An adaptive filtering and intelligent separation method for multi-modal partial discharge signals

Abstract

The application discloses a kind of multi-modal partial discharge signal's adaptive filtering and intelligent separation method, and the application is related to power equipment insulation state monitoring and signal processing technical field, comprising the following steps: step S1: real-time acquisition power equipment partial discharge signal, and synchronously obtain equipment operating parameter and environmental noise signal.The multi-modal partial discharge signal's adaptive filtering and intelligent separation method, by dynamic coupling degree modeling and the synergistic mechanism of double-channel confrontation type decoupling, effectively solves the mode aliasing and feature distortion problem caused by signal and noise time-frequency coupling in dynamic noise environment in traditional method.Dynamic coupling path perception module quantifies the interactive characteristics of noise and signal in real time, combined with the double-channel architecture of complex domain phase-sensitive filtering and vibration trajectory matching filtering, realizes the accurate suppression of high-frequency transient interference and low-frequency periodic noise.

Description

Technical Field

[0001] This invention relates to the field of insulation condition monitoring and signal processing technology for power equipment, specifically an adaptive filtering and intelligent separation method for multimodal partial discharge signals. Background Technology

[0002] Partial discharge detection is a core method for assessing the insulation condition of power equipment. However, in actual operating conditions, the signals collected by sensors often contain multiple discharge types and complex noise, such as the dynamic superposition of different modes of signals like corona discharge and internal discharge with high-frequency electromagnetic interference and low-frequency mechanical vibration noise. Traditional adaptive filtering methods are usually based on static noise assumptions, separating signals and noise through fixed frequency band segmentation or preset thresholds. However, in dynamic noise environments, the time-frequency coupling characteristics of signals and noise will nonlinearly shift with factors such as equipment load fluctuations and changes in ambient temperature and humidity. This dynamic coupling makes it difficult for traditional algorithms to accurately distinguish the cross-frequency bands of signals and noise. For example, high-frequency transient interference may intrude into the main frequency range of the discharge pulse, while low-frequency mechanical vibration noise may overlap with the periodic discharge signal in the time domain. After conventional filtering, the two still retain cross-interference components. Residual noise further destroys the independence of multi-mode signals, causing the time-frequency characteristics of different discharge types to overlap during intelligent separation, leading to misjudgment of discharge types or missed detection of weak signals by pattern recognition-based classification models. While existing improvement schemes attempt to enhance separation accuracy by optimizing thresholds or introducing deep learning, they have not fundamentally broken through the static noise separation framework. They cannot track the dynamic coupling relationship between noise and signal in real time, resulting in cascading error accumulation in the filtering and separation stages, which seriously affects the reliability of diagnosis. Summary of the Invention

[0003] (a) Technical problems to be solved

[0004] To address the shortcomings of existing technologies, this invention provides an adaptive filtering and intelligent separation method for multimodal partial discharge signals, solving the problem of adaptive decoupling and high-fidelity separation of multimodal partial discharge signals in dynamic noise environments.

[0005] (II) Technical Solution

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive filtering and intelligent separation method for multimodal partial discharge signals, comprising the following steps:

[0007] Step S1: Real-time acquisition of partial discharge signals from power equipment, and simultaneous acquisition of equipment operating parameters and environmental noise signals;

[0008] Step S2: Calculate the dynamic coupling index based on the time-frequency matrix of signal and noise, extract the time-frequency energy distribution through a sliding window, and generate a quantized value of dynamic coupling strength based on the inverse correlation between the high-frequency interference energy gradient and the signal main frequency gradient.

[0009] Step S3: Based on the time-series prediction results of the dynamic coupling index, the signal is divided into a high-frequency transient interference channel and a low-frequency periodic noise channel, and noise suppression is performed by complex domain phase-sensitive filtering and vibration trajectory matching filtering, respectively. Among them, the complex domain phase-sensitive filtering maps the signal to the complex domain through the transient pulse compression encoder and dynamically adjusts the filter cutoff frequency, while the vibration trajectory matching filtering constructs an adaptive notch filter bank based on the prior knowledge of the mechanical vibration spectrum of the equipment.

[0010] Step S4: Input the output signals of the high-frequency channel and the low-frequency channel into the adversarial mode separation network, and force the separated signals to meet the preset independent mode statistical distribution through the mode discriminator;

[0011] Step S5: Cross-validate the separation results based on physical feature compliance detection and multi-model classification confidence verification. If the impulse parameter exceeds the threshold or the classification confidence variance exceeds the standard, trigger parameter recalibration and re-iterate decoupling.

[0012] Preferably, the time-series prediction of the dynamic coupling index is modeled by a bidirectional gated cyclic unit to model the time-frequency mutual information entropy gradient sequence, predicting the direction of the coupling path offset between noise and signal within the next three time windows. During the calculation of the time-frequency mutual information entropy gradient, the dynamic coupling strength is identified by comparing the rate of change of the overlapping area of ​​the energy distribution between the signal's main frequency band and the noise frequency band. When the high-frequency interference energy gradient and the signal's main frequency gradient show an inverse correlation, the coupling risk is determined to be increased, triggering a pre-adjustment of the high-frequency channel filter cutoff frequency. For low-frequency vibration noise, the possibility of the coupling path penetrating to higher frequencies is predicted by analyzing the slope of the vibration signal envelope and the periodic phase difference of the discharge signal, and the notch filter parameters are optimized accordingly.

[0013] Preferably, the specific implementation of the complex domain phase-sensitive filtering includes:

[0014] Nonlinear phase compensation is applied to both the real and imaginary parts of the signal. Combined with dynamic coupling degree prediction, the cutoff frequency of the complex domain filter is offset to preserve the steep pulse leading edge and suppress cross-coupling components in high-frequency interference bands. During the complex domain mapping stage, nonlinear phase compensation is applied to both the real and imaginary parts of the signal. The compensation coefficients are dynamically adjusted based on the phase derivative of the instantaneous frequency to ensure the consistency of the high-frequency pulse waveform's delay. When high-frequency noise is detected intruding into the main frequency band, the filter's cutoff frequency coverage is expanded based on the coupling path prediction results, while the adjustment step size is limited to prevent pulse waveform passivation. For instantaneous frequency jumps caused by load abrupt changes, the phase derivative calculation window length is shortened to improve tracking sensitivity. Phase alignment verification of the real and imaginary parts is performed during the signal reconstruction stage, and compensation is iteratively executed when the deviation exceeds the limit.

[0015] Preferably, the specific implementation of the vibration trajectory matching filter includes:

[0016] The notch center frequency is dynamically generated based on the fundamental frequency of the equipment vibration and its harmonic components. The notch bandwidth is adaptively adjusted by the slope of the vibration signal envelope, and the time-domain amplitude of the residual components is corrected by Kalman filtering. The notch center frequency of the adaptive notch filter bank is dynamically generated based on the fundamental frequency of the equipment vibration and its harmonic components. The notch bandwidth is adaptively adjusted with the slope of the vibration signal envelope. When the slope exceeds a threshold, the bandwidth is expanded and the notch depth is increased; conversely, the bandwidth is compressed to focus on the main vibration frequency. The time-domain amplitude of the vibration noise is predicted by Kalman filtering, and the state equation is updated by combining the residual components of the notch at the previous moment, and the amplitude compensation coefficient is dynamically corrected. When a periodic discharge signal is detected to overlap with the vibration noise in the time domain, the delay parameters of the notch filter group are adjusted according to the phase difference to offset the overlapping interval.

[0017] Preferably, the adversarial mode separation network is implemented in the following way:

[0018] Modal discriminators are set at the output terminals of the high and low frequency channels respectively. The discriminators are preset with physical constraints that the pulse interval of the corona discharge follows a Poisson distribution and the periodicity of the internal discharge follows a Gaussian distribution.

[0019] The KL divergence is used to calculate the difference between the separation signal and the target modal distribution, and the pulse interval standard deviation is introduced as an additional constraint term. The filter parameters are then optimized inversely using an adversarial loss function. The modal discriminator is pre-set with physical statistical constraints that corona discharge follows a Poisson distribution and internal discharge follows a Gaussian distribution. The difference between the separation signal and the target distribution is quantified using KL divergence, and the pulse interval standard deviation is introduced as an additional constraint term. When the statistical characteristics deviate from the target distribution, the filter parameters are optimized inversely to restore the discharge type characteristics. For distribution shifts caused by sudden changes in environmental noise, the coefficient of variation of statistical parameters is monitored through a sliding window, triggering a robust optimization mode and widening the distribution tolerance range. For unknown modal interference, a manual annotation and model parameter update mechanism is initiated.

[0020] Preferably, the physical feature compliance detection includes:

[0021] The threshold range is set based on the quantitative relationship between the rise time of the partial discharge pulse and the size of the insulation defect. When the rise time of the separated signal pulse is less than 1 ns and there is no subsequent oscillation waveform, it is judged as a distorted signal and recalibration is triggered. The physical characteristic compliance detection module sets a dynamic threshold based on the consistency of pulse rise time, peak amplitude and half-wave width. When the parameters exceed the limit, it first checks whether the filtering intensity of the high-frequency and low-frequency channels is unbalanced and triggers parameter recalibration. The multi-model classification confidence verification module analyzes the risk of mode mixing through the confidence variance of the time domain and frequency domain classification models. When the variance exceeds the standard, it returns to the dynamic coupling degree modeling stage for re-decoupling. For abnormal signals with low historical data matching degree, the manual auxiliary annotation and model iteration optimization process is initiated.

[0022] When a sudden mechanical impact or a sudden change in temperature and humidity is detected, the system switches to a wideband notch filter mode or an anti-temperature drift mode by monitoring the instantaneous kurtosis value of the vibration signal and environmental sensor data, limiting the parameter adjustment range to prevent misoperation; when the hardware acquisition channel is abnormal, it automatically switches to the backup channel and resets the compensation parameters; when convergence still fails after multiple iterations, the model is frozen and static filtering of the historical best parameters is enabled, while triggering an alarm signal to wait for manual intervention.

[0023] Preferably, the multi-model classification confidence verification includes:

[0024] The separated signals are input in parallel into a pre-trained time-domain convolutional network and a frequency-domain graph neural network. The confidence variance of the classification results of each model is calculated. When the variance exceeds a preset threshold, it is determined that there is modal aliasing.

[0025] Preferably, the nonlinear phase compensation is achieved through the following steps:

[0026] Extract the phase derivative of the instantaneous frequency of the signal, and apply a compensation coefficient that is inversely proportional to the rate of change of the phase of the real part to the imaginary part of the complex domain signal to ensure the time delay consistency of the high-frequency pulse waveform.

[0027] Preferably, the adjustment rule for the notch bandwidth is as follows: when the slope of the vibration signal envelope exceeds a set threshold, the notch bandwidth is extended to cover the range of vibration noise harmonic diffusion; when the slope of the envelope is lower than the threshold, the notch bandwidth is compressed to the matching range of the focusing main vibration frequency.

[0028] Preferably, the optimization process of the adversarial loss function includes:

[0029] In each iteration, the distribution parameters of the mode discriminator are updated synchronously, so that the standard deviation of the pulse interval of the corona discharge converges to the theoretical variance of the Poisson distribution, and the periodic error of the internal discharge converges to the standard deviation range of the Gaussian distribution.

[0030] By using dynamic coupling quantitative modeling to perceive the interaction characteristics of noise and signal in real time, dual-channel noise suppression is achieved by using complex domain filtering and vibration trajectory matching. Then, the independence of discharge type characteristics is enhanced by adversarial mode separation. Finally, the reliability of the separation results is ensured by dual verification of physical laws and data models. A closed-loop feedback of "perception-decoupling-separation-verification" is formed between the modules, and the contradiction between signal fidelity and mode independence under dynamic noise environment is systematically resolved.

[0031] (III) Beneficial Effects

[0032] This invention provides an adaptive filtering and intelligent separation method for multimodal partial discharge signals. It offers the following advantages:

[0033] (I) This adaptive filtering and intelligent separation method for multimodal partial discharge signals effectively solves the problems of modal aliasing and feature distortion caused by the time-frequency coupling of signals and noise in dynamic noise environments through a collaborative mechanism of dynamic coupling degree modeling and dual-channel adversarial decoupling. The dynamic coupling path sensing module quantifies the interaction characteristics of noise and signals in real time, and combined with the dual-channel architecture of complex domain phase-sensitive filtering and vibration trajectory matching filtering, it achieves accurate suppression of high-frequency transient interference and low-frequency periodic noise. The adversarial mode separation network forces the separated signals to meet the independent distribution characteristics of discharge types through physical statistical constraints and adversarial training mechanisms, which significantly improves the separation accuracy and anti-interference capability of multimodal signals.

[0034] (II) This adaptive filtering and intelligent separation method for multimodal partial discharge signals solves the problem of insufficient diagnostic reliability caused by the accumulation of cascaded errors in traditional algorithms by introducing a dual guarantee mechanism of physical feature compliance detection and multi-model confidence verification. This ensures that the separation results simultaneously comply with the dual constraints of physical laws and data models. Under complex operating conditions of power equipment, this method can adaptively track the dynamic coupling changes of noise and signals, reducing false positives and false negatives. It provides highly reliable data support for early warning and accurate location of insulation defects, while reducing the need for manual intervention and improving the practicality and engineering deployment efficiency of the intelligent diagnostic system. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0036] Figure 2 This is the timing diagram of the control logic of the present invention. Detailed Implementation

[0037] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see Figure 1 and Figure 2 This invention provides a technical solution: an adaptive filtering and intelligent separation method for multimodal partial discharge signals, comprising the following steps:

[0039] Step S1: Real-time acquisition of partial discharge signals from power equipment, and simultaneous acquisition of equipment operating parameters and environmental noise signals;

[0040] Step S2: Calculate the dynamic coupling index based on the time-frequency matrix of signal and noise, extract the time-frequency energy distribution through a sliding window, and generate a quantized value of dynamic coupling strength based on the inverse correlation between the high-frequency interference energy gradient and the signal main frequency gradient.

[0041] Step S3: Based on the time-series prediction results of the dynamic coupling index, the signal is divided into a high-frequency transient interference channel and a low-frequency periodic noise channel, and noise suppression is performed by complex domain phase-sensitive filtering and vibration trajectory matching filtering, respectively. Among them, the complex domain phase-sensitive filtering maps the signal to the complex domain through the transient pulse compression encoder and dynamically adjusts the filter cutoff frequency, while the vibration trajectory matching filtering constructs an adaptive notch filter bank based on the prior knowledge of the mechanical vibration spectrum of the equipment.

[0042] Step S4: Input the output signals of the high-frequency channel and the low-frequency channel into the adversarial mode separation network, and force the separated signals to meet the preset independent mode statistical distribution through the mode discriminator;

[0043] Step S5: Cross-validate the separation results based on physical feature compliance detection and multi-model classification confidence verification. If the impulse parameter exceeds the threshold or the classification confidence variance exceeds the standard, trigger parameter recalibration and re-iterate decoupling.

[0044] The time-series prediction of the dynamic coupling degree index uses a bidirectional gated cyclic unit to model the time-frequency mutual information entropy gradient sequence, predicting the direction of the noise-signal coupling path offset within the next three time windows. Further explanation is needed: in the specific implementation, the time-series prediction of the dynamic coupling degree index uses a bidirectional gated cyclic unit to model the time-frequency mutual information entropy gradient sequence. A sliding window is used to extract the time-frequency energy distribution characteristics of the current and historical moments, capturing the coupling trend of signal and noise in the time and frequency domains. Based on the forward and backward time-series dependency analysis of the bidirectional gated cyclic unit, the direction of the noise-signal coupling path offset within the next three time windows is predicted, including two typical modes: high-frequency electromagnetic interference spreading to low frequencies or low-frequency mechanical vibration noise penetrating to high frequencies. According to the prediction results, when the coupling path spreads to low frequencies, the cutoff frequency of the complex domain filter of the high-frequency transient interference channel is adjusted in advance. To prevent intrusion into the main frequency band, the high-frequency suppression range is expanded, while the notch bandwidth of the low-frequency channel is reduced to match the slowly varying characteristics of vibration noise. When the coupling path penetrates into the high frequency range, the center frequency offset of the low-frequency notch filter is dynamically increased, and the cutoff frequency adjustment step of the high-frequency filter is reduced to ensure the fidelity of the high-frequency pulse waveform. For operating conditions with sudden load changes or rapid fluctuations in ambient temperature and humidity, the correlation between equipment operating parameters and noise signals is monitored in real time, and a short-time coupling strength correction coefficient is added to the prediction model. If the rate of change of the dynamic coupling degree index exceeds the preset threshold within two consecutive time windows, the parameters of the coupling path prediction model are updated online to avoid the failure of historical data due to sudden changes in operating conditions.

[0045] The specific implementation of complex domain phase-sensitive filtering includes:

[0046] Nonlinear phase compensation is performed on the real and imaginary parts of the signal, and the cutoff frequency of the complex domain filter is offset by combining the dynamic coupling degree prediction value, so as to preserve the steep pulse leading edge characteristics and suppress the cross-coupling components of the high-frequency interference band.

[0047] It needs further explanation that, in the specific implementation process, during the complex domain phase-sensitive filtering process, the time-domain signal of the high-frequency transient interference channel is first mapped to the complex domain. Nonlinear phase compensation is then performed on the real and imaginary parts of the signal. By extracting the phase derivative of the instantaneous frequency of the signal, the imaginary compensation coefficient is dynamically adjusted to ensure that the rate of change of the imaginary phase is inversely proportional to the rate of change of the real phase, thus ensuring the consistency of the time delay of the high-frequency pulse waveform. Based on the dynamic coupling degree prediction results, when the high-frequency interference energy gradient and the signal main frequency gradient show an enhanced inverse correlation, it is determined that the risk of high-frequency noise intruding into the main frequency band increases. At this time, the cutoff frequency of the complex domain filter is extended towards the high-frequency direction to increase the high-frequency suppression bandwidth to cover the cross-coupling frequency band. At the same time, the step size of the cutoff frequency adjustment is dynamically reduced according to the real-time monitoring value of the pulse leading edge steepness to avoid over-filtering leading to pulse waveform passivation. When the signal main frequency... When intermittent electromagnetic interference exists near the frequency band, a dynamic weighting factor is introduced into the complex domain filter based on the predicted value of the coupling path offset direction to selectively attenuate the signal components in the edge region of the main frequency band, preserving the pulse energy integrity of the core region of the main frequency. For instantaneous high-frequency noise mutations caused by load fluctuations, the correlation between the equipment current change rate and the noise spectrum offset is monitored in real time. If the current change rate exceeds the threshold and the main peak of the noise spectrum shifts to a lower frequency, the phase compensation coefficient of the complex domain filter is adaptively reset to prioritize the fidelity of the pulse leading edge characteristics. In addition, during the signal reconstruction stage, the phase alignment of the real and imaginary parts of the filtered complex domain signal is checked. If the phase deviation of the pulse rising edge exceeds the allowable range, the nonlinear compensation parameters are recalculated and the complex domain filtering is iteratively executed until the pulse waveform meets the preset time-domain distortion tolerance threshold.

[0048] The specific implementation of vibration trajectory matching filtering includes:

[0049] The notch center frequency is dynamically generated based on the fundamental frequency of the equipment vibration and its harmonic components. The notch bandwidth is adaptively adjusted by the slope of the vibration signal envelope, and the time-domain amplitude of the residual components is corrected by Kalman filtering.

[0050] It should be further explained that, in the specific implementation process, during the vibration trajectory matching filtering of the low-frequency periodic noise channel, the distribution characteristics of the fundamental frequency and its harmonic components are first extracted based on the historical data of the equipment's mechanical vibration spectrum, and a notch center frequency matching the current operating conditions is dynamically generated. Based on the real-time acquisition of the vibration signal envelope slope change, the intensity fluctuation trend of the mechanical vibration is judged. If the envelope slope exceeds a preset threshold, it is determined that the vibration noise energy is positively accumulating in the low-frequency region. At this time, the notch bandwidth is expanded to 1.5 times the fundamental frequency to cover the harmonic diffusion range. Conversely, if the slope is below the threshold, the notch bandwidth is compressed to 0.8 times the fundamental frequency to focus on the main vibration frequency. For the vibration frequency shift caused by temperature and humidity changes, a vibration noise time-domain amplitude prediction model is constructed using Kalman filtering. The state equation is updated using the correlation between the residual notch component from the previous moment and the current vibration acceleration signal, dynamically correcting the residual noise. The amplitude compensation coefficient is calculated. When periodic discharge signals and vibration noise are detected to overlap in the time domain, the group delay parameter of the notch filter is adjusted in real time according to the phase difference between the two. If the phase of the discharge signal leads the vibration noise, the phase hysteresis of the notch filter is increased to offset the overlapping interval, and vice versa to reduce the hysteresis to preserve the integrity of the discharge pulse. For vibration spectrum distortion caused by sudden mechanical shock, the instantaneous kurtosis value of the vibration signal is monitored. If the kurtosis exceeds three times the standard deviation of the normal operating condition, the wideband notch filter mode is temporarily switched and the Kalman filter correction is suspended. The adaptive tracking is resumed after the shock ends. In addition, during the signal reconstruction stage after notch filtering, the time-domain waveform of the residual component is periodically checked. If a residual oscillation component consistent with the equipment rotation period is detected, the fine-tuning mechanism of the notch filter center frequency is triggered, and the fundamental frequency value is shifted by 0.5% successively until the amplitude of the residual component drops below the noise baseline.

[0051] Adversarial mode separation networks are implemented in the following ways:

[0052] Modal discriminators are set at the output terminals of the high and low frequency channels respectively. The discriminators are preset with physical constraints that the pulse interval of the corona discharge follows a Poisson distribution and the periodicity of the internal discharge follows a Gaussian distribution.

[0053] The difference between the separated signal and the target mode distribution is calculated based on KL divergence, and the pulse interval standard deviation is introduced as an additional constraint term. The filter parameters are then optimized inversely through the adversarial loss function.

[0054] It needs further explanation that, in the specific implementation process, during the physical characteristic compliance detection, a dynamic threshold range is set based on the quantitative relationship between the rise time of the partial discharge pulse and the size of the insulation defect. When the rise time of the separated signal pulse is less than 1 ns and no subsequent oscillation waveform is detected, it is determined that the waveform distortion is caused by high-frequency filter overshoot or low-frequency notch residue. At this time, the parameter recalibration mechanism is triggered, which first checks whether the adjustment step size of the complex domain filter cutoff frequency of the high-frequency channel exceeds the allowable range. If the step size exceeds 80% of the preset maximum value, the cutoff frequency is rolled back to the average value within the first three time windows and the adjustment step size is reduced to 50%. If the pulse rise time is within the threshold range but there is periodic oscillation aftershock, it is determined that the vibration trajectory matching filter of the low-frequency channel has failed. The instantaneous estimate of the vibration fundamental frequency is recalculated and... The offset of the notch filter center frequency is reset to eliminate residual oscillation components. For changes in the dielectric constant of insulating materials caused by sudden temperature and humidity changes, the reference value of the rise time threshold is dynamically adjusted. When the ambient humidity increases by 10% and the temperature fluctuation exceeds 5°C, the upper limit of the threshold is relaxed to 1.2ns to match the changes in material properties. During the recalibration process, if the pulse parameters are still detected to be out of standard after two consecutive calibrations, it is determined that the multi-mode coupling is not completely decoupled, and the original signal of the full frequency band is forcibly switched to re-execute dynamic coupling degree modeling and dual-channel filtering. In addition, the pulse waveform is jointly verified by multiple parameters, including the consistency of rise time, peak amplitude, and half-wave width. If the deviation of any two of the three exceeds 20%, it is determined that the separated signal has irreversible distortion, the current window data is discarded, and the historical best parameters are used to reconstruct the output signal.

[0055] Physical characteristic compliance testing includes:

[0056] The threshold range is set based on the quantitative relationship between the rise time of the partial discharge pulse and the size of the insulation defect. When the rise time of the separated signal pulse is less than 1ns and there is no subsequent oscillation waveform, it is judged as a distorted signal and recalibration is triggered.

[0057] It needs further explanation that, in the specific implementation process, during the physical characteristic compliance detection, a dynamic threshold range is set based on the quantitative relationship between the rise time of the partial discharge pulse and the size of the insulation defect. When the rise time of the separated signal pulse is less than 1 ns and no subsequent oscillation waveform is detected, it is determined that the waveform distortion is caused by high-frequency filter overshoot or low-frequency notch residue. At this time, the parameter recalibration mechanism is triggered, which first checks whether the adjustment step size of the complex domain filter cutoff frequency of the high-frequency channel exceeds the allowable range. If the step size exceeds 80% of the preset maximum value, the cutoff frequency is rolled back to the average value within the first three time windows and the adjustment step size is reduced to 50%. If the pulse rise time is within the threshold range but there is periodic oscillation aftershock, it is determined that the vibration trajectory matching filter of the low-frequency channel has failed. The instantaneous estimate of the vibration fundamental frequency is recalculated and... The offset of the notch filter center frequency is reset to eliminate residual oscillation components. For changes in the dielectric constant of insulating materials caused by sudden temperature and humidity changes, the reference value of the rise time threshold is dynamically adjusted. When the ambient humidity increases by 10% and the temperature fluctuation exceeds 5°C, the upper limit of the threshold is relaxed to 1.2ns to match the changes in material properties. During the recalibration process, if the pulse parameters are still detected to be out of standard after two consecutive calibrations, it is determined that the multi-mode coupling is not completely decoupled, and the original signal of the full frequency band is forcibly switched to re-execute dynamic coupling degree modeling and dual-channel filtering. In addition, the pulse waveform is jointly verified by multiple parameters, including the consistency of rise time, peak amplitude, and half-wave width. If the deviation of any two of the three exceeds 20%, it is determined that the separated signal has irreversible distortion, the current window data is discarded, and the historical best parameters are used to reconstruct the output signal.

[0058] Multi-model classification confidence verification includes:

[0059] The separated signals are input in parallel into a pre-trained time-domain convolutional network and a frequency-domain graph neural network. The confidence variance of the classification results of each model is calculated. When the variance exceeds a preset threshold, it is determined that there is modal aliasing.

[0060] It needs further explanation that, in the specific implementation process, during the multi-model classification confidence verification, the separated signals are input in parallel into the pre-trained time-domain convolutional network and frequency-domain graph neural network to extract the time-domain pulse sequence features and frequency-domain energy distribution features of the signal, respectively. When the corona discharge confidence of the output of the time-domain convolutional network is higher than 90%, while the internal discharge confidence of the frequency-domain graph neural network exceeds 85%, it is determined that there is a conflict in the two models' understanding of the signal modes, and the variance of the two confidence levels is further calculated. If the variance exceeds a preset threshold, it is determined that the current signal has mode aliasing, triggering a re-decoupling instruction and returning to the dynamic coupling degree modeling stage. For the distortion of time-domain pulse features caused by high-frequency transient interference residues, the fluctuation amplitude of the confidence output of the time-domain convolutional network is monitored. If the fluctuation amplitude exceeds 30% within three consecutive time windows, it is determined that the time-domain feature extraction has failed, and the time-domain model is temporarily frozen. The model relies solely on the classification results of the frequency domain model for verification. When the environmental noise spectrum undergoes abrupt changes, the frequency band weight allocation of the frequency domain graph neural network is dynamically adjusted. If the noise peak frequency shifts to within ±10% of the signal's main frequency, the weight coefficient of the corresponding frequency band is reduced by 50% to suppress the impact of noise interference on the classification results. For frequency domain energy tailing caused by incomplete filtering of low-frequency vibration noise, the energy proportion of subharmonic components is detected through the frequency domain graph neural network. If the subharmonic energy exceeds 15% of the fundamental frequency energy, the low-frequency channel is deemed insufficiently filtered, and the notch depth of the vibration trajectory matching filter is adjusted first. In cases where the credibility of the verification results is insufficient, a similarity matching mechanism based on historical successful separation cases is introduced. If the time-frequency joint features of the current signal match the historical successful case library by less than 60%, it is determined to be a new aliasing mode, and the manual auxiliary annotation process is initiated and the model parameter library is updated.

[0061] Nonlinear phase compensation is achieved through the following steps:

[0062] Extract the phase derivative of the instantaneous frequency of the signal, and apply a compensation coefficient that is inversely proportional to the rate of change of the phase of the real part to the imaginary part of the complex domain signal to ensure the time delay consistency of the high-frequency pulse waveform.

[0063] It needs further explanation that, in the specific implementation process, during the nonlinear phase compensation process of phase-sensitive filtering in the complex domain, the phase derivative of the instantaneous frequency of the signal is extracted, the difference in the phase change rate between the real and imaginary parts is calculated in real time, and the compensation coefficient is dynamically allocated according to the direction of the difference. When the phase change rate of the real part is higher than that of the imaginary part, a compensation coefficient inversely proportional to the difference in the change rate is applied to the imaginary part to accelerate the phase change of the imaginary part to match the phase trend of the real part; conversely, the compensation coefficient of the imaginary part is reduced to delay the phase shift. For phase nonlinear distortion caused by abrupt changes in the energy gradient of high-frequency interference, the attenuation rate of the steepness of the pulse leading edge is monitored. If the rate exceeds a preset critical value, it is determined that the phase compensation is insufficient, triggering an adaptive incremental adjustment of the compensation coefficient until the time delay fluctuation of the pulse leading edge converges to the allowable range. For instantaneous frequency jumps caused by load fluctuations, the calculation window length of the phase derivative is synchronously corrected according to the dynamic coupling degree prediction result. Predictions indicate an increased risk of high-frequency noise intrusion, so the window length is shortened to 1 / 4 of the current period to improve phase tracking sensitivity. During signal reconstruction, the phase alignment of the real and imaginary parts of the compensated complex domain signal is checked. If the peak time deviation between the real and imaginary parts of the pulse rising edge exceeds 0.1 microseconds, the compensation coefficients are recalculated and phase compensation is iteratively performed. The deviation direction is recorded as a reference for subsequent compensation coefficient initialization. To address the differences in medium propagation delay caused by temperature and humidity changes, the reference value of the phase derivative is dynamically corrected using environmental sensor data. If the ambient humidity increases by 15% and the temperature drops by more than 10°C, the reference phase derivative is multiplied by a correction factor of 1.2 to offset the impact of changes in medium characteristics. If the phase alignment requirement still cannot be met after multiple iterations, it is determined that the hardware acquisition channel is abnormal. The system switches to the backup signal channel and resets all compensation parameters to ensure the continuous stability of the filtering process.

[0064] The adjustment rules for notch bandwidth are as follows:

[0065] When the slope of the vibration signal envelope exceeds a set threshold, the notch bandwidth is extended to 1.5 times the fundamental frequency; when the slope of the envelope is below the threshold, the notch bandwidth is compressed to 0.8 times the fundamental frequency.

[0066] It should be further explained that, in the specific implementation process, during the adjustment of the notch bandwidth for vibration trajectory matching filtering, the instantaneous rate of change of the slope of the vibration signal envelope is monitored in real time. When the slope exceeds the set threshold, it is determined that the vibration noise energy is positively accumulating in the low-frequency harmonic region. At this time, the notch bandwidth is expanded to 1.5 times the fundamental frequency to cover the harmonic diffusion range, and the notch depth is increased to 120% of the original value to enhance low-frequency suppression. If the slope is below the threshold and the energy fluctuation of the main peak of the vibration spectrum is less than 5%, it is determined that the vibration noise is in a stable state. The notch bandwidth is compressed to 0.8 times the fundamental frequency and the notch depth is reduced to 80% to avoid over-filtering leading to attenuation of the discharge signal fundamental frequency. For the instantaneous shift of the vibration fundamental frequency caused by sudden changes in equipment load, the fundamental frequency change trend of the next time window is predicted by Kalman filtering. If the predicted shift exceeds 2% of the current fundamental frequency, the tracking step size of the notch center frequency is adjusted in advance, and... During the bandwidth expansion phase, offset compensation is superimposed to prevent notch mismatch. When a sudden increase in ambient temperature is detected, causing an abnormally steep increase in the slope of the vibration signal envelope, an anti-temperature drift mode is temporarily activated, dynamically raising the slope judgment threshold to 1.3 times the standard value and limiting the maximum notch bandwidth expansion to 1.2 times the fundamental frequency to prevent temperature drift interference from falsely triggering the filter. For intermittent broadband vibration noise caused by mechanical loosening, the spectral flatness of the residual component after notch filtering is analyzed. If the flatness index exceeds 0.7 and lasts for three time windows, it is determined to be broadband noise leakage. The mode is switched to multi-stage series notch filtering and the bandwidth overlap area is increased step by step. After the notch filter parameters are updated, the integrity of the periodic discharge pulse of the separated signal is reverse-verified. If the standard deviation of the pulse interval increases by more than 15% compared to before filtering, it is determined that the notch filtering is excessive, causing waveform distortion. The mode is then rolled back to the previous parameter set and the bandwidth adjustment amplitude is reduced to 50% of the original value.

[0067] The optimization process of the adversarial loss function includes:

[0068] In each iteration, the distribution parameters of the mode discriminator are updated synchronously, so that the standard deviation of the pulse interval of the corona discharge converges to the theoretical variance of the Poisson distribution, and the periodic error of the internal discharge converges to the standard deviation range of the Gaussian distribution.

[0069] It needs further explanation that, in the specific implementation process, during the optimization of the adversarial loss function, the weight coefficient of KL divergence in the adversarial loss function is dynamically adjusted by monitoring the deviation of the standard deviation of the pulse interval of corona discharge from the theoretical variance of the Poisson distribution in real time. When the standard deviation of the pulse interval exceeds 15% of the theoretical variance, it is determined that the random characteristics of the high-frequency channel are not retained sufficiently, and the weight of KL divergence is increased to 1.2 times the initial value to strengthen the distribution matching constraint, while reducing the optimization weight of the periodic error term. For the periodic peak-to-peak error of the internal discharge, if it exceeds three times the standard deviation of the Gaussian distribution, it is determined that the group delay parameter of the notch filter in the low-frequency channel is mismatched. The gradient of the periodic error term is backpropagated to the low-frequency channel through adversarial training, and the tracking step size of the notch center frequency is dynamically adjusted. When the statistical characteristics of corona discharge and internal discharge deviate from the target distribution simultaneously, a joint evaluation mechanism of distribution coupling degree is introduced. If the product of the two types of deviation exceeds a preset threshold, then... If the physical constraints of the modal discriminator are deemed invalid, adversarial training is paused and the historically optimal distribution parameters are reloaded. To address the abnormally increased randomness of pulse intervals caused by load fluctuations, the moving average of the coefficient of variation is calculated using a sliding window. If the average value increases by more than 20% within three consecutive time windows, a robust optimization mode is triggered, relaxing the theoretical variance tolerance range of the Poisson distribution to 1.3 times and increasing the correction frequency for periodic errors. In each iteration, the threshold values ​​of the discriminator's distribution parameters are updated synchronously, gradually converging the standard deviation of the corona discharge pulse interval to the theoretical variance range of the Poisson distribution, while simultaneously forcing the peak value of the periodic error of internal discharge to be compressed to within twice the standard deviation of the Gaussian distribution. For cases where parameters cannot converge under extreme conditions, if the standard deviation still exceeds the target range after five iterations, it is determined to be due to environmental interference or hardware anomaly. The current model parameters are frozen, and the system switches to a static filtering mode based on the historical optimal solution, while simultaneously triggering an alarm signal to await manual intervention and calibration.

[0070] When implementing adaptive filtering and intelligent separation of multimodal partial discharge signals in a dynamic noise environment, the mixed signals are first collected in real time by sensors, and the equipment operating parameters and environmental noise signals are acquired simultaneously as auxiliary inputs. Based on the time-frequency energy distribution characteristics of the signal and noise, the time-frequency mutual information entropy gradient sequence of the current and historical moments is extracted by using a sliding window. The temporal dependency is modeled by a bidirectional gated cyclic unit to predict the coupling path offset direction of noise and signal in multiple future time windows.

[0071] Based on the prediction results, the signal is divided into two independent channels: high-frequency transient interference and low-frequency periodic noise. The high-frequency channel adopts complex domain phase-sensitive filtering technology to map the time domain signal to the complex domain and implement nonlinear phase compensation. By dynamically adjusting the filter cutoff frequency, high-frequency interference is suppressed from intruding into the main frequency band, while retaining the steep pulse leading edge characteristics.

[0072] The low-frequency channel constructs an adaptive notch filter bank based on prior knowledge of the equipment's mechanical vibration spectrum. By matching the vibration noise trajectory in real time and correcting the residual component amplitude, it eliminates the time-domain overlap between low-frequency noise and periodic discharge signals. The output signals of the two channels are input into an adversarial mode separation network. By pre-setting mode discriminators with different physical statistical characteristics of discharge types, the separated signals are forced to satisfy the randomness of the pulse interval of corona discharge and the periodic regularity of internal discharge. The adversarial loss function is used to inversely optimize the filter parameters. After separation, the signal needs to undergo a dual verification mechanism. The physical feature compliance detection module sets a dynamic threshold based on the quantitative relationship between partial discharge pulse parameters and insulation defects. When waveform distortion is detected, parameter recalibration is triggered. The multi-model classification confidence verification module inputs the signal into the time-domain and frequency-domain classification models in parallel. The confidence variance analysis is used to determine the risk of mode mixing and returns to iterative decoupling.

[0073] For operating conditions such as load fluctuations and sudden changes in temperature and humidity, the correlation between equipment parameters and noise signals is monitored in real time, and the coupling degree prediction model and filter parameter adjustment step size are dynamically corrected. When a sudden mechanical shock or unknown modal interference is detected, the system switches to anti-interference mode and initiates a manual intervention protocol to ensure system robustness. All modules form a closed-loop feedback, and through the synergistic effect of dynamic coupling sensing, dual-channel decoupling filtering, adversarial modal separation, and dual verification, high-fidelity separation of multimodal signals under dynamic noise environment is achieved.

[0074] By employing a collaborative mechanism of dynamic coupling degree modeling and dual-channel adversarial decoupling, this study effectively addresses the modal aliasing and feature distortion issues caused by time-frequency coupling of signals and noise in dynamic noise environments, a problem inherent in traditional methods. The dynamic coupling path sensing module quantifies the interaction characteristics of noise and signals in real time. Combined with a dual-channel architecture of complex-domain phase-sensitive filtering and vibration trajectory matching filtering, it achieves precise suppression of high-frequency transient interference and low-frequency periodic noise. The adversarial modal separation network, through physical statistical constraints and adversarial training mechanisms, forces the separated signals to satisfy the independent distribution characteristics of discharge types, significantly improving the separation accuracy and anti-interference capability of multimodal signals.

[0075] By introducing a dual guarantee mechanism of physical feature compliance detection and multi-model confidence verification, this method solves the problem of insufficient diagnostic reliability caused by the accumulation of cascaded errors in traditional algorithms, ensuring that the separation results simultaneously comply with the dual constraints of physical laws and data models. Under the complex operating conditions of power equipment, this method can adaptively track the dynamic coupling changes of noise and signals, reduce the false positive and false negative rates, provide highly reliable data support for early warning and accurate location of insulation defects, reduce the need for manual intervention, and improve the practicality and engineering deployment efficiency of intelligent diagnostic systems.

[0076] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive filtering and intelligent separation method for multimodal partial discharge signals, characterized in that, Includes the following steps: Step S1: Real-time acquisition of partial discharge signals from power equipment, and simultaneous acquisition of equipment operating parameters and environmental noise signals; Step S2: Calculate the dynamic coupling index based on the time-frequency matrix of signal and noise, extract the time-frequency energy distribution through a sliding window, and generate a quantized value of dynamic coupling strength based on the inverse correlation between the high-frequency interference energy gradient and the signal main frequency gradient. Step S3: Based on the time-series prediction results of the dynamic coupling index, the signal is divided into a high-frequency transient interference channel and a low-frequency periodic noise channel, and noise suppression is performed by complex domain phase-sensitive filtering and vibration trajectory matching filtering, respectively. Among them, the complex domain phase-sensitive filtering maps the signal to the complex domain through the transient pulse compression encoder and dynamically adjusts the filter cutoff frequency, while the vibration trajectory matching filtering constructs an adaptive notch filter bank based on the prior knowledge of the mechanical vibration spectrum of the equipment. Step S4: Input the output signals of the high-frequency channel and the low-frequency channel into the adversarial mode separation network, and force the separated signals to meet the preset independent mode statistical distribution through the mode discriminator; Step S5: Cross-validate the separation results based on physical feature compliance detection and multi-model classification confidence verification. If the impulse parameter exceeds the threshold or the classification confidence variance exceeds the standard, trigger parameter recalibration and re-iterate decoupling.

2. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 1, characterized in that: The time-series prediction of the dynamic coupling degree index is modeled by a bidirectional gated cyclic unit to model the time-frequency mutual information entropy gradient sequence, predicting the direction of the coupling path offset between noise and signal within the next three time windows.

3. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 1, characterized in that: The implementation of the complex domain phase-sensitive filtering includes: Nonlinear phase compensation is performed on the real and imaginary parts of the signal, and the cutoff frequency of the complex domain filter is offset by combining the dynamic coupling degree prediction value, so as to preserve the steep pulse leading edge characteristics and suppress the cross-coupling components of the high-frequency interference band.

4. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 1, characterized in that: The implementation of the vibration trajectory matching filter includes: The notch center frequency is dynamically generated based on the fundamental frequency of the equipment vibration and its harmonic components. The notch bandwidth is adaptively adjusted by the slope of the vibration signal envelope, and the time-domain amplitude of the residual components is corrected by Kalman filtering.

5. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 1, characterized in that: The adversarial mode separation network is implemented in the following way: Modal discriminators are set at the output terminals of the high and low frequency channels respectively. The discriminators are preset with physical constraints that the pulse interval of the corona discharge follows a Poisson distribution and the periodicity of the internal discharge follows a Gaussian distribution. The difference between the separated signal and the target mode distribution is calculated based on KL divergence, and the pulse interval standard deviation is introduced as an additional constraint term. The filter parameters are then optimized inversely through the adversarial loss function.

6. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 1, characterized in that: The physical characteristic compliance detection includes: setting a threshold range based on the quantitative relationship between the rise time of the partial discharge pulse and the size of the insulation defect; when the rise time of the separated signal pulse is less than 1ns and there is no subsequent oscillation waveform, it is determined to be a distorted signal and recalibration is triggered.

7. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 1, characterized in that: The multi-model classification confidence verification includes: inputting the separated signals into a pre-trained time-domain convolutional network and a frequency-domain graph neural network in parallel, calculating the confidence variance of each model classification result, and determining that mode aliasing exists when the variance exceeds a preset threshold.

8. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 3, characterized in that: The nonlinear phase compensation is achieved through the following process: extracting the phase derivative of the instantaneous frequency of the signal, applying a compensation coefficient inversely proportional to the rate of change of the real phase to the imaginary part of the complex domain signal, and ensuring the time delay consistency of the high-frequency pulse waveform.

9. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 4, characterized in that: The adjustment rule for the notch bandwidth is as follows: when the slope of the vibration signal envelope exceeds a set threshold, the notch bandwidth is extended to cover the range of vibration noise harmonic diffusion; when the slope of the envelope is lower than the threshold, the notch bandwidth is compressed to the matching range of the focusing main vibration frequency.

10. The adaptive filtering and intelligent separation method for multimodal partial discharge signals according to claim 5, characterized in that: The optimization process of the adversarial loss function includes: In each iteration, the distribution parameters of the mode discriminator are updated synchronously, so that the standard deviation of the pulse interval of the corona discharge converges to the theoretical variance of the Poisson distribution, and the periodic error of the internal discharge converges to the standard deviation range of the Gaussian distribution.

Images