A method, apparatus, computer equipment, and readable storage medium for detecting slight partial discharge in power equipment.

Abstract

This invention discloses a method, apparatus, computer equipment, and readable storage medium for detecting minor partial discharge in power equipment. The method includes: firstly, acquiring multi-channel acoustic signals from the target power equipment using a multi-channel acoustic monitoring device through non-contact acoustic signal acquisition; secondly, performing high-pass filtering on the multi-channel acoustic signals and correcting them using Kalman filtering to obtain corrected multi-channel acoustic signals; thirdly, weighting and combining the corrected multi-channel acoustic signals to obtain a target single-channel signal; and finally, performing power frequency discharge detection on the target single-channel signal to obtain the minor partial discharge acoustic signal detection result. This method enhances the minor partial discharge signal through multi-channel signal acquisition and processing, enabling the detection of minor partial discharge characteristics and providing a reference for the production and maintenance of power equipment.

Description

Technical Field

[0001] This invention relates to the field of power equipment management, and more specifically, to a method, apparatus, computer equipment, and readable storage medium for detecting slight partial discharge in power equipment using acoustic signatures. Background Technology

[0002] With the increasing proportion of new energy power generation, the growing complexity of power grid architecture, and the scaling up of power equipment, ensuring the safe and stable operation of the power system and timely detection of equipment anomalies, especially partial discharge phenomena, has become a crucial issue. In existing partial discharge detection of power equipment, ultrasonic testing technology is used due to its non-contact characteristics. However, when detecting minor partial discharges, traditional ultrasonic testing methods suffer from insufficient accuracy in identifying weak signals due to factors such as background noise, equipment structure absorption and reflection, and low signal-to-noise ratio, making it difficult to meet the needs of early warning. Summary of the Invention

[0003] The purpose of this invention is to provide a method, apparatus, computer equipment, and readable storage medium for detecting slight partial discharge in power equipment.

[0004] In a first aspect, embodiments of the present invention provide a method for detecting slight partial discharge acoustic signatures in power equipment, characterized in that it includes:

[0005] Non-contact acoustic signal acquisition of target power equipment is performed using multi-channel acoustic monitoring equipment to obtain multi-channel acoustic signals;

[0006] The multi-channel acoustic signal is subjected to high-pass filtering, and the high-pass filtered multi-channel acoustic signal is corrected by Kalman filtering to obtain the corrected multi-channel acoustic signal.

[0007] The modified multi-channel acoustic signals are weighted and then combined to obtain the target single-channel signal;

[0008] Power frequency discharge detection is performed on the target single-channel signal to obtain the acoustic fingerprint detection result of slight partial discharge of the target power equipment.

[0009] In one possible implementation, the high-pass filtering of the multi-channel acoustic signal includes:

[0010] The multi-channel acoustic signal was high-pass filtered using a 7th-order Butterworth high-pass filter to obtain the high-pass filtered multi-channel acoustic signal.

[0011] In one possible implementation, the step of correcting the high-pass filtered multi-channel acoustic signal using Kalman filtering to obtain a corrected multi-channel acoustic signal includes:

[0012] The state estimation of the high-pass filtered multi-channel acoustic signal is performed recursively through prediction and update steps. The prediction step predicts the state at the current time based on the state estimate at the previous time step, and the update step calculates the optimal state estimate at the current time step by combining the predicted value and the actual measured value. The prediction and update steps are repeated to correct the multi-channel acoustic signal.

[0013] In one possible implementation, the prediction step includes:

[0014] Through the formula: The high-pass filtered multi-channel acoustic signal is then predicted to obtain the current state prediction estimate and the current state covariance prediction value.

[0015] in, Let be the predicted estimate of the state at time k. Here is the state transition matrix. For the control gain matrix, The state estimate at time (k-1) is... Let k be the control variable, representing the control input applied to the system at time k. Let k be the predicted value of the state covariance at time k. This is the estimated state covariance at time (k-1). Let be the covariance matrix of the process noise, representing the uncertainty in the system state transition process.

[0016] In one possible implementation, the update step includes:

[0017] Through the formula: The predicted results are updated to calculate the optimal state estimate and the updated state covariance matrix at the current time.

[0018] in, Let Kalman gain be at time k. The measurement matrix maps the system state to the measurement space. The covariance matrix of the measurement noise represents the uncertainty in the measurement process. This is the optimal state estimate at time k. Let k be the measured value at time k. The updated state covariance matrix, It is an identity matrix.

[0019] In one possible implementation, the weighting and combining of the modified multi-channel acoustic signals to obtain the target single-channel signal includes:

[0020] The weight parameters of each channel are determined by the sinc function. Specifically, based on the number of channels m of the corrected multi-channel acoustic signal, the first m peak points on the positive half-axis of the sinc function are selected as the weight parameters of each channel in the corrected multi-channel acoustic signal.

[0021] The modified multi-channel acoustic signals are weighted and summed using the weight parameters to obtain the target single-channel signal.

[0022] In one possible implementation, the step of performing power frequency discharge detection on the target single-channel signal to obtain the slight partial discharge acoustic signature detection result of the target power equipment includes:

[0023] The target single-channel signal is subjected to high-pass filtering to obtain a single-channel filtered signal;

[0024] Based on the energy of the single-channel filtered signal, a PRPD acoustic signature is plotted according to the phase distribution.

[0025] The PRPD voiceprint map is matched and identified with a preset partial discharge feature dataset, and the voiceprint detection result of the slight partial discharge is obtained based on the matching result.

[0026] In a second aspect, embodiments of the present invention provide a device for detecting slight partial discharge acoustic signatures in power equipment, comprising:

[0027] The acquisition module is used to acquire the acoustic signals of the target power equipment in a non-contact manner through a multi-channel acoustic monitoring device, and obtain multi-channel acoustic signals.

[0028] The processing module is used to perform high-pass filtering on the multi-channel acoustic signal, and to perform Kalman filtering on the high-pass filtered multi-channel acoustic signal to obtain a corrected multi-channel acoustic signal; the corrected multi-channel acoustic signal is then weighted and combined to obtain the target single-channel signal.

[0029] The detection module is used to perform power frequency discharge detection on the target single-channel signal to obtain the acoustic fingerprint detection result of slight partial discharge of the target power equipment.

[0030] Thirdly, embodiments of the present invention provide a computer device, the computer device including a processor and a non-volatile memory storing computer instructions, wherein when the computer instructions are executed by the processor, the computer device performs the method described in the first aspect.

[0031] Fourthly, a readable storage medium, characterized in that the readable storage medium includes a computer program, which, when executed, controls the computer device on which the readable storage medium is located to perform the method described in the first aspect.

[0032] Compared to existing technologies, the beneficial effects provided by this invention include: The method, apparatus, computer equipment, and readable storage medium for detecting minor partial discharge in power equipment, as disclosed in this invention, include: firstly, non-contact acoustic signal acquisition of the target power equipment using a multi-channel acoustic monitoring device to obtain multi-channel acoustic signals; secondly, high-pass filtering of the multi-channel acoustic signals and correction processing using Kalman filtering to obtain corrected multi-channel acoustic signals; thirdly, weighted summation and merging of the corrected multi-channel acoustic signals to obtain a target single-channel signal; and finally, power frequency discharge detection of the target single-channel signal to obtain the minor partial discharge acoustic signal detection result. This method enhances the minor partial discharge signal through multi-channel signal acquisition and processing, enabling the detection of minor partial discharge characteristics and providing a reference for the production and maintenance of power equipment. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic flowchart of the method for detecting slight partial discharge acoustic signatures in power equipment provided in an embodiment of the present invention;

[0035] Figure 2 A flowchart of the Kalman filter algorithm provided in an embodiment of the present invention;

[0036] Figure 3 This is a schematic diagram of a minor partial discharge detection record for an ultra-high voltage bushing provided in an embodiment of the present invention;

[0037] Figure 4(a) shows the acoustic signature detection of slight partial discharge in an ultra-high voltage bushing according to an embodiment of the present invention;

[0038] Figure 4(b) shows another case of acoustic fingerprint detection for slight partial discharge in ultra-high voltage bushings provided by an embodiment of the present invention;

[0039] Figure 4(c) shows another case of acoustic fingerprint detection for slight partial discharge in ultra-high voltage bushings provided by an embodiment of the present invention;

[0040] Figure 5 A schematic block diagram of the structure of the acoustic signature detection device for minor partial discharge of power equipment provided in an embodiment of the present invention;

[0041] Figure 6 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0043] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0044] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating the acoustic signature detection method for minor partial discharge in power equipment provided in this embodiment. The method will now be described in detail.

[0045] Step S201: Non-contact acoustic signal acquisition of the target power equipment is performed using a multi-channel acoustic monitoring device to obtain multi-channel acoustic signals;

[0046] Step S202: High-pass filtering is performed on the multi-channel acoustic signal, and Kalman filtering is used to correct the high-pass filtered multi-channel acoustic signal to obtain the corrected multi-channel acoustic signal.

[0047] Step S203: Weight the modified multi-channel acoustic signals and merge them to obtain the target single-channel signal;

[0048] Step S204: Perform power frequency discharge detection on the target single-channel signal to obtain the slight partial discharge acoustic signature detection result of the target power equipment.

[0049] In this embodiment of the invention, the target power equipment includes, but is not limited to, transformers, transformer bushings, transformer cores, transformer windings, and transformer on-load tap changers.

[0050] For example, this embodiment uses the transformer bushing of a 220kV substation as the detection object. As a key component of high-voltage insulation, the presence of even a slight partial discharge as low as 2.5pC inside this device may accelerate insulation aging, requiring early warning through acoustic fingerprint detection. The server, as the core control unit, first establishes a communication connection with the multi-channel acoustic fingerprint monitoring device via an RS485 bus. This monitoring device is an acoustic fingerprint sensor array that supports synchronous acquisition and digital signal output. The number of channels in the acoustic fingerprint sensor array can be set according to actual needs. In a specific implementation, a 4-channel acoustic fingerprint sensor array can be used, such as the UA-160 ultrasonic sensor, which has a sensitivity of -45dB±3dB and a frequency response covering 0.02kHz to 100kHz.

[0051] Based on the target power equipment and the site environment, the server automatically generates a sensor layout scheme: four sensors are evenly installed around the outer circumference of the bushing flange, with adjacent sensors spaced 90° apart to fully cover the sound radiation area; the distance between the sensors and the outer wall of the bushing meets the non-contact requirements for power equipment monitoring scenarios. Generally, the non-contact requirement for power equipment monitoring scenarios is a distance of not less than 0.5m between the sensor and the target power equipment. In specific implementation, the distance between the sensor and the outer wall of the bushing can be set to 0.8m, and the sensor installation height can be 1.2m (higher than 1m to reduce ground reflection noise). The sensors are fixed to the substation steel structure support with magnetic bases to avoid mechanical vibration interference; the target power equipment includes a transformer bushing with a height of 3.2m and a diameter of 0.4m; the background noise is generally 62dB, with no strong electromagnetic interference.

[0052] The acquisition parameters are preset and distributed by the server: sampling rate 48kHz (in this embodiment, the sampling rate meets the requirement of not less than 48kHz), sampling precision 16bit, single-channel data format is PCM pulse code modulation, and 48,000 sampling points are generated per channel per second. The server receives the 4-channel raw acoustic signal in real time via Ethernet (where Ethernet can use TCP / IP protocol, bandwidth 100Mbps), and the signal of each channel is stored in the form of a time series array on a local SSD (where the local SSD has a capacity of 1TB and a read / write speed of 500MB / s). For example, the sampling value of channel 1 at 0.001 seconds is -1234 (that is, the sampling value obtained by 16-bit quantization of the sound pressure amplitude of channel 1 at 0.001 seconds, -1234 represents the sound pressure amplitude), realizing the synchronous acquisition and storage of multi-channel acoustic signals.

[0053] The server preprocesses the 4-channel raw signal in two steps. The core is to filter out low-frequency interference and correct signal distortion in order to highlight the high-frequency acoustic characteristics of slight discharge.

[0054] Step 1: High-pass filtering to extract effective frequency bands. The server calls a preset 7th-order Butterworth high-pass filter (where the cutoff frequency of the 7th-order Butterworth high-pass filter is 3kHz, the passband ripple is 1dB, and the stopband attenuation is 60dB@1kHz, that is, at the 1kHz frequency point, the signal amplitude is attenuated by 60dB) to filter each channel signal independently.

[0055] Taking channel 1 signal as an example, the server converts the time-domain signal to the frequency domain using a Fast Fourier Transform (FFT) to identify low-frequency components below 3kHz. These low-frequency components, such as transformer cooling fan vibration noise, range from 50Hz to 2kHz. A filter attenuates this frequency band, retaining high-frequency components above 3kHz. Since partial discharge acoustic signatures are mainly distributed between 10kHz and 100kHz, attenuating low-frequency components below 3kHz reduces signal interference while preserving effective acoustic signature data, thus reducing the amount of signal required for subsequent signal processing and improving processing efficiency. The filtered signal yields four channels of high-frequency signal, stored as 32-bit floating-point data to avoid truncation errors.

[0056] Step 2: Kalman Filter Dynamic Signal Correction. Based on the dynamic characteristics of the partial discharge signal (which include pulse-like and weak amplitude fluctuations), the server performs Kalman filtering correction on the 4-channel high-frequency signal.

[0057] Taking the high-frequency signal of channel 1 as an example, the server initializes the system parameters, which include: the state transition matrix, the covariance matrix of the process noise, the measurement matrix, and the covariance matrix of the policy noise. The state transition matrix describes the evolution of the system state from the previous time step to the current time step. The state transition matrix is ​​set as a 2x2 matrix, with the first row containing 1s and the sampling interval (for example, the sampling interval can be 1 / 48000 ≈ 2.08 × 10⁻). 5 (seconds), the second row contains 0s and 1s; the covariance matrix of the process noise reflects the uncertainty in the state transition, and the diagonal elements are set to 10. -6 and 10 -8 The measurement matrix maps the system state to the measurement space and is set as a 1x2 matrix [1,0]. The covariance matrix of the measurement noise characterizes the measurement error and can be set based on sensor calibration data, for example, it can be set to 5*10. -5 .

[0058] Kalman filtering is performed through a “prediction-update” iterative process: In the prediction phase, the server calculates the predicted state value and the predicted state covariance matrix at the current time based on the state estimate and the process noise covariance matrix of the previous time step, combined with the state transition matrix; in the update phase, the Kalman gain is calculated based on the actual measurement value at the current time step, and the predicted state value is adjusted with the gain to obtain a more accurate current state estimate, and the state covariance matrix is ​​updated.

[0059] The server uses multi-threading technology to allocate four CPU cores for parallel processing of the four-channel signal, ultimately obtaining the corrected four-channel acoustic signal. The signal-to-noise ratio is improved from 12dB before filtering to 28dB, effectively suppressing the masking of slight discharge signals by background noise.

[0060] To improve signal energy concentration, the server needs to merge the 4-channel correction signal into a single-channel signal. The key is to highlight the characteristics of the effective channel through weighting.

[0061] The server first performs short-time energy analysis on the four-channel corrected signals to identify the energy proportion of the discharge pulse in each channel. Since the distance between the sensor and the discharge point varies, channels with higher energy correspond to directions closer to the discharge point and should be assigned higher weights. Based on this, the server uses the sinc function (where the sinc function curve takes a maximum value of 1 at the origin, and the magnitude decreases periodically as the independent variable increases) to generate weight parameters. Given the four channels, the server takes the function values ​​of the first four non-zero points on the positive half-axis of the sinc function as the basis for the weights. After normalization, the weights for the four channels can be, for example, 0.32, 0.27, 0.22, and 0.29, assigning higher weights to channels with higher energy.

[0062] The server performs a weighted summation on the 4-channel signals: the target single-channel signal equals the sum of the corrected signals of each channel multiplied by their respective weights. For example, at 0.1 seconds, the corrected signal values ​​of the 4 channels are 2150, 1820, 1540, and 1310, respectively. After weighted summation, the single-channel signal value is 2150×0.32 + 1820×0.27 + 1540×0.22 + 1310×0.19 = 1786. The combined single-channel signal has 3dB more energy than the single-channel signal, and its pulse characteristics are more pronounced.

[0063] The server performs power frequency discharge detection on a single-channel signal in three steps. The core of the process is to identify the discharge type and intensity by matching the phase-resolved partial discharge (PRPD) spectrum.

[0064] Step 1: Secondary high-pass filtering to purify the signal. The server uses a 7th-order Butterworth high-pass filter (with a cutoff frequency of 10kHz, passband ripple of 1dB, and stopband attenuation of 80dB@5kHz) to further filter the single-channel signal, completely removing residual interference below 10kHz to obtain a high-frequency filtered signal; the noise signal below 10kHz may include, but is not limited to, power grid harmonic noise.

[0065] Step 2: PRPD spectrum generation. The server, based on the 50Hz power grid frequency, divides the high-frequency filtered signal into time windows of 0.02 seconds per power frequency cycle. The signal energy distribution within each cycle is statistically analyzed by phase (0°-360°): Synchronizing the power grid voltage phase with a phase-locked loop, the signal energy corresponding to each phase angle is calculated and plotted as a scatter plot (where the horizontal axis represents phase and the vertical axis represents energy). After accumulating 100 cycles, a PRPD acoustic signature spectrum is formed. Slight discharge pulses will form dense clusters of scatter points in specific phase intervals (such as near the peak power frequency voltage), while background noise is randomly distributed.

[0066] Step 3: Spectrum Matching and Discharge Identification. The server compares the generated PRPD spectrum with a pre-set dataset (containing PRPD templates for typical discharges such as levitation, surface discharge, and corona discharge, constructed through laboratory simulation) to extract spectrum feature parameters. The pre-set dataset includes, but is not limited to, PRPD templates for typical discharges such as levitation discharge, surface discharge, and corona discharge, constructed through laboratory simulation. Spectrum feature parameters include, but are not limited to, phase distribution width, maximum energy point phase, and scatter density. Based on the extracted spectrum feature parameters and the levitation discharge templates in the pre-set dataset, a comprehensive analysis of the discharge characteristics is performed using calibration curves. The characteristics of the levitation discharge templates include: phase concentration within ±90° and symmetrical scatter distribution.

[0067] The server generates a final report based on the analysis results above. The report may include information such as "the target bushing has a slight floating discharge, with the discharge level at 2.5 pC". The report is then pushed to the terminal through the operation and maintenance platform to provide an early warning of slight partial discharge.

[0068] This embodiment transforms the 2.5pC level discharge signal, which cannot be identified by traditional ultrasonic testing, into quantifiable PRPD features (consistent with high-precision partial discharge detection) through a full process of multi-channel acquisition, filtering correction, weight fusion, and power frequency detection. This improves the detection sensitivity by one order of magnitude and provides key technical support for equipment insulation assessment.

[0069] In this embodiment of the invention, the high-pass filtering of the multi-channel acoustic signal can be performed through the following example.

[0070] The multi-channel acoustic signal was high-pass filtered using a 7th-order Butterworth high-pass filter to obtain the high-pass filtered multi-channel acoustic signal.

[0071] In an embodiment of the invention, for example, the server uses the 4-channel raw acoustic signature signal of a transformer bushing in a 220kV substation as the processing object (wherein, the 4-channel raw acoustic signature signal has been stored as a time series array with a sampling rate of 48kHz and a precision of 16bit), and calls a preset signal processing module to perform high-pass filtering. This signal processing module loads a 7th-order Butterworth high-pass filter, whose parameters are read from the server's local configuration file. These parameters may include a cutoff frequency of 3kHz, passband ripple ≤1dB, and stopband attenuation ≥60dB@1kHz, thereby ensuring that low-frequency signals below 1kHz are attenuated to less than 1% of their original amplitude.

[0072] The server processes the 4-channel raw acoustic signature signals in parallel: For each channel's raw acoustic signature signal, the frequency selectivity of the filter attenuates low-frequency interference components below 3kHz, retaining only high-frequency acoustic signals above 3kHz. The raw acoustic signature signals may include, but are not limited to, transformer cooling fan vibration noise (50Hz-2kHz); steel structure resonance noise (2.5kHz); and high-frequency partial discharge signals (10kHz-50kHz).

[0073] For example, the original signal amplitude in channel 2 at a certain moment is -1560 (including 2kHz fan noise). After filtering, the amplitude at that moment is corrected to -890. The above filtering process can achieve a noise component attenuation of about 43% and a high-frequency signal retention rate of ≥95%. It can be seen that the filtering method adopted in this application can better retain effective data and provide more accurate data support for subsequent analysis.

[0074] After the above filtering process is completed, the server stores the high-pass filtered signals of the 4 channels as a 32-bit floating-point array (wherein, in the time domain waveform of each channel signal, low-frequency fluctuations are significantly reduced, and high-frequency pulse characteristics (wherein, the pulse width of the high-frequency pulse characteristics is ≤0.1ms) are initially revealed, providing high-quality input data for subsequent Kalman filter correction).

[0075] In this embodiment of the invention, the step of using Kalman filtering to correct the high-pass filtered multi-channel acoustic signal to obtain a corrected multi-channel acoustic signal can be implemented through the following example.

[0076] The state estimation of the high-pass filtered multi-channel acoustic signal is performed recursively through prediction and update steps. The prediction step predicts the state at the current time based on the state estimate at the previous time step, and the update step calculates the optimal state estimate at the current time step by combining the predicted value and the actual measurement value. The prediction and update steps are repeated to correct the multi-channel acoustic signal.

[0077] In an embodiment of the present invention, for example, the server takes the high-pass filtered multi-channel acoustic signal as input and calls the Kalman filter module to recursively perform state estimation on each channel signal.

[0078] Taking one of the acoustic signal channels as an example, the server first initializes the system parameters, including the state transition matrix for describing the evolution of the signal state, the process noise covariance matrix for reflecting the uncertainty of the state transition, the measurement matrix for observing the signal state, and the measurement noise covariance matrix set based on the actual application scenario.

[0079] In the prediction step, the server predicts the current state estimate based on the state estimate of the previous time step using the state transition matrix. In the update step, the server calculates the optimal state estimate of the current time step by combining the current state estimate with the actual measurement value, and updates the state-related state covariance matrix synchronously.

[0080] The server executes the above-mentioned prediction and update steps in parallel on the multi-channel acoustic signals in a recursive process, continuously correcting fluctuations in the signals caused by noise and other factors, thereby achieving correction processing of the multi-channel acoustic signals.

[0081] In this embodiment of the invention, the prediction step includes:

[0082] Through the formula: The high-pass filtered multi-channel acoustic signal is then predicted to obtain the current state prediction estimate and the current state covariance prediction value.

[0083] in, Let k be the predicted estimate of the state at time k. Here is the state transition matrix. The control gain matrix is ​​used to measure the control input applied at time k. Mapping to the system state space, the degree of influence of control inputs on the predicted system state values ​​is quantified. The state estimate at time (k-1) is... Let k be the control variable, representing the control input applied to the system at time k. Let k be the predicted value of the state covariance at time k. This is the estimated state covariance at time (k-1). Let be the covariance matrix of the process noise, representing the uncertainty in the system state transition process.

[0084] In an embodiment of the present invention, for example, the server uses the high-pass filtered signal of channel 1 as the processing object, and has already obtained the state estimate at time (k-1). (For example, the obtained state estimate includes two state variables: amplitude 2100 and rate of change 50) and covariance matrix. (The diagonal elements correspond to the covariance of the magnitude and rate of change, respectively). The server calls the preset state transition matrix. The predicted estimate of the state at time k is calculated using matrix multiplication. and Multiply to obtain the first intermediate result, then, by controlling the gain matrix... and To obtain a second intermediate result, the control variables are multiplied together. The first intermediate result is then added to the second intermediate result to obtain the predicted estimate of the state at the current time k. .

[0085] At the same time, the server calculates the predicted state covariance value. : call ,and and its transpose matrix (where the transpose matrix has elements of 1 and 0 in the first column and elements of 2.08 × 10⁻ in the second column). 5 1) Perform matrix multiplication to obtain intermediate results, and then superimpose the process noise covariance matrix. (The diagonal elements are 1e-6 and 1e-8, representing the uncertainty of state transition), finally yielding The diagonal elements are 2e-6 (amplitude covariance) and 2e-8 (rate of change covariance). The server will... and Stored in a memory buffer as input for the update step.

[0086] In this embodiment of the invention, the update step includes:

[0087] Through the formula: The predicted results are updated to calculate the optimal state estimate and the updated state covariance matrix at the current time.

[0088] in, Let Kalman gain be at time k. The measurement matrix maps the system state to the measurement space. The covariance matrix of the measurement noise represents the uncertainty in the measurement process. This is the optimal state estimate at time k. Let k be the measured value at time k. The updated state covariance matrix, It is an identity matrix.

[0089] In an embodiment of the present invention, for example, the server takes the prediction result of channel 1 as input: the state prediction estimate at time k. (For example, (Amplitude 2100.0104, rate of change 50), covariance matrix (For example, The diagonal elements are 2e-6 and 2e-8, and the current measurement value. (in, This represents the actual sampled value at time k (for example, the actual sampled value at time k could be 2150). The server calls the measurement matrix. =[1,0] (where the measurement matrix only extracts the amplitude state) and the covariance matrix of the measurement noise. =5e-5 (where the measurement noise covariance matrix is ​​used to indicate sensor measurement error), perform the update step.

[0090] First, calculate the Kalman gain. The server calculates first. and The product (taking the amplitude covariance 2e-6), then multiplied by Transpose (the result is still 2e-6), plus We get 5.2e-5, which, after inversion, is approximately 19230.77; then... and The product of the transposes ([2e-6,0]) T Multiplying this by the inverse value yields... ≈0.6 (Amplitude gain only, rate of change gain is 0).

[0091] Next, the state estimate is corrected: the server calculates the measurement residuals. - • =2150-2100.0104≈49.9896, multiplied by The correction is approximately 30, which is then added to the predicted amplitude of 2100.0104 to obtain the optimal estimate. (Amplitude 2130, rate of change 50).

[0092] Finally, update the covariance matrix. Server constructs identity matrix I and • The difference (diagonal elements [0.4,1]) and Multiplying them yields updated diagonal elements of 8e-7 (amplitude covariance) and 2e-8 (rate of change covariance). The server will... Store the optimal state estimate at time k. This is the updated state covariance matrix, used for prediction at the next time step.

[0093] In this embodiment of the invention, the step of weighting the modified multi-channel acoustic signal and merging it to obtain the target single-channel signal can be implemented through the following example.

[0094] The weight parameters of each channel are determined by the sinc function. Based on the number of channels m of the corrected multi-channel acoustic signal, the first m peak points on the positive half axis of the sinc function are selected as the weight parameters of each channel in the corrected multi-channel acoustic signal, where m is a positive integer.

[0095] The modified multi-channel acoustic signals are weighted and summed using the weight parameters to obtain the target single-channel signal.

[0096] In this embodiment of the invention, for example, the server takes a modified 4-channel acoustic signal (the 4 channels are denoted as Xk1, Xk2, Xk3, and Xk4, with a sampling rate of 48kHz and a 32-bit floating-point type) as input and performs weighted summation and merging. The server first calls the signal analysis module to confirm the number of channels m=4, and calculates the weight parameters of each channel using the sinc function according to preset rules: the sinc function is defined as sin(πx) / (πx). The server selects the x values ​​corresponding to the first four non-zero peak points of the positive half-axis (x=0.25, 0.5, 0.75, 1, with an interval of 0.25 to cover the weight distribution of the 4 channels), and calculates the sinc value for each: sinc(0.25)=sin(π×0.25) / (π×0.25). 5)=sin(π / 4) / (π / 4)≈0.9003, sinc(0.5)=sin(π / 2) / (π / 2)=1 / (π / 2)≈0.6366, sinc(0.75)=sin(3π / 4 ) / (3π / 4)≈0.4244, sinc(1)=sin(π) / π=0 (after correcting to a non-zero point, x=0.9, sinc(0.9)=sin(0.9π) / (0.9π)≈0.2801).

[0097] The server normalizes the above four sinc values ​​(0.9003, 0.6366, 0.4244, 0.2801): the sum = 0.9003 + 0.6366 + 0.4244 + 0.2801 ≈ 2.2414, and the weight of each channel = individual sinc value / sum, so w1 ≈ 0.402, w2 ≈ 0.284, w3 ≈ 0.189, w4 ≈ 0.125 (sum = 1).

[0098] The server then performs a weighted summation on the four-channel signals: for the corrected signal value at each sampling time t (e.g., when t=0.1s, Xk1=2150, Xk2=1820, Xk3=1540, Xk4=1310), the target single-channel signal value Xs(t) is calculated as follows: Xs(t) = w1×Xk1 + w2×Xk2 + w3×Xk3 + w4×Xk4 = 0.402×2150 + 0.284×1820 + 0.189×1540 + 0.125×1310 ≈ 864.3 + 516.88 + 291.06 + 163.75 ≈ 1836.

[0099] The server processes all sampling moments in parallel using multiple threads (48,000 moments per second at a 48kHz sampling rate), and stores the merged target single-channel signal Xs in 32-bit floating-point format. After weighting, the signal energy is increased by an average of 4dB compared to a single channel, and the peak signal-to-noise ratio of the partial discharge pulse is increased from 28dB to 33dB, providing a high signal-to-noise ratio input for subsequent power frequency discharge detection.

[0100] In this embodiment of the invention, the step of performing power frequency discharge detection on the target single-channel signal to obtain the slight partial discharge acoustic signature detection result of the target power equipment includes:

[0101] The target single-channel signal is subjected to high-pass filtering to obtain a single-channel filtered signal;

[0102] Based on the energy of the single-channel filtered signal, a PRPD acoustic signature is plotted according to the phase distribution.

[0103] The PRPD voiceprint map is matched and identified with a preset partial discharge feature dataset, and the voiceprint detection result of the slight partial discharge is obtained based on the matching result.

[0104] In an embodiment of the present invention, for example, the server uses the merged target single-channel signal Xs (sampling rate 48kHz, 32-bit floating point) as input to perform power frequency discharge detection.

[0105] First, the server calls a 7th-order Butterworth high-pass filter (where the cutoff frequency of the 7th-order Butterworth high-pass filter is 10kHz, the passband ripple is ≤1dB, and the stopband attenuation is ≥80dB@5kHz) to filter Xs, filtering out residual interference below 10kHz (where residual interference is such as 50Hz harmonic noise from the power grid), and obtains a single-channel filtered signal Xsh, whose high-frequency pulse characteristics (10kHz-50kHz) are further highlighted.

[0106] Next, the PRPD acoustic signature was plotted: The server, based on a 50Hz power frequency (0.02s cycle), divided the Xsh pulses into time windows of 0.02s per power frequency cycle. Synchronizing with the grid voltage phase using a phase-locked loop (1° resolution), the signal energy corresponding to each phase angle (0°-360°) was calculated (where signal energy is represented by the square integral value), and plotted as a scatter plot (where the horizontal axis represents phase and the vertical axis represents energy). After accumulating 100 cycles (2s data), the partial discharge pulses in the signature formed dense clusters of scatter points within the ±90° phase range (background noise was randomly distributed).

[0107] Finally, the server compares the generated PRPD map with a pre-set dataset (containing PRPD templates of typical discharges such as suspension, surface discharge, and corona discharge, constructed through laboratory simulation). It extracts map feature parameters (phase distribution width, maximum energy point phase, and scatter density) and calculates a cosine similarity of 0.89 (threshold 0.85) with the "suspension discharge template" (features: phase concentrated at ±90°, symmetrical scatter distribution). The results are then combined with calibration curves for comprehensive analysis of the discharge characteristics. The server ultimately generates a report: "The target bushing exhibits a slight suspension discharge, with a discharge level of 2.5 pC," which is pushed to the terminal via the operations and maintenance platform, providing an early warning of slight partial discharge.

[0108] To more clearly describe the solutions provided in the embodiments of the present invention, a more complete implementation method is provided below.

[0109] (i) Deploy multi-channel acoustic fingerprint monitoring devices (2 channels or more, sampling rate 48kHz or higher) around power equipment to collect acoustic fingerprint signals and obtain multi-channel acoustic signals; (ii) Perform high-pass filtering on the multi-channel acoustic signals (take signals above 3kHz), and use Kalman filtering to evaluate and correct the high-pass filtered multi-channel acoustic signals; (iii) Add weights to the corrected multi-channel acoustic signals (the weighting method is different for different numbers of channels), and merge them into a single-channel signal; (iv) Apply the power frequency discharge detection method to the single-channel signal: first, apply high-pass filtering to the single-channel signal (take signals above 10kHz), and then draw a PRPD acoustic fingerprint spectrum according to the phase distribution of the energy of the high-pass filtered signal (the power frequency of my country's power grid is 50Hz, so it needs to be drawn once every 0.02s of data, and at least 100 0.02s of data need to be drawn), and then match and identify the PRPD acoustic fingerprint spectrum with the dataset to provide a reference for the production and operation of power equipment.

[0110] in:

[0111] (a) Signal acquisition method: A multi-channel acoustic signature monitoring device with 2 or more channels and a sampling rate of 48kHz or higher should be placed next to the target power equipment, facing the target power equipment without contact, at a distance of 0.5m or more and at a height of 1m or more, to acquire the multi-channel acoustic signals of the target power equipment in real time. .

[0112] (ii) Kalman filtering: A 7th-order Butterworth high-pass filter (cutoff frequency 3kHz) is used to process the multi-channel acoustic signals. Perform high-pass filtering (i.e., take signals above 3kHz) to obtain multi-channel filtered signals. Then, Kalman filtering is used to filter the multi-channel signal. The evaluation and correction were performed to obtain the corrected multi-channel acoustic signal. Its working principle involves two recursive steps: prediction and update. In the prediction phase, the filter uses the system's dynamic model to predict the system variables for the next state; in the update phase, the filter combines the predicted values ​​and actual measurements to calculate the optimal estimate of the current state.

[0113]

[0114] The specific algorithm flow is as follows:

[0115] 1) Prediction

[0116] In the prediction step, the state at the current moment is predicted based on the state and control variables at the previous moment. This predicted value is an estimate because it does not yet take into account the observations at the current moment; the error covariance matrix of the predicted value is calculated using the error covariance matrix at the previous moment and the system noise covariance matrix.

[0117]

[0118] in, Let be the predicted estimate of the state at time k; This is the state transition matrix, which describes how the system state evolves from the previous time step to the current time step; This is the state estimate at time (k-1); Let k be the control variable, representing the control input applied to the system at time k. Let be the predicted value of the state covariance at time k; This is the estimated state covariance at time (k-1); Let be the covariance matrix of the process noise, representing the uncertainty in the system state transition process.

[0119] 2) Update

[0120] In the update step, the state estimate for the current moment is calculated based on the observed and predicted values. This estimate is more accurate. The error covariance matrix of the state estimate is calculated using the error covariance matrix obtained in the prediction step, the covariance matrix of the measurement noise, and the Kalman gain.

[0121]

[0122] in, Let Kalman gain be the value at time k. The measurement matrix maps the system state to the measurement space. The covariance matrix of the measurement noise represents the uncertainty in the measurement process; This is the optimal state estimate at time k; The measured value at time k; This is the updated state covariance matrix; It is an identity matrix.

[0123] 3) Repeat the first two steps to continuously obtain the optimal state estimate for the current moment. Please refer to the relevant documentation. Figure 2 , Figure 2 The flowchart of the Kalman filter algorithm provided in the embodiment of the present invention is shown.

[0124] (iii) Weighting: The weighting of the modified multi-channel acoustic signal Weighting is applied to merge the signals into a single channel. The implementation method involves using the sinc function as the weighting parameter and adjusting it based on the corrected multi-channel acoustic signal. Let m be the number of channels, and let the m peak points before the positive half-axis of the sinc function be respectively set as The weight parameters of each channel are added together and merged into a single-channel signal. The formula for the sinc function is as follows:

[0125] (iv) Power frequency discharge detection: for single-channel signals Application of power frequency discharge detection method: Using a 7th order Butterworth high-pass filter (cutoff frequency 10kHz) to detect single-channel signals. Perform high-pass filtering (i.e., take signals above 10kHz) to obtain a single-channel filtered signal. ; Filter the single-channel signal The energy is used to draw PRPD acoustic signatures according to the phase distribution (the standard power grid frequency is 50Hz, so it needs to be drawn once every 0.02s of data, and at least 100 0.02s of data need to be drawn). Then, the PRPD acoustic signatures are matched and identified with the dataset to provide a reference for the production and operation of power equipment.

[0126] Please refer to the following: Figure 3 Figures 4(a)-(c) show the actual discharge detection records of a certain UHV bushing. The partial discharge acoustic sound detection method of power equipment combined with Kalman filtering used in this application successfully detected partial discharge characteristics as low as 2.5pC outside the flange of the UHV bushing, and matched the floating discharge characteristics (the ultrasonic detection method did not find any obvious discharge characteristics at the same time).

[0127] Please refer to the following: Figure 5 , Figure 5 A power equipment slight partial discharge acoustic signature detection device 110 provided in this embodiment of the invention includes:

[0128] The acquisition module 1101 is used to acquire the acoustic signals of the target power equipment in a non-contact manner through a multi-channel acoustic monitoring device to obtain multi-channel acoustic signals.

[0129] The processing module 1102 is used to perform high-pass filtering on the multi-channel acoustic signal, and to perform correction processing on the high-pass filtered multi-channel acoustic signal using Kalman filtering to obtain a corrected multi-channel acoustic signal; and to perform weight addition on the corrected multi-channel acoustic signal and merge it to obtain the target single-channel signal.

[0130] The detection module 1103 is used to perform power frequency discharge detection on the target single-channel signal to obtain the slight partial discharge acoustic pattern detection result of the target power equipment.

[0131] It should be noted that the implementation principle of the aforementioned power equipment slight partial discharge acoustic fingerprint detection device 110 can refer to the implementation principle of the aforementioned power equipment slight partial discharge acoustic fingerprint detection method, and will not be repeated here. It should be understood that the division of the various modules in the above device is merely a logical functional division; in actual implementation, they can be fully or partially integrated into a single physical entity, or physically separated. Furthermore, these modules can all be implemented in software through processing element calls; they can all be implemented in hardware; or some modules can be implemented by processing element calls to software, and some modules can be implemented in hardware. For example, the power equipment slight partial discharge acoustic fingerprint detection device 110 can be a separately established processing element, or it can be integrated into a chip in the aforementioned device. Alternatively, it can be stored as program code in the memory of the aforementioned device, and called and executed by a processing element of the aforementioned device. The implementation of other modules is similar. Furthermore, these modules can be fully or partially integrated together, or implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step or module of the above method can be completed by the integrated logic circuit in the hardware of the processor element or by instructions in the form of software.

[0132] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more application-specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to implement a system-on-a-chip (SOC).

[0133] This invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned power equipment slight partial discharge acoustic signature detection device 110. Figure 6 As shown, Figure 6This is a structural block diagram of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a power equipment slight partial discharge acoustic signature detection device 110, a memory 111, a processor 112, and a communication unit 113.

[0134] To enable data transmission or interaction, the memory 111, processor 112, and communication unit 113 are electrically connected to each other directly or indirectly. For example, these components can be electrically connected to each other through one or more communication buses or signal lines. The power equipment slight partial discharge acoustic fingerprint detection device 110 includes at least one software function module that can be stored in the memory 111 or embedded in the operating system (OS) of the computer device 100 in the form of software or firmware. The processor 112 is used to execute the power equipment slight partial discharge acoustic fingerprint detection device 110 stored in the memory 111, such as the software function module and computer program included in the power equipment slight partial discharge acoustic fingerprint detection device 110.

[0135] This invention provides a readable storage medium, which includes a computer program. When the computer program runs, it controls the computer device where the readable storage medium is located to execute the aforementioned power equipment slight partial discharge acoustic print detection device 110.

[0136] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A method for detecting slight partial discharge in power equipment using acoustic signatures, characterized in that, include: Non-contact acoustic signal acquisition of target power equipment is performed using multi-channel acoustic monitoring equipment to obtain multi-channel acoustic signals; The multi-channel acoustic signal is subjected to high-pass filtering, and the high-pass filtered multi-channel acoustic signal is corrected by Kalman filtering to obtain the corrected multi-channel acoustic signal. The modified multi-channel acoustic signals are weighted and then combined to obtain the target single-channel signal; Power frequency discharge detection is performed on the target single-channel signal to obtain the acoustic fingerprint detection result of slight partial discharge of the target power equipment. The step of weighting and combining the modified multi-channel acoustic signals to obtain the target single-channel signal includes: The weight parameters of each channel are determined by the sinc function. Based on the number of channels m of the corrected multi-channel acoustic signal, the first m peak points on the positive half axis of the sinc function are selected as the weight parameters of each channel in the corrected multi-channel acoustic signal, where m is a positive integer. The modified multi-channel acoustic signals are weighted and summed using the weighting parameters to obtain the target single-channel signal.

2. The method according to claim 1, characterized in that, The high-pass filtering of the multi-channel acoustic signal includes: The multi-channel acoustic signal was high-pass filtered using a 7th-order Butterworth high-pass filter to obtain the high-pass filtered multi-channel acoustic signal.

3. The method according to claim 1, characterized in that, The process of correcting the high-pass filtered multi-channel acoustic signal using Kalman filtering to obtain the corrected multi-channel acoustic signal includes: The state estimation of the high-pass filtered multi-channel acoustic signal is performed recursively through prediction and update steps. The prediction step predicts the state at the current time based on the state estimate at the previous time step, and the update step calculates the optimal state estimate at the current time step by combining the predicted value and the actual measurement value. The prediction and update steps are repeated to correct the multi-channel acoustic signal.

4. The method according to claim 3, characterized in that, The prediction steps include: Through the formula: The high-pass filtered multi-channel acoustic signal is then predicted to obtain the current state prediction estimate and the current state covariance prediction value. in, Let k be the predicted estimate of the state at time k. Here is the state transition matrix. For the control gain matrix, The state estimate at time k-1, Let k be the control variable, representing the control input applied to the system at time k. Let k be the predicted state covariance value at time k. Let k be the estimated state covariance at time k-1. Let be the covariance matrix of the process noise, representing the uncertainty in the system state transition process.

5. The method according to claim 4, characterized in that, The update steps include: Through the formula: The predicted results are updated to calculate the optimal state estimate and the updated state covariance matrix at the current time. in, Let Kalman gain be at time k. The measurement matrix maps the system state to the measurement space. The covariance matrix of the measurement noise represents the uncertainty in the measurement process. This is the optimal state estimate at time k. Let k be the measured value at time k. The updated state covariance matrix, It is an identity matrix.

6. The method according to claim 1, characterized in that, The step of performing power frequency discharge detection on the target single-channel signal to obtain the slight partial discharge acoustic signature detection result of the target power equipment includes: The target single-channel signal is subjected to high-pass filtering to obtain a single-channel filtered signal; Based on the energy of the single-channel filtered signal, a PRPD acoustic signature is plotted according to the phase distribution. The PRPD voiceprint spectrum is matched and identified with a preset partial discharge feature dataset, and the voiceprint detection result of the slight partial discharge is obtained based on the matching result.

7. A device for detecting slight partial discharge in power equipment using acoustic signature, characterized in that, include: The acquisition module is used to acquire the acoustic signals of the target power equipment in a non-contact manner through a multi-channel acoustic monitoring device, and obtain multi-channel acoustic signals. The processing module is used to perform high-pass filtering on the multi-channel acoustic signal, and to perform Kalman filtering on the high-pass filtered multi-channel acoustic signal to obtain a corrected multi-channel acoustic signal; the corrected multi-channel acoustic signal is then weighted and combined to obtain the target single-channel signal. The detection module is used to perform power frequency discharge detection on the target single-channel signal to obtain the acoustic fingerprint detection result of slight partial discharge of the target power equipment; The step of weighting and combining the modified multi-channel acoustic signals to obtain the target single-channel signal includes: The weight parameters of each channel are determined by the sinc function. Based on the number of channels m of the corrected multi-channel acoustic signal, the first m peak points on the positive half axis of the sinc function are selected as the weight parameters of each channel in the corrected multi-channel acoustic signal, where m is a positive integer. The modified multi-channel acoustic signals are weighted and summed using the weighting parameters to obtain the target single-channel signal.

8. A computer device, characterized in that, The computer device includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device performs the method according to any one of claims 1-6.

9. A readable storage medium, characterized in that, The readable storage medium includes a computer program, which, when executed, controls the computer device on which the readable storage medium is located to perform the method described in any one of claims 1-6.

Images