and multi-dimensional partial discharge feature extraction, GIS virtual modeling and meshing of the air chamber based on spatial pose gradient projection, feature vector spatial normalization correction incorporating the acoustic wave propagation characteristics of the shell material, FPGA-based parallel matching discrimination and classifier dynamic threshold calibration, and an integrated technical approach combining partial discharge classification results and feature vector amplitude for fault level determination and data encapsulation. Therefore, it effectively overcomes the shortcomings of traditional GIS partial discharge acoustic signature processing technologies, such as weak on-site interference suppression and failure to consider acoustic signature feature extraction. The technical challenges of feature distortion caused by the spatial geometry and acoustic wave propagation characteristics of equipment, insufficient real-time signal processing modes, misjudgment due to fixed classification thresholds, and lack of precise quantitative basis for fault classification have been addressed. This solution achieves real-time hardware-level processing of GIS partial discharge acoustic signature signals from acquisition to identification, classification, and fault classification. It improves the accuracy of partial discharge acoustic signature feature extraction and the precision of partial discharge type classification, reliably determines the level of partial discharge faults, and completes standardized result encapsulation. This enables real-time and accurate identification and classification of partial discharge acoustic signatures, providing reliable technical support for GIS equipment partial discharge status monitoring, fault early warning, and operation and maintenance decisions, and ensuring the stable operation of GIS equipment.

[0022] In a preferred embodiment of the present invention, step 1 above may include: Step 1.1: Perform programmable gain amplification on the received raw voiceprint signal to obtain an amplitude-adjusted analog voiceprint signal. Specifically, this includes: using the programmable gain amplifier integrated within the FPGA as the core execution carrier, firstly, continuously sampling the received raw analog voiceprint signal for instantaneous amplitude, collecting all instantaneous amplitude data of the raw voiceprint signal within a preset sampling period; using each sampling time point as the abscissa and the corresponding collected instantaneous amplitude as the ordinate, generating a series of discrete amplitude coordinate points in a two-dimensional plane; continuously linearly fitting these discrete coordinate points to form a smooth amplitude variation fitting curve; then, extracting a curve segment within the preset sampling period from this fitting curve, and calculating... The area of ​​the closed geometric region enclosed by the curve segment and the horizontal axis is divided by the time length of the preset sampling period to obtain the optimized real-time amplitude of the original voiceprint signal. Then, the system adaptation reference amplitude stored in the FPGA is retrieved, and the real-time gain coefficient is calculated by division. The real-time gain coefficient is calculated by dividing the system adaptation reference amplitude by the real-time gain coefficient of the original voiceprint signal. The instantaneous voltage value corresponding to each time node in the original voiceprint signal is multiplied by the calculated real-time gain coefficient to obtain the amplitude-adjusted voltage value corresponding to each time node. All the adjusted voltage values ​​are continuously output according to the original time sequence to finally obtain the amplitude-adjusted analog voiceprint signal.

[0023] Step 1.2: The amplitude-adjusted analog acoustic signal is subjected to anti-aliasing filtering to remove high-frequency interference, resulting in a filtered analog acoustic signal. This operation is performed using a low-pass anti-aliasing filter integrated within the FPGA. The cutoff frequency of the low-pass anti-aliasing filter is determined based on the sampling frequency set by the analog-to-digital conversion stage. The cutoff frequency is calculated as half of the analog-to-digital conversion sampling frequency. The frequency value of the acoustic signal is used as the horizontal axis, and the corresponding signal amplitude is used as the vertical axis to construct a frequency-amplitude geometric plane in a two-dimensional plane. This plane is divided into equidistant geometric grids according to the frequency values. The determined cutoff frequency is used as the geometric dividing line to divide the entire frequency-amplitude geometric plane into a low-frequency effective signal region and a high-frequency interference signal region. Then, according to the attenuation rule of 20 dB per octave, the high-frequency interference signal is filtered. Within the interference signal region, a geometrically fitted straight line for amplitude attenuation is drawn according to the frequency multiple relationship. Based on this line, the amplitude attenuation thresholds corresponding to different high-frequency bands are determined. The filtering characteristics of the filter are set according to the geometrically divided region and the determined attenuation threshold. The frequency-amplitude coordinates of the amplitude-adjusted simulated acoustic signature signal are mapped one by one to the frequency-amplitude geometric plane. For coordinates falling into the high-frequency interference signal region, amplitude attenuation is performed according to the attenuation threshold determined by the corresponding geometrically fitted straight line. For coordinates falling into the low-frequency effective signal region, the amplitude remains unchanged. In this way, signal components with frequencies higher than the cutoff frequency are attenuated by a fixed proportion, retaining only the high-frequency ultrasonic components in the signal that are within the effective detection range. The signal after geometric region determination and amplitude processing is continuously output to obtain the filtered simulated acoustic signature signal.

[0024] Step 1.3: Convert the filtered analog voiceprint signal into a digital voiceprint signal using an analog-to-digital converter (ADC). Use the digital voiceprint signal as the conditioned voiceprint signal. Specifically, this includes: converting the analog signal to a digital signal using the FPGA's integrated ADC; determining the fixed sampling accuracy and sampling frequency of the ADC; ensuring the sampling frequency strictly follows the Nyquist sampling theorem and is at least twice the highest effective frequency of the partial discharge voiceprint signal; and calculating the quantization interval of the ADC based on the preset reference voltage and the determined sampling accuracy. The quantization interval is calculated by dividing the reference voltage by... The analog-to-digital converter (ADC) continuously samples the filtered analog voiceprint signal at equal time intervals using a fixed sampling frequency with a sampling precision of 2 bits. The analog voltage value corresponding to each sampling point is obtained by dividing the analog voltage value of each sampling point by the quantization interval. All quantization values ​​are then binary-coded to obtain the digital encoded data corresponding to each sampling point. Finally, the digital encoded data of all continuous sampling points are arranged sequentially according to the original time sequence to form a continuous digital voiceprint signal, which is the voiceprint signal after pre-conditioning.

[0025] In this embodiment of the invention, the programmable gain amplification process realizes the dynamic standardization adjustment of the original acoustic signature signal amplitude, effectively solving the problem of uneven and fluctuating original acoustic signature signal amplitude caused by different partial discharge locations, sound wave propagation paths and distances in traditional conditioning methods. By calculating and matching the gain coefficient in real time, the signal amplitude is uniformly adjusted to the adaptation range of the FPGA system, which avoids signal distortion caused by signal amplitude overshoot and prevents the subsequent stages from being unable to effectively identify and extract features due to excessively small signal amplitude, thus ensuring the stability of the processed signal amplitude. The anti-aliasing filtering process realizes the accurate and directional filtering of high-frequency interference in the original signal. By strictly calculating and determining the cutoff frequency and setting a fixed attenuation rule, the signal aliasing distortion problem that is very easy to occur in the subsequent analog-to-digital conversion stage is avoided from the root. While filtering out invalid high-frequency interference, the effective high-frequency ultrasonic features of the partial discharge acoustic signature signal are completely preserved, improving the purity of the acoustic signature signal and allowing the extracted signal features to truly reflect the actual state of partial discharge.

[0026] The analog-to-digital conversion process achieves high-precision conversion from analog to digital voiceprint signals. Through precise calculation of quantization intervals and binary encoding, it preserves the effective information in the analog signal to the greatest extent and avoids information loss during the signal conversion process. The converted digital voiceprint signal is perfectly adapted to the FPGA's digital signal processing architecture, providing a suitable signal form for subsequent fully digital processing stages such as adaptive noise reduction, voiceprint feature extraction, and parallel matching and discrimination. This ensures the smooth operation of the processing flow from a technical perspective. The hardware-level pre-conditioning, which is completed entirely within the FPGA, improves the real-time performance and integration of signal conditioning compared to the traditional method of using external independent conditioning modules combined with processors. It also reduces signal loss, delay, and additional interference during transmission between external modules, ensuring high synchronization and high accuracy of the voiceprint signal throughout the entire process from acquisition to conditioning. The hardware processing method within the FPGA has stronger stability and anti-interference capabilities, adapting to the complex industrial environment of substations and further ensuring the quality of the pre-conditioned signal. This improves the accuracy and reliability of the entire GIS partial discharge voiceprint recognition and classification process from the source of processing.

[0027] In a preferred embodiment of the present invention, step 2 above may include: Step 2.1: Input the digital voiceprint signal into an adaptive filter. An iterative algorithm dynamically tracks and suppresses background noise in the signal to obtain the denoised voiceprint signal. Specifically, this includes: inputting the digital voiceprint signal obtained after pre-conditioning into an adaptive filter built inside the FPGA; initializing the adaptive filter by determining its order and initial weight vector; setting the convergence factor of the iterative algorithm based on the noise characteristics of the GIS site. This convergence factor needs to balance the convergence speed and stability of the algorithm to meet real-time processing requirements; decomposing the digital voiceprint signal into a main input signal and a reference input signal, where the reference input signal is selected from signal components strongly correlated with the background noise; and starting the iterative algorithm for noise reduction calculation: calculating the filter output signal, which has... The process involves multiplying each weight in the filter's weight vector by the corresponding sample value in the reference input signal vector, then summing all the products to obtain the filter's output signal at the current moment. An error signal is calculated by subtracting the filter's output signal from the main input signal, resulting in an error signal characterizing the noise reduction effect. The weight vector is then updated by adding a convergence factor to the original weight vector, multiplying by the error signal, and then multiplying by the reference input signal vector, completing one weight update. This iterative calculation process is repeated, continuously tracking changes in background noise and dynamically adjusting the weight vector until the mean square value of the error signal drops to a preset stable threshold. At this point, the filter's output signal is the noise-reduced voiceprint signal, which effectively suppresses background noise while preserving the core voiceprint characteristics of partial discharge.

[0028] Step 2.2: Perform Hilbert transform on the denoised acoustic signature signal to extract the time-domain envelope parameters characterizing the amplitude variation trend of the signal; perform Fast Fourier Transform on the denoised acoustic signature signal to obtain the spectral energy distribution parameters of each frequency band; perform pulse detection on the denoised acoustic signature signal to extract pulse waveform parameters including pulse amplitude, rise time, and decay constant, specifically including: performing Hilbert transform on the denoised acoustic signature signal to determine the discrete sample sequence of the denoised acoustic signature signal, constructing the corresponding Hilbert transform kernel function, and performing convolution operation on each signal sample with the kernel function to obtain the Hilbert transform sequence of the signal; using the original denoised acoustic signature signal as the real part and the Hilbert transform sequence as the imaginary part, combine them to form an analytic signal; calculate the magnitude value for each sample point of the analytic signal, that is, first perform squaring operation on the real part value and squaring operation on the imaginary part value, then add the two squaring results, and finally take the square root of the summation result. The resulting magnitude value sequence is the time-domain envelope parameter, which can accurately characterize the amplitude variation trend of the partial discharge acoustic signature signal.

[0029] The noise-reduced voiceprint signal is subjected to a Fast Fourier Transform (FFT): First, the noise-reduced voiceprint signal sample sequence is padded with zeros to a length equal to a power of 2, satisfying the FFT operation rules; following the butterfly operation logic, the time-domain voiceprint signal is converted into a complex sequence in the frequency domain; the amplitude is calculated for each frequency point in the complex sequence, i.e., the square root of the sum of the squares of the real and imaginary parts of the complex number at that frequency point; the entire ultrasonic frequency band is divided into several continuous sub-bands, and the energy value of each sub-band is calculated, specifically by squaring the amplitude of all frequency points within the sub-band and summing all the squaring results. The energy values ​​of all sub-bands together constitute the spectral energy distribution parameters. Pulse detection is then performed on the noise-reduced voiceprint signal: a pulse detection threshold is determined, which is... The mean of the background noise amplitude during normal operation of the GIS equipment is added to three times the standard deviation of the background noise amplitude. The noise-reduced voiceprint signal sample sequence is traversed, and when a sample amplitude exceeds the threshold, it is marked as the pulse start point. The subsequent samples are traversed from the pulse start point to find the maximum amplitude, which is the pulse amplitude. The number of samples from the pulse start point to the maximum amplitude point is counted, and this number is multiplied by the signal sampling period to obtain the pulse rise time. The signal in the attenuation stage after the pulse peak is subjected to exponential fitting, with the signal amplitude in the attenuation stage as the dependent variable and time as the independent variable. The attenuation constant is calculated by least squares fitting. The pulse amplitude, rise time, and attenuation constant together constitute the pulse waveform parameters.

[0030] Step 2.3 involves fusing the extracted time-domain envelope parameters, spectral energy distribution parameters, and pulse waveform parameters to form an initial acoustic signature feature vector describing the characteristics of partial discharge. Specifically, this includes normalizing the extracted time-domain envelope parameters, spectral energy distribution parameters, and pulse waveform parameters to eliminate dimensional and dimensional differences between the parameters: Time-domain envelope parameter normalization: Take the minimum and maximum values ​​of all sample values ​​of the time-domain envelope parameters, subtract the minimum value from each sample value, and then divide by the difference between the maximum and minimum values ​​to obtain the normalized time-domain envelope parameters; Spectral energy distribution parameter normalization: Divide the energy value of each sub-band by the sum of the energy values ​​of all sub-bands to obtain the normalized spectral energy distribution parameters. Pulse waveform parameters are normalized by dividing the pulse amplitude by the preset maximum value of the partial discharge pulse amplitude, the rise time by the preset maximum value of the partial discharge rise time, and the attenuation constant by the preset maximum value of the partial discharge attenuation constant. After normalization, the three types of normalized parameters are concatenated in the preset order of time-domain envelope parameters, spectral energy distribution parameters, and pulse waveform parameters to form a one-dimensional parameter sequence. This sequence is the initial acoustic signature feature vector describing the characteristics of partial discharge. The length of the vector is equal to the sum of the number of samples of the three types of parameters, and the value at each position corresponds to the normalization result of one type of parameter.

[0031] In this embodiment of the invention, the adaptive noise reduction stage dynamically updates the filter weights through an iterative algorithm, which can track and suppress background noise such as mechanical vibration and electromagnetic interference in the GIS site in real time. Compared with the traditional fixed filtering method, it can retain the effective acoustic fingerprint signal of partial discharge to the greatest extent and solve the problem of effective signal being masked by background noise superposition. The feature extraction stage comprehensively extracts parameters from three dimensions: time domain, frequency domain, and pulse characteristics. Each parameter is obtained through precise mathematical calculation, which can completely characterize the acoustic fingerprint features of partial discharge and avoid the defect that single-dimensional parameters cannot fully reflect the characteristics of partial discharge. The normalization process during feature vector construction eliminates the differences in the dimensions and units of different parameters, so that the initial acoustic fingerprint feature vector can uniformly and accurately describe the characteristics of partial discharge. This lays a reliable data foundation for feature correction and classification based on spatial parameters and reduces the risk of feature distortion. The entire process is implemented based on FPGA hardware architecture. All calculation steps are executed in parallel through hardware logic. Compared with the serial processing mode of general processors, the signal processing speed is improved, which meets the engineering requirements of real-time monitoring of partial discharge acoustic fingerprints in GIS.

[0032] In a preferred embodiment of the present invention, step 3 above may include: Step 3.1: Obtain the pre-stored spatial coordinate datasets of each high-frequency ultrasonic acoustic sensor and the corresponding volume parameters of the air chamber. Specifically, this includes: retrieving the configuration parameters that were calibrated and stored before the device left the factory from the read-only memory unit inside the FPGA; sequentially reading the spatial coordinate dataset of each high-frequency ultrasonic acoustic sensor, which contains the horizontal, vertical, and longitudinal spatial coordinates of each sensor inside the GIS air chamber; and simultaneously reading the GIS air chamber volume parameters corresponding to the installation position of each sensor, which includes the overall length, width, height, and effective volume of the air chamber filled with insulating gas. All parameters are completely matched with the actual physical structure of the GIS device and are directly used as the reference data for spatial geometric modeling and acoustic path analysis.

[0033] Step 3.2: The initial acoustic signature feature vector is fused with the spatial coordinate dataset and volume parameters. Based on the spatial geometric relationship between the sound source point and each sensor determined after fusion, the three-dimensional geometric structure of the GIS air chamber and the path trajectory of the sound wave from the sound source point to each sensor are constructed by calculating the projection method of the spatial gradient, thus obtaining a virtual spatial reference configuration. Specifically, this includes: fusing the initial acoustic signature feature vector with the sensor spatial coordinate dataset and air chamber volume parameters; determining the spatial location of the sound source point corresponding to partial discharge based on the signal intensity distribution characteristics represented by the initial acoustic signature feature vector; first calculating the spatial straight-line distance between the sound source point and a single high-frequency ultrasonic acoustic signature sensor; squaring the difference between the sensor's horizontal coordinate and the sound source point's horizontal coordinate; squaring the difference between the sensor's vertical coordinate and the sound source point's vertical coordinate; squaring the difference between the sensor's vertical coordinate and the sound source point's vertical coordinate; adding the results of the three squaring operations; and then calculating the total sum. The square root operation is performed to obtain the spatial straight-line distance from the sound source point to the sensor. Following the same calculation method, the spatial straight-line distance between the sound source point and all high-frequency ultrasonic acoustic sensors is calculated sequentially. Based on the spatial straight-line distance, the spatial gradient from the sound source point to each sensor is calculated. The spatial gradient is the continuous rate of change of the spatial straight-line distance along the direction of sound wave propagation. The calculated spatial gradient is projected onto the three-dimensional geometric plane corresponding to the GIS air chamber. The projection process is to multiply the spatial gradient value by the normal component of the corresponding three-dimensional geometric plane to obtain the projected spatial gradient value. Based on the projected spatial gradient value, combined with the sensor spatial coordinate dataset and air chamber volume parameters, the complete three-dimensional geometric structure of the GIS air chamber is gradually restored. At the same time, the complete path trajectory of the sound wave from the sound source point to each high-frequency ultrasonic acoustic sensor is determined. The three-dimensional geometric structure of the GIS air chamber and all sound wave propagation path trajectories are integrated to form a virtual spatial reference configuration that truly reflects the actual spatial form of the equipment and the law of sound wave propagation.

[0034] Step 3.3 involves dividing the virtual space reference configuration into finite element meshes according to a preset spatial resolution to obtain multiple continuous meshed regions. Specifically, this includes: setting a unified spatial resolution in advance based on the accuracy requirements of partial discharge monitoring of GIS equipment. This spatial resolution serves as the standardized division scale for three-dimensional spatial units. The virtual space reference configuration is then divided into finite element meshes according to the preset spatial resolution. Along the lateral direction of the GIS air chamber, the overall length of the air chamber is divided by the spatial resolution to obtain the number of mesh segments in the lateral direction. Along the longitudinal direction of the air chamber, the overall width of the air chamber is divided by the spatial resolution to obtain the number of mesh segments in the longitudinal direction. Along the vertical direction of the air chamber, the overall height of the air chamber is divided by the spatial resolution to obtain the number of mesh segments in the vertical direction. The virtual space reference configuration is then divided region by region according to the number of mesh segments in the lateral, longitudinal, and vertical directions. Each spatial unit formed by the division is a regular three-dimensional mesh unit. All mesh units are continuously arranged without overlapping areas or gaps, ultimately resulting in multiple continuous meshed regions covering the entire virtual space reference configuration.

[0035] In this embodiment of the invention, accurate sensor spatial coordinates and air chamber volume parameters are obtained in advance, providing a reliable physical benchmark for spatial modeling. This solves the modeling deviation problem caused by the lack of spatial parameters in traditional technologies, and avoids the risk of data distortion from the basic stage. By fusing the initial acoustic feature vector with spatial parameters and calculating spatial gradient projection, a virtual spatial reference configuration is constructed, which realistically restores the geometric structure of the GIS air chamber and the sound wave propagation path. This makes up for the technical defects of traditional technologies that do not integrate the spatial geometric characteristics of the equipment, and provides an accurate spatial reference for the spatial correction of acoustic features. Finite element meshing is completed according to the preset spatial resolution, and the continuous virtual space is discretized into regular meshed regions, which facilitates the analysis of shell material constraints and sound wave propagation characteristics region by region, improving the accuracy and execution efficiency of spatial propagation compensation calculation. The entire process of spatial modeling and meshing is executed in parallel by FPGA hardware logic, which has fast computing speed and strong real-time performance, and is suitable for the engineering application requirements of online monitoring of partial discharge of GIS equipment. At the same time, it lays the foundation for eliminating feature distortion caused by spatial propagation.

[0036] In a preferred embodiment of the present invention, step 4 above may include: Step 4.1: Analyze the geometric constraints of the shell material in each region of the meshed area to obtain the attenuation and reflection parameters of the sound wave propagating in the corresponding shell material. Specifically, this includes: for all meshed areas obtained after finite element analysis, traversing and analyzing each meshed area one by one according to the sound wave propagation sequence from the local discharge point to the high-frequency ultrasonic acoustic sensor. During the traversal, the geometric constraints of the GIS shell within each individual meshed area are determined. These geometric constraints specifically include the actual wall thickness of the shell within the current meshed area, the curvature parameters of the shell wall, the joint structure of the shell splicing points, and the metal material of the shell. The contact interface morphology and interface flatness between the sound wave and the internal insulating gas are determined through the analysis of the above geometric constraints. This clarifies the physical propagation limitations imposed by the shell structure and interface morphology when the sound wave passes through the current gridded area. After completing the analysis of the geometric constraints, the basic acoustic parameters matching the shell material of the current gridded area are retrieved from the material acoustic parameter library pre-stored inside the FPGA. The basic acoustic parameters include the sound wave attenuation coefficient per unit thickness of the shell metal material, the acoustic impedance value of the shell metal material, and the acoustic impedance value of the insulating gas inside the GIS. After obtaining the above basic parameters, the sound wave attenuation parameters and sound wave reflection parameters are calculated respectively.

[0037] The specific calculation method for the sound wave attenuation parameter is as follows: multiply the actual wall thickness of the shell within the current gridded area by the sound wave attenuation coefficient per unit thickness corresponding to the shell material. The product is the sound wave attenuation parameter generated when the sound wave passes through the shell within the current gridded area. This parameter directly characterizes the degree of energy attenuation of the sound wave within the current gridded area. The specific calculation method for the sound wave reflection parameter is as follows: extract the acoustic impedance values ​​of the shell metal material and the internal insulating gas. Subtract the acoustic impedance value of the insulating gas from the acoustic impedance value of the shell metal material to obtain the acoustic impedance difference between the two media. Then add the acoustic impedance values ​​of the shell metal material and the insulating gas to obtain the total acoustic impedance between the two media. Divide the acoustic impedance difference by the total acoustic impedance. The result of the division operation is the sound wave reflection parameter generated at the shell-gas interface within the current gridded area. This parameter directly characterizes the degree of energy loss after the sound wave is reflected at the interface. Following the above analysis and calculation process, the attenuation and reflection parameters of all gridded areas are extracted.

[0038] Step 4.2 involves associating and fusing the attenuation parameters and reflection parameters with the spatial location of each gridded region to obtain a set of spatial propagation compensation coefficients corresponding to each gridded region. Specifically, this includes associating and binding the independently calculated acoustic attenuation parameters and acoustic reflection parameters of each gridded region with its three-dimensional spatial location in the virtual space reference configuration, thereby achieving precise fusion of propagation characteristic parameters and spatial location. Based on this, the spatial propagation compensation coefficient of a single gridded region is calculated. During the calculation process, the distance weight and path weight corresponding to the current gridded region are first determined. The distance weight is the ratio obtained by dividing the straight-line distance from the current gridded region to the local discharge point by the total propagation distance from the sound source point to the corresponding high-frequency ultrasonic acoustic fingerprint sensor. The path weight is the proportion of the path length occupied by the current gridded region in the entire acoustic propagation path.

[0039] The compensation coefficient is calculated by multiplying the acoustic attenuation parameter of the current gridded area by the corresponding distance weight to obtain the attenuation compensation component; multiplying the acoustic reflection parameter of the current gridded area by the corresponding path weight to obtain the reflection compensation component; and then adding the attenuation compensation component and the reflection compensation component together. The sum of the two is the spatial propagation compensation coefficient specific to the current gridded area. Following a unified calculation logic, the spatial propagation compensation coefficient is calculated sequentially for all gridded areas. The calculation process relies on the FPGA hardware parallel computing unit to execute synchronously, ensuring computational efficiency. The spatial propagation compensation coefficients calculated for all gridded areas are systematically summarized and arranged according to the order of the acoustic propagation path, ultimately forming a complete set of spatial propagation compensation coefficients covering the entire virtual space reference configuration and corresponding one-to-one with all gridded areas.

[0040] Step 4.3: Based on the spatial propagation compensation coefficient set, perform spatial normalization correction on the initial acoustic feature vector to eliminate the signal attenuation caused by differences in propagation path and shell material, and obtain the corrected acoustic feature vector. Specifically, this includes: using the spatial propagation compensation coefficient set as the basis for correction, performing targeted spatial normalization correction operation on the extracted initial acoustic feature vector. The purpose is to completely eliminate signal attenuation and energy distortion interference caused by differences in sound wave propagation path length, shell material, and geometric structure changes. During the correction process, all feature components in the initial acoustic feature vector, such as the time domain envelope parameter, spectral energy distribution parameter, and pulse waveform parameter, are first accurately matched with the spatial propagation compensation coefficients of each gridded region on the corresponding sound wave propagation path.

[0041] After matching is completed, each feature component in the initial voiceprint feature vector is multiplied sequentially with the spatial propagation compensation coefficient of the corresponding gridded region on the propagation path. After multiplying each feature component with the corresponding compensation coefficient, the intermediate value of the feature component after spatial propagation compensation is obtained. The intermediate values ​​of all feature components after compensation are summed to obtain the total value of all feature components after compensation. Then, the total value is divided by the total number of spatial propagation compensation coefficients participating in this correction operation to complete the mean normalization process, eliminating the difference in numerical magnitude caused by different compensation coefficients. The final numerical sequence obtained after multiplication, summation and normalization is the voiceprint feature vector after spatial correction. This vector completely eliminates the interference caused by spatial propagation and only retains the inherent voiceprint characteristics of partial discharge itself.

[0042] In this embodiment of the invention, by analyzing the geometric constraints of the shell material region by region and using a precise textual calculation method to obtain the sound wave attenuation parameters and reflection parameters, the true physical laws of sound wave propagation in the GIS shell and insulating gas are fully restored. This solves the technical shortcoming of traditional partial discharge monitoring technology that completely ignores the influence of shell material and geometry on sound wave signals. It ensures the authenticity and accuracy of propagation characteristic analysis from the parameter level. The attenuation parameters, reflection parameters and the spatial location of the gridded area are correlated and fused to calculate and generate a set of spatial propagation compensation coefficients. This realizes the differentiated quantification of propagation loss under different spatial locations and different propagation paths, avoiding the analysis error caused by the traditional technology of using uniform parameters to process all propagation scenarios. This allows the spatial propagation compensation to fit the actual propagation conditions of each region.

[0043] By employing a normalization correction method that multiplies the characteristic components with compensation coefficients and then sums and averages them, signal attenuation and waveform distortion introduced by differences in propagation distance, shell material, and geometric structure are accurately eliminated. This fundamentally solves the core problem of severe distortion in the initial acoustic signature feature vector in the background technology, allowing the corrected acoustic signature feature vector to truly reflect the inherent characteristics of partial discharge itself. The acoustic signature feature vector after spatial normalization correction eliminates interference factors not caused by the discharge itself, improving the reliability of subsequent partial discharge type identification and fault level determination. It avoids the problems of traditional technology misjudging air gap discharge as floating potential discharge and excessive deviation in fault level determination due to feature distortion, enabling staff to promptly open The system provides precise data support for operation and maintenance. The entire spatial propagation compensation and feature correction process is completed in hardware logic within the FPGA. All multiplication and addition operations are executed in parallel, eliminating the latency issues associated with serial operations in general-purpose processors. While ensuring correction accuracy, it meets the engineering requirements for real-time monitoring of partial discharge in GIS. It can quickly complete feature correction, reserving sufficient time for subsequent fault identification and reporting. Through spatial propagation compensation, it achieves accurate correction of acoustic features, effectively avoiding faults such as accelerated insulation degradation and power outages in GIS equipment caused by distorted monitoring data. This improves the stability and security of GIS equipment status monitoring in power systems and meets the actual needs of long-term stable operation of high-voltage substations.

[0044] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1: Input the corrected voiceprint feature vector into the parallel matching and discrimination logic embedded in the FPGA. Obtain the initial matching result by performing parallel matching calculations with pre-stored templates for various partial discharge types. Specifically, this includes: inputting the spatially normalized corrected voiceprint feature vector into the parallel matching and discrimination logic embedded in the FPGA using a hardware description language. The FPGA's internal storage unit pre-stores standard voiceprint feature templates for various typical partial discharge types, such as corona discharge, floating potential discharge, surface discharge, and air gap discharge. Each type of partial discharge corresponds to a complete set of standard feature templates that have been calibrated on-site. At the same time, the parallel matching and discrimination logic performs similarity matching calculations between the corrected voiceprint feature vector and the standard feature templates for all types of partial discharge. The specific calculation method is as follows: for each feature component in the corrected voiceprint feature vector, the difference operation is performed with the feature component at the same position in the corresponding partial discharge standard feature template. The difference is squared and then summed sequentially. The square root of the sum is then performed to obtain the matching distance value between the current feature vector and the template of that type. The smaller the matching distance value, the higher the similarity between the corrected voiceprint feature vector and the type of partial discharge. According to the above calculation method, the matching distance value between the corrected voiceprint feature vector and each type of partial discharge template is obtained. All matching distance values ​​are sorted in ascending order, and the discharge type corresponding to the template with the highest ranking is taken as the current initial matching result.

[0045] Step 5.2 involves statistically analyzing the distribution characteristics of historically corrected voiceprint feature vectors and dynamically adjusting the discrimination thresholds for each category in the parallel matching and discrimination logic based on these distribution characteristics. This process includes: establishing a dedicated data cache area within the FPGA to continuously store and statistically analyze the voiceprint feature vectors generated and spatially corrected during multiple historical identification processes. This continuous accumulation forms a historical voiceprint feature sample set. The distribution characteristics of all corrected voiceprint feature vectors in the historical sample set are statistically analyzed, and the mean and dispersion of the feature vectors corresponding to each type of partial discharge are calculated. Specifically, the calculation method involves sequentially matching the components at the same position of all historical feature vectors corresponding to the same type of discharge. The mean of the feature vectors for that class is obtained by adding the mean and dividing by the total number of samples in that class. The difference between each historical feature vector and the mean is calculated, the difference is squared and summed, and then divided by the total number of samples to obtain the dispersion of the feature vectors for that class. Based on the statistically obtained mean and dispersion, the discrimination thresholds for each class in the parallel matching discrimination logic are dynamically adjusted. Specifically, the mean of each type of partial discharge is added to the product of the dispersion and the dynamic coefficient to obtain the adjusted discrimination threshold for that class. The dynamic coefficient is selected in real time according to the on-site noise intensity and equipment operating conditions to ensure that the threshold can adapt to the changes in the current monitoring environment. Finally, the adjusted discrimination thresholds for each type of partial discharge are obtained.

[0046] Step 5.3 involves applying the adjusted discrimination threshold to the matching calculation process of the parallel matching discrimination logic to calibrate the initial matching result and obtain the classification result. Specifically, this includes: writing the adjusted discrimination threshold into the threshold register of the parallel matching discrimination logic, replacing the original fixed discrimination threshold, and applying it to the matching calculation and result determination process; comparing the matching distance value in the initial matching result with the adjusted discrimination threshold corresponding to the discharge type; if the matching distance value corresponding to the initial matching result is less than or equal to the adjusted discrimination threshold, the initial matching result is deemed valid, and the category is directly used as the final classification result; if the matching distance value corresponding to the initial matching result is greater than the adjusted discrimination threshold, the initial matching result is deemed invalid, and the next highest-ranked template type is selected and compared again with the corresponding adjusted discrimination threshold until the condition that the distance value is less than or equal to the discrimination threshold is met. Through the comparison, screening, and calibration process, the calibration of the initial matching result is completed, and finally, a reliable and accurate partial discharge classification result is obtained, clearly distinguishing the current signal as a specific type among corona discharge, floating potential discharge, surface discharge, or air gap discharge.

[0047] In this embodiment of the invention, parallel matching and discrimination logic embedded within the FPGA is used to simultaneously calculate the matching distance between the corrected acoustic signature feature vector and templates of various partial discharge types. This fully leverages the advantages of parallel processing in FPGA hardware, shortening the matching calculation time compared to the serial, line-by-line matching method of traditional processors. This ensures the real-time performance of partial discharge acoustic signature recognition and meets the rapid response requirements of online monitoring for GIS equipment. Similarity is determined by calculating the matching distance between the feature vector and the template. The calculation process is standardized and quantifiable, avoiding subjective errors caused by manual or fuzzy judgments. This provides a unified judgment standard for the initial matching results and offers an objective and reliable data foundation for result calibration. By continuously statistically analyzing the distribution characteristics of historically corrected acoustic signature feature vectors, the discrimination thresholds for various partial discharges are dynamically calculated and updated. This overcomes the shortcomings of traditional technologies that use fixed discrimination thresholds and cannot adapt to changes in the field environment. It can adaptively adjust to factors such as noise intensity, equipment operating status, and sensor aging, effectively improving the discrimination stability under complex working conditions.

[0048] By calibrating the initial matching results using dynamically adjusted discrimination thresholds, erroneous matching results caused by environmental interference and signal fluctuations can be effectively eliminated. This fundamentally solves the problem in the background technology where air gap discharge is misjudged as floating potential discharge and the classification results are distorted due to fixed thresholds. It improves the accuracy of partial discharge type identification. The classification results after dynamic threshold calibration are closer to the actual insulation state of the equipment, providing a reliable basis for fault level determination. This allows maintenance personnel to promptly grasp the internal insulation defects of GIS equipment, avoiding the aggravation of insulation deterioration due to misjudgment or omission, and reducing the risk of insulation breakdown and power outage accidents in GIS equipment. The entire classification and identification process is implemented in hardware circuitry within the FPGA, without relying on external processors and software programs. It features stable operation, strong anti-electromagnetic interference capability, and fast response speed, making it more suitable for the harsh industrial environment of strong electromagnetic fields and strong vibrations in substations, thus improving the reliability and engineering applicability of the entire monitoring system.

[0049] In a preferred embodiment of the present invention, step 6 above may include: Step 6.1: Based on at least one classification result among corona discharge, floating potential discharge, surface discharge, and air gap discharge, and combined with the amplitude parameter in the corrected acoustic signature feature vector, determine the fault level of the current partial discharge according to the preset fault level determination rules. Specifically, this includes: after the partial discharge classification result is one of corona discharge, floating potential discharge, surface discharge, or air gap discharge, extracting the amplitude parameter representing the signal energy intensity from the corrected acoustic signature feature vector. This amplitude parameter includes the maximum and minimum values ​​of the time-domain envelope and the pulse peak amplitude. Retrieving the preset fault level determination rules from the FPGA's internal storage unit, which divides the corresponding fault level amplitude range for different partial discharge types, and using the extracted maximum and minimum values ​​of the time-domain envelope... The value is summed with the pulse peak amplitude, and the sum is divided by two to obtain the comprehensive amplitude parameter of the current partial discharge. The calculated comprehensive amplitude parameter is compared with the amplitude range of each level of fault corresponding to the current discharge type. If the comprehensive amplitude parameter is less than the upper limit of the first-level fault amplitude of the corresponding discharge type, it is judged as a level one minor fault; if the comprehensive amplitude parameter is greater than or equal to the upper limit of the first-level fault amplitude and less than the upper limit of the second-level fault amplitude, it is judged as a level two general fault; if the comprehensive amplitude parameter is greater than or equal to the upper limit of the second-level fault amplitude and less than the upper limit of the third-level fault amplitude, it is judged as a level three severe fault; if the comprehensive amplitude parameter is greater than or equal to the upper limit of the third-level fault amplitude, it is judged as a level four critical fault. Through the comparison and judgment process, the fault level corresponding to the current partial discharge is finally determined.

[0050] Step 6.2: Encapsulate the classification results and determined fault levels according to a preset data frame format to generate an upload data frame. Specifically, this includes: encapsulating the partial discharge classification results and fault levels according to a unified data frame format pre-defined within the FPGA. The data encapsulation is performed sequentially in the following order: frame header, device identifier, classification result, fault level, characteristic amplitude, verification segment, and frame tail. The frame header is filled with a fixed frame start identifier; the device identifier is filled with the current GIS air chamber and sensor array number; the classification result is filled with the corresponding discharge type encoding information; the fault level is filled with the determined fault level value; and the characteristic amplitude is filled with the comprehensive amplitude parameter calculated in this operation. The verification segment is generated through a verification operation. All data from the frame header to the characteristic amplitude segment are summed sequentially, and the remainder obtained by dividing the sum by a fixed value is used as the verification value. After completing the orderly concatenation of all data segments, a complete upload data frame conforming to the communication transmission standard is generated.

[0051] Step 6.3 involves sending the uploaded data frame to the station control platform via the communication interface to complete the real-time identification, classification, and fault level determination of the partial discharge acoustic pattern. Specifically, this includes: activating the communication interface driver module within the FPGA; transmitting the uploaded data frame to the transmission buffer unit of the communication interface; sending the uploaded data frame bit by bit to the station control platform via Ethernet or serial communication; synchronously verifying the transmitted data in real time during data transmission; comparing the verification value from the sending end with the verification value calculated by the receiving end of the station control platform; if the verification values ​​match, the data transmission is considered complete and distortion-free; if the verification values ​​do not match, the currently uploaded data frame is automatically retransmitted until the transmission verification passes. After the station control platform successfully receives and parses the uploaded data frame, it sends a reception completion signal to the FPGA. Upon receiving the feedback signal, the FPGA determines that the reporting process is complete, thus completing the full reporting of the real-time identification, classification, and fault level determination of the partial discharge acoustic pattern.

[0052] In this embodiment of the invention, fault level determination is performed by combining the classification results of different discharge types with comprehensive amplitude parameters. Different amplitude determination intervals are set for different partial discharge types. Simultaneously, the comprehensive amplitude parameter is obtained by averaging the maximum and minimum values ​​of the time-domain envelope and the pulse peak amplitude, avoiding determination errors caused by fluctuations in a single amplitude parameter. This solves the problem of large deviations in fault level determination in traditional technologies, allowing the fault level to accurately reflect the severity of internal insulation defects in GIS equipment. The fault level determination rules are embedded within the FPGA, and the entire determination process is hardware logic operation without manual intervention or software delay. This enables rapid level determination after partial discharge occurs, ensuring real-time fault assessment. This provides maintenance personnel with accurate information to quickly assess the extent of equipment malfunctions. Data is encapsulated according to a standardized data frame format, and data integrity and standardization are ensured through segmented splicing and verification operations. This prevents data loss and corruption during transmission, providing a standard data format for subsequent parsing and display by the station control platform. It also improves the stability of communication between the upper and lower level computers. Identification, classification, and judgment results are uploaded to the station control platform in real time via a communication interface. Combined with a transmission verification mechanism, this ensures the accuracy of data transmission, solving the shortcomings of traditional monitoring systems such as untimely data reporting and unreliable transmission. This allows staff to obtain insulation status information of GIS equipment immediately and take timely maintenance and repair measures.

[0053] The entire process achieves a closed-loop processing from partial discharge identification, classification, and level determination to data reporting. It fundamentally avoids the problem of accelerated insulation deterioration of GIS equipment due to errors in fault level determination and untimely reporting of results, eliminates the occurrence of partial power supply interruptions in substations, and improves the safety and reliability of power system GIS equipment status monitoring. The entire process of fault level determination and data reporting is implemented using FPGA hardware, which has strong anti-electromagnetic interference capabilities and is suitable for the complex working conditions of strong vibration and strong electromagnetic interference in substations. At the same time, the entire process is automated and requires no manual intervention, reducing the operation and maintenance costs of the monitoring system and meeting the engineering requirements for long-term stable operation of high-voltage substations.

[0054] like Figure 2 As shown, embodiments of the present invention also provide an FPGA-based real-time partial discharge acoustic print recognition and classification integrated circuit processing system, including: The acquisition module is used to receive the raw acoustic fingerprint signals from the high-frequency ultrasonic acoustic fingerprint sensor array installed on the side wall of each air chamber of the GIS; and to pre-condition the raw acoustic fingerprint signals inside the FPGA to obtain the conditioned acoustic fingerprint signals. The noise reduction module is used to perform adaptive noise reduction processing on the conditioned voiceprint signal to obtain the noise-reduced voiceprint signal; the time domain envelope, spectral energy distribution and pulse waveform parameters are extracted from the noise-reduced voiceprint signal to obtain the initial voiceprint feature vector of partial discharge characteristics. The projection module is used to construct a virtual spatial reference configuration of the GIS air cell geometry and sound wave propagation path by using an initial feature vector, combined with the spatial coordinates of each sensor and the corresponding air cell volume parameters, and adopting a parametric modeling method of spatial pose gradient projection; and to divide the virtual spatial reference configuration into gridded regions to obtain the divided gridded regions. The correction module is used to analyze the geometric constraints of the shell material within the meshed area, integrate the attenuation and reflection characteristics of sound waves in the material, and generate a set of spatial propagation compensation coefficients. The initial soundprint feature vector is spatially normalized and corrected using the set of spatial propagation compensation coefficients to obtain the corrected soundprint feature vector. The calibration module is used to input the corrected voiceprint feature vector into the parallel matching and discrimination logic embedded in the FPGA, and dynamically calibrate the discrimination threshold of the classifier according to the statistical voiceprint feature distribution to obtain the classification result. The processing module is used to determine the fault level of partial discharge based on the classification results and the magnitude of the corrected acoustic signature feature vector; it encapsulates the classification results and fault level into data and uploads them to the station control platform through the communication interface.

[0055] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for real-time identification and classification of partial discharge acoustic prints based on FPGA, characterized in that, The method includes: The system receives the raw acoustic signature signal from the high-frequency ultrasonic acoustic signature sensor array installed on the side wall of each air chamber of the GIS; and pre-conditions the raw acoustic signature signal inside the FPGA to obtain the conditioned acoustic signature signal. Adaptive noise reduction is performed on the conditioned voiceprint signal to obtain the noise-reduced voiceprint signal; the time-domain envelope, spectral energy distribution and pulse waveform parameters are extracted from the noise-reduced voiceprint signal to obtain the initial voiceprint feature vector of partial discharge characteristics. By using initial feature vectors and combining the spatial coordinates of each sensor with the corresponding air chamber volume parameters, a parametric modeling method using spatial position gradient projection is adopted to construct a virtual spatial reference configuration of the GIS air chamber geometry and sound wave propagation path. In this process, the spatial straight-line distance between the sound source point and all high-frequency ultrasonic acoustic sensors is calculated sequentially using the same calculation method. Based on the spatial straight-line distance, the spatial gradient from the sound source point to each sensor is calculated. The spatial gradient is the continuous rate of change of the spatial straight-line distance along the sound wave propagation direction. The calculated spatial gradient is then projected onto the three-dimensional geometric plane corresponding to the GIS air chamber. The projection process involves multiplying the spatial gradient value by the normal component of the corresponding three-dimensional geometric plane to obtain the projected spatial gradient value. This forms a virtual spatial reference configuration that truly reflects the actual spatial shape of the equipment and the sound wave propagation law. The virtual spatial reference configuration is then divided into gridded regions to obtain the divided gridded regions. The geometric constraints of the shell material within the gridded area are analyzed, and the attenuation and reflection characteristics of sound waves in the material are integrated to generate a set of spatial propagation compensation coefficients. The initial acoustic feature vector is spatially normalized and corrected using the set of spatial propagation compensation coefficients to obtain the corrected acoustic feature vector. The corrected voiceprint feature vector is input into the parallel matching and discrimination logic embedded in the FPGA, and the discrimination threshold of the classifier is dynamically calibrated according to the statistical voiceprint feature distribution to obtain the classification result. Based on the classification results and the magnitude of the corrected acoustic signature feature vector, the fault level of partial discharge is determined; the classification results and fault level are encapsulated to achieve real-time identification and classification of partial discharge acoustic signatures.

2. The FPGA-based real-time partial discharge acoustic print recognition and classification integrated circuit processing method according to claim 1, characterized in that, The system receives raw acoustic signature signals from a high-frequency ultrasonic acoustic signature sensor array installed on the sidewalls of each air chamber in the GIS; it then pre-conditions the raw acoustic signature signals within the FPGA to obtain conditioned acoustic signature signals, including: The received raw voiceprint signal is amplified by programmable gain control to obtain an amplitude-adjusted analog voiceprint signal; The amplitude-adjusted analog voiceprint signal is subjected to anti-aliasing filtering to remove high-frequency interference, resulting in a filtered analog voiceprint signal. The filtered analog voiceprint signal is converted into a digital voiceprint signal through an analog-to-digital converter, and the digital voiceprint signal is used as the conditioned voiceprint signal.

3. The FPGA-based real-time partial discharge acoustic print recognition and classification integrated circuit processing method according to claim 2, characterized in that, Adaptive denoising is performed on the conditioned voiceprint signal to obtain the denoised voiceprint signal. The temporal envelope, spectral energy distribution, and pulse waveform parameters are extracted from the denoised voiceprint signal to obtain the initial voiceprint feature vector of partial discharge characteristics, including: The digital voiceprint signal is input into an adaptive filter, and the background noise in the signal is dynamically tracked and suppressed through an iterative algorithm to obtain the noise-reduced voiceprint signal. The noise-reduced voiceprint signal is subjected to Hilbert transform to extract the time-domain envelope parameter that characterizes the signal amplitude variation trend; the noise-reduced voiceprint signal is subjected to fast Fourier transform to obtain the spectral energy distribution parameters of each frequency band; the noise-reduced voiceprint signal is subjected to pulse detection to extract pulse waveform parameters including pulse amplitude, rise time and decay constant. The extracted time-domain envelope parameters, spectral energy distribution parameters, and pulse waveform parameters are fused to form an initial acoustic signature feature vector describing the characteristics of partial discharge.

4. The FPGA-based real-time partial discharge acoustic print recognition and classification integrated circuit processing method according to claim 3, characterized in that, By using the initial feature vector, combined with the spatial coordinates of each sensor and the corresponding air chamber volume parameters, a parametric modeling method using spatial position gradient projection is adopted to construct a virtual spatial reference configuration of the GIS air chamber geometry and sound wave propagation path. The virtual space reference configuration is then divided into meshed regions, resulting in meshed regions including: Obtain the pre-stored spatial coordinate dataset of each high-frequency ultrasonic acoustic sensor and the corresponding volume parameters of the air chamber; The initial acoustic feature vector is fused with the spatial coordinate dataset and volume parameters. Based on the spatial geometric relationship between the sound source point and each sensor determined after fusion, the three-dimensional geometric structure of the GIS air chamber and the path trajectory of the sound wave from the sound source point to each sensor are constructed by calculating the projection of the spatial gradient, thus obtaining the virtual space reference configuration. The virtual space reference configuration is divided into finite element meshes according to a preset spatial resolution to obtain multiple continuous meshed regions.

5. The FPGA-based real-time partial discharge acoustic print recognition and classification integrated circuit processing method according to claim 4, characterized in that, The geometric constraints of the shell material within the meshed region are analyzed, and the attenuation and reflection characteristics of sound waves in the material are integrated to generate a set of spatial propagation compensation coefficients. The initial acoustic signature feature vector is then spatially normalized using this set of coefficients to obtain the corrected acoustic signature feature vector, which includes: The geometric constraints of the shell material in each region of the meshed area are analyzed to obtain the attenuation and reflection parameters of the sound wave when it propagates in the corresponding shell material; By associating and fusing the attenuation parameters and reflection parameters with the spatial location of each gridded region, a set of spatial propagation compensation coefficients corresponding to each gridded region is obtained; Based on the spatial propagation compensation coefficient set, the initial voiceprint feature vector is spatially normalized to eliminate the signal attenuation caused by the difference in propagation path and shell material, thus obtaining the corrected voiceprint feature vector.

6. The FPGA-based real-time partial discharge acoustic print recognition and classification integrated circuit processing method according to claim 5, characterized in that, The corrected voiceprint feature vector is input into the parallel matching and discrimination logic embedded in the FPGA, and the discrimination threshold of the classifier is dynamically calibrated according to the statistical voiceprint feature distribution to obtain the classification result, including: The corrected voiceprint feature vector is input into the parallel matching and discrimination logic embedded in the FPGA. The initial matching result is obtained by performing parallel matching calculations with pre-stored templates of various partial discharge types. The distribution characteristics of the historically corrected voiceprint feature vectors are statistically analyzed, and the discrimination thresholds corresponding to each category in the parallel matching and discrimination logic are dynamically adjusted based on the distribution characteristics to obtain the adjusted discrimination thresholds. The adjusted discrimination threshold is applied to the matching calculation process of the parallel matching discrimination logic to calibrate the initial matching results and obtain the classification results.

7. The FPGA-based real-time partial discharge acoustic print recognition and classification integrated circuit processing method according to claim 6, characterized in that, Based on the classification results and the magnitude of the corrected acoustic signature feature vector, the fault level of partial discharge is determined; the classification results and fault level are encapsulated and uploaded to the station control platform via a communication interface, including: Based on at least one classification result among corona discharge, floating potential discharge, surface discharge, and air gap discharge, combined with the amplitude parameter in the corrected acoustic signature feature vector, and in accordance with the preset fault level judgment rules, the fault level of the current partial discharge is determined. The classification results and the determined fault levels are encapsulated according to a preset data frame format to generate an upload data frame; The uploaded data frames are sent to the station control platform through the communication interface to complete the real-time identification, classification, and reporting of fault level determination results of partial discharge acoustic patterns.

8. A real-time partial discharge acoustic print recognition and classification integrated circuit processing system based on FPGA, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to receive the raw acoustic fingerprint signals from the high-frequency ultrasonic acoustic fingerprint sensor array installed on the side wall of each air chamber of the GIS; and to pre-condition the raw acoustic fingerprint signals inside the FPGA to obtain the conditioned acoustic fingerprint signals. The noise reduction module is used to perform adaptive noise reduction processing on the conditioned voiceprint signal to obtain the noise-reduced voiceprint signal; the time domain envelope, spectral energy distribution and pulse waveform parameters are extracted from the noise-reduced voiceprint signal to obtain the initial voiceprint feature vector of partial discharge characteristics. The projection module is used to construct a virtual spatial reference configuration of the GIS air cell geometry and sound wave propagation path by using an initial feature vector, combined with the spatial coordinates of each sensor and the corresponding air cell volume parameters, and adopting a parametric modeling method of spatial position gradient projection; and to divide the virtual spatial reference configuration into gridded regions to obtain the divided gridded regions. The correction module is used to analyze the geometric constraints of the shell material within the meshed area, integrate the attenuation and reflection characteristics of sound waves in the material, and generate a set of spatial propagation compensation coefficients. The initial soundprint feature vector is spatially normalized and corrected using the set of spatial propagation compensation coefficients to obtain the corrected soundprint feature vector. The calibration module is used to input the corrected voiceprint feature vector into the parallel matching and discrimination logic embedded in the FPGA, and dynamically calibrate the discrimination threshold of the classifier according to the statistical voiceprint feature distribution to obtain the classification result. The processing module is used to determine the fault level of partial discharge based on the classification results and the magnitude of the corrected acoustic signature feature vector; it encapsulates the classification results and fault level into data and uploads them to the station control platform through the communication interface.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images