A risk identification method based on short wave gate excitation and defect discharge response fingerprint

Abstract

The application discloses a risk identification method based on short-wave gate-controlled excitation and defect discharge response fingerprint, relates to the technical field of power equipment insulation state detection and discharge signal intelligent diagnosis, and comprises the following steps: acquiring information of an insulation part to be detected and initial detection configuration, generating a gate-controlled excitation time sequence table and a collection condition set, synchronously collecting multi-source discharge signals and synchronous parameters, outputting a synchronous multi-source data set and preprocessing, and extracting candidate pulses; screening a reliable discharge event set based on the candidate pulses and cross-channel correlation conditions, performing multi-dimensional feature extraction, constructing a discharge response fingerprint vector based on a multi-domain feature set, and calculating the reliability weight of each dimension feature; obtaining a defect risk index and determining a risk grade based on the discharge response fingerprint vector and the reliability weight, and determining a defect type based on the discharge response fingerprint vector and the reliability weight. The application improves the detection sensitivity and diagnosis reliability of early micro insulation defects in high-voltage power equipment.

Description

Technical Field

[0001] This application relates to the field of power equipment insulation condition detection and intelligent discharge signal diagnosis technology, and in particular to a risk identification method based on shortwave gating excitation and defect discharge response fingerprinting, which is applicable to the early identification of defects such as air gaps, cracks, interface peeling, metal particle adhesion and surface contamination in gas-insulated metal-enclosed switchgear (GIS), gas-insulated metal-enclosed transmission lines, gas-insulated cable terminals, epoxy insulation components, pot insulators and other high-voltage insulation components. Background Technology

[0002] Partial discharge detection is an important technical means to determine the insulation status of high-voltage power equipment. Traditional detection methods mostly rely on empirical judgment based on apparent discharge quantity, discharge frequency, or phase-resolved partial discharge (PRPD) spectrum density distribution. While these methods can identify typical discharge phenomena, they still have significant shortcomings when dealing with early, minute defects. On the one hand, under conventional power frequency voltage excitation, the probability of initial electron generation for minute defects is low, and the discharge statistical time lag is long, resulting in only a small number of pulses appearing within a limited acquisition window. On the other hand, the complex field environment, electromagnetic noise, switching operation interference, and differences in sensor bandwidth can easily distort the pulse waveform, making it difficult for a single discharge quantity or a single spectrum feature to stably and reliably reflect the true state of the defect.

[0003] In recent years, shortwave radiation (such as X-rays or extreme ultraviolet radiation) has been introduced into the field of partial discharge detection. It can provide exogenous initial electrons to the defect region, thereby exciting latent defects to produce a more stable and repeatable discharge response under lower or conventional electric field conditions. Existing research mainly focuses on the effect of shortwave excitation on increasing the discharge probability, or on the design of specific experimental devices and defect simulation structures. However, systematic technical solutions are still lacking for how to effectively fuse discharge data before and after excitation, how to suppress noise and redundancy from multi-source signals, and how to construct quantitative fingerprint features with discriminative capabilities.

[0004] Especially under complex field conditions, the discharge response of the same defect may exhibit various changes such as amplitude fluctuations, phase drift, spectral energy migration, and discharge sequence aggregation under different voltage levels, different shortwave doses, and different excitation cycles. If only a single indicator is used for analysis, it may misjudge active noise as defect deterioration, or underestimate dangerous defects with weak amplitude but high repeatability as low-risk conditions.

[0005] Therefore, it is necessary to provide a risk identification method based on short-wavelength gating excitation and defect discharge response fingerprints to solve the above problems. Summary of the Invention

[0006] The purpose of this application is to provide a risk identification method based on shortwave gating excitation and defect discharge response fingerprinting, which improves the detection sensitivity and diagnostic reliability of early micro-insulation defects in high-voltage power equipment and reduces the impact of noise interference on the identification results in complex field environments.

[0007] To achieve the above objectives, this application provides the following solution: This application provides a risk identification method based on short-wavelength gated excitation and defect discharge response fingerprinting, the method comprising: Acquire information about the insulation component under test and the initial test configuration, and generate a gating excitation timing table and a set of acquisition conditions; each acquisition cycle of the gating excitation timing table includes a shortwave excitation segment; Based on the gating excitation timing table and the acquisition condition set, multi-source discharge signals and synchronization parameters are acquired synchronously. After time alignment and the addition of acquisition stage labels, a synchronous multi-source dataset is output. The synchronous multi-source dataset is preprocessed, and candidate pulses for each channel are extracted; Discharge events are filtered based on candidate pulses and cross-channel correlation conditions, and a set of reliable discharge events is output. Multi-dimensional feature extraction is performed on the set of credible discharge events to obtain a multi-domain feature set; A discharge response fingerprint vector is constructed based on a multi-domain feature set, and the reliability weights of each dimension of the features in the discharge response fingerprint vector are calculated. The defect risk index is obtained based on the discharge response fingerprint vector and reliability weights; The risk level is determined based on the defect risk index, and the defect type is determined based on the discharge response fingerprint vector and reliability weight.

[0008] In one embodiment, the information of the insulation component to be tested includes the device type, the area under test, and the area of ​​concern for defects. The initial detection configuration includes the applied voltage method, sensor channel, sampling rate, shortwave source type, allowable dose range, and number of acquisition cycles. Acquire information about the insulating component under test and the initial test configuration, and generate a gating excitation timing table and a set of acquisition conditions, specifically including: Acquire information about the insulation component under test and the initial detection configuration. Based on the information about the insulation component under test and the initial detection configuration, set at least one acquisition cycle and divide each acquisition cycle into a dark field baseline segment, a shortwave excitation segment and a recovery observation segment in sequence, and set the start and end times of each stage respectively. Set the applied voltage, shortwave dose, gating pulse width, repetition interval, and number of repetitions for each acquisition cycle; The start and end times of each stage are arranged with the applied voltage, shortwave dose, gating pulse width, repetition interval and repetition number of each acquisition cycle according to a unified clock to form a gating excitation timing table. The applied voltage method, sensor channel, sampling rate, shortwave source type, allowable dose range and number of acquisition cycles are set as acquisition conditions and output.

[0009] In one embodiment, the multi-source discharge signal includes the original waveforms of each partial discharge sensor channel, and the synchronization parameters include: applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp, and environmental parameters. Based on the gating excitation timing table and acquisition condition set, multi-source discharge signals and synchronization parameters are acquired synchronously. After time alignment and the addition of acquisition stage labels, a synchronous multi-source dataset is output, which specifically includes: Based on the gating excitation timing table and the acquisition condition set, the applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp, environmental parameters, and the original waveforms of each partial discharge sensor channel are synchronously acquired. The data collected by each partial discharge sensor channel is time-aligned using a unified sampling clock, hardware trigger signal or timestamp, and the arrival time deviation caused by each partial discharge sensor and transmission link is corrected by using a pre-calibrated fixed channel delay. Based on the start and end times of each stage in the gating excitation timing table, each sampling point is assigned a sampling cycle number and a sampling stage label. The applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp, and environmental parameters are associated with the same time base, and a synchronous multi-source dataset is output.

[0010] In one embodiment, the synchronous multi-source dataset is preprocessed, and candidate pulses for each channel are extracted, specifically including: Bandpass filtering and baseline drift correction matching the effective frequency band of the corresponding sensor are performed on each partial discharge sensor channel in the synchronous multi-source dataset to obtain the preprocessed waveform of each channel; Based on the data from the dark field baseline segment of the synchronous multi-source dataset, the noise mean, noise standard deviation, and noise energy of each partial discharge sensor channel are estimated. Based on the noise standard deviation and the first coefficient of each channel, the noise peak threshold corresponding to each channel is determined; Based on the noise energy and second coefficient of each channel, the noise energy threshold corresponding to each channel is determined. When the peak value of the preprocessed waveform of any channel is greater than or equal to the peak value threshold of the corresponding channel, or when the energy within the preset pulse window of the preprocessed waveform of any channel is greater than or equal to the noise energy threshold of the corresponding channel, a local waveform segment of preset length located on both sides of the trigger time on the preprocessed waveform of the corresponding channel is extracted as the candidate pulse of the corresponding channel.

[0011] In one embodiment, the cross-channel association conditions include cross-channel time consistency conditions, tag consistency during the acquisition phase, and waveform and phase similarity conditions; Discharge events are filtered based on candidate pulses and cross-channel correlation conditions, and a set of reliable discharge events is output, specifically including: The morphological characteristics and power frequency phase of each candidate pulse are calculated. Based on the morphological characteristics and power frequency phase, each candidate pulse is initially screened according to preset criteria to remove random spikes and non-discharge noise that do not meet any preset criteria, thus obtaining the initially screened candidate pulses. The morphological characteristics include peak value, pulse energy, rise time, fall time, half peak width, and waveform correlation coefficient. The preset criteria include one or more of the following: preset amplitude lower limit, preset energy lower limit, preset pulse width range, and preset rise edge range. After removing duplicate reflected pulses and switching interference from the initial candidate pulses, valid candidate pulses are obtained. Each valid candidate pulse that simultaneously meets the cross-channel time consistency condition, the acquisition stage tag consistency condition, and the waveform and phase similarity condition is associated as the same discharge event. All discharge events that meet the cross-channel association condition and are successfully associated are judged as high-confidence discharge events, and the event quality coefficient of each high-confidence discharge event is determined. Valid candidate pulses that do not meet the cross-channel association condition, meet the preset repeatability condition, and are triggered by a single channel are judged as low-confidence discharge events, and the event quality coefficient of each low-confidence discharge event is determined. The event quality coefficient of a low-confidence discharge event is less than that of a high-confidence discharge event. The preset repeatability condition is that the event pulses recur in multiple acquisition cycles at similar power frequency phases or after shortwave gating is enabled. Based on all high-confidence and low-confidence discharge events, a set of credible discharge events is constructed; each discharge event carries a corresponding event quality coefficient.

[0012] In one embodiment, multi-dimensional feature extraction is performed on the set of credible discharge events to obtain a multi-domain feature set, specifically including: For each discharge event in the set of credible discharge events, time-domain morphological features, frequency-domain energy features, phase statistical features, excitation response features, and sequence dynamic features are extracted to form a multi-domain feature set; Among them, the temporal morphological characteristics include one or more of the following: pulse peak value, apparent discharge equivalent value, rise time, fall time, half width at half maximum (WWHM), pulse area, kurtosis, and skewness. Frequency domain energy characteristics include one or more of the following: dominant frequency, frequency band energy proportion, spectral centroid, spectral entropy, short-time energy peak, and wavelet packet energy distribution. Phase statistical characteristics include one or more of the following: discharge initiation phase, phase concentration, positive and negative half-cycle asymmetry, PRPD spectrum density distribution, and phase drift. Excitation response characteristics include one or more of the following: discharge number gain from dark field to excitation segment, amplitude gain, time delay reduction rate, dose response slope, response saturation coefficient, and recovery decay time constant; Sequence dynamics characteristics include one or more of the following: adjacent discharge time interval, clustering degree, burst duration, repetition rate stability, inter-pulse energy correlation, and discharge memory factor.

[0013] In one embodiment, a discharge response fingerprint vector is constructed based on a multi-domain feature set, and the reliability weights of each dimension of the features in the discharge response fingerprint vector are calculated, specifically including: Physical quantity correction and statistical normalization are performed on the features of each dimension in the multi-domain feature set to obtain the normalized features of each dimension. Based on the normalized features of each dimension, the discharge response fingerprint vector is constructed. Based on the event quality coefficient of each discharge event, the normalized signal-to-noise ratio, cross-channel consistency coefficient and repeatability coefficient of each dimension of the discharge response fingerprint vector are calculated respectively. Based on synchronous multi-source data, the acquisition window integrity coefficient of each dimension of features in the discharge response fingerprint vector is calculated; Based on the pre-calibrated sensor channel calibration parameters, the calibration quality coefficients of each dimension of the discharge response fingerprint vector are calculated. The weighted geometric mean method is used to determine the reliability weights of each dimension of the discharge response fingerprint vector based on the normalized signal-to-noise ratio, cross-channel consistency coefficient, repeatability coefficient, acquisition window integrity coefficient, and calibration quality coefficient.

[0014] In one embodiment, a defect risk index is obtained based on the discharge response fingerprint vector and reliability weights, specifically including: Based on the information of the insulation component under test, the set of acquisition conditions and environmental parameters, a matching reference subset is selected from the preset risk model; the matching reference subset includes the safe endpoints and risk endpoints of each feature; Each dimension of the discharge response fingerprint vector is compared with the safe endpoint and risk endpoint in the matching reference subset, and each dimension of the feature is mapped to a single risk score in the range of 0 to 1. The individual risk score of each feature in the discharge response fingerprint vector is multiplied by the corresponding reliability weight and the preset feature fusion weight to obtain the unnormalized contribution value of each feature. The shortwave excitation comprehensive term is obtained by weighting and averaging the individual risk scores corresponding to the excitation response sub-vectors in the discharge response fingerprint vector. The memory term for the recovery segment is determined based on the ratio of the event rates of the recovery observation segment to the shortwave excitation segment and the recovery time constant. The defect risk index is obtained by summing and normalizing the unnormalized contribution values ​​of each dimension's features, the shortwave excitation comprehensive term, and the recovery segment memory term.

[0015] In one embodiment, the risk level is determined based on a defect risk index, and the defect type is determined based on a discharge response fingerprint vector and a reliability weight, specifically including: When the defect risk index is greater than or equal to 0 and less than the first preset risk threshold, the risk level of the insulation component to be tested is determined to be low risk level. When the defect risk index is greater than or equal to the first preset risk threshold and less than the second preset risk threshold, the risk level of the insulation component to be tested is determined to be the risk level of concern. When the defect risk index is greater than or equal to the second preset risk threshold and less than the third preset risk threshold, the risk level of the insulation component to be tested is determined to be the warning risk level. When the defect risk index is greater than or equal to the third preset risk threshold and less than 1, the risk level of the insulation component to be tested is determined to be a severe risk level. Based on the reference defect category model, obtain the reference center vector corresponding to each preset defect category; The reliability-weighted distance between the discharge response fingerprint vector and the reference center vector of each preset defect category is calculated one by one, and the defect category with the smallest reliability-weighted distance is taken as the final defect type. The reliability-weighted distance is obtained by weighting the distance based on the reliability weight of each dimension feature.

[0016] In one embodiment, the method further includes: Based on the defect risk index, individual risk score vector, reliability weight, and preset feature fusion weight, the normalized contribution value of each dimension feature is determined, thereby obtaining the feature contribution vector. The normalized contribution values ​​of each dimension feature are then summarized according to the feature domain to obtain the contribution degree of each feature domain. The individual risk score vector is obtained based on the individual risk scores of all dimension features. Based on the defect risk index, individual risk score vector, feature contribution vector, contribution of each feature domain, and reference defect category model, preliminary diagnostic results are determined. The preliminary diagnostic results include candidate defect type, risk level, defect risk index, category similarity, key contribution features, and contribution of each feature domain. Based on the preliminary diagnostic results, reliability weight vector, acquisition window integrity coefficient, repetition period judgment consistency rate, and defect category similarity, the diagnostic confidence is determined and output.

[0017] According to the specific embodiments provided in this application, this application has the following technical effects: This application discloses a risk identification method based on shortwave gated excitation and defect discharge response fingerprinting. By setting a gated excitation timing table that includes a shortwave excitation segment, external initial electrons are provided to the defect region, enabling early-stage small defects to generate stable and repeatable discharge responses under a lower electric field. This solves the problem of sparse discharge pulses and difficulty in capturing them in conventional detection. Discharge events are screened through cross-channel correlation conditions, requiring that the same discharge event be detected simultaneously by at least two independent channels to be identified. This effectively suppresses single-channel sporadic noise and external interference, solving the problem of misjudgment based on a single indicator in complex field environments. Furthermore, a multi-dimensional analysis of high-confidence discharge events is conducted. Feature extraction comprehensively characterizes discharge characteristics from multiple dimensions, including time-domain morphology, frequency-domain energy, phase statistics, excitation response, and sequence dynamics, avoiding the limitations of traditional methods that rely on a single discharge quantity or a single PRPD map for judgment. By constructing a unified discharge response fingerprint vector, multi-domain features are fused into a unified representation that is calculable, comparable, and traceable, solving the problem of the lack of a systematic fusion scheme for discharge data before and after excitation in existing methods. By calculating the reliability weights of features in each dimension, the contribution of features with low signal-to-noise ratio, poor cross-channel consistency, and weak repeatability to the fusion results is reduced, further improving the stability and anti-interference ability of the diagnostic results. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A schematic flowchart of a risk identification method based on shortwave gating excitation and defect discharge response fingerprinting provided in an embodiment of this application; Figure 2 This is a schematic diagram of a gating excitation acquisition sequence provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] In one exemplary embodiment, such as Figure 1 As shown, a risk identification method based on short-wavelength gated excitation and defect discharge response fingerprinting is provided, including the following steps: Wherein: Step S1: Obtain information on the insulation component under test and the initial test configuration, and generate a gating excitation timing table and a set of acquisition conditions. Each acquisition cycle of the gating excitation timing table includes a shortwave excitation segment.

[0023] As an optional implementation, the information of the insulation component under test includes the device type, the area under test and the area of ​​concern for defects, and the initial detection configuration includes the external voltage method, sensor channel, sampling rate, shortwave source type, allowable dose range and number of acquisition cycles; Step S1 specifically includes: Step S11: Obtain the information of the insulation component under test and the initial detection configuration. Based on the information of the insulation component under test and the initial detection configuration, set at least one acquisition cycle and divide each acquisition cycle into a dark field baseline segment, a shortwave excitation segment and a recovery observation segment in sequence, and set the start and end times of each stage respectively. Step S12: Set the applied voltage, shortwave dose, gating pulse width, repetition interval, and number of repetitions for each acquisition cycle; Step S13: Arrange the start and end times of each stage with the applied voltage, shortwave dose, gating pulse width, repetition interval and repetition number of each acquisition cycle according to a unified clock to form a gating excitation timing table, and output the applied voltage method, sensor channel, sampling rate, shortwave source type, allowable dose range and number of acquisition cycles as the acquisition condition set.

[0024] Specifically, based on the information of the insulating component under test and initial detection configuration Set at least one acquisition cycle, and divide each acquisition cycle sequentially into the following: Figure 2 The dark field baseline segment shown Shortwave excitation segment and recovery observation segment Three acquisition stages; start and end times for each stage are set separately. , , , , and The dark-field baseline segment is used to acquire the background noise and spontaneous discharge response without shortwave application; the shortwave excitation segment is used to acquire the discharge response under the influence of exogenous initial electrons; and the recovery observation segment is used to acquire the discharge attenuation or memory response after shortwave removal. The shortwave excitation segment applies shortwave-gated pulses using fixed dose, incremental dose, pulse width modulation, or intermittent repetition methods, and records the on-time, off-time, and dose parameters of the shortwave-gated pulses.

[0025] According to the purpose of the test, the first Set the external voltage for each acquisition cycle. shortwave dose or flux Gating pulse width Repetition interval and number of repetitions The applied voltage can remain stable or vary according to a preset step.

[0026] Arrange the above parameters according to a unified clock to form a gating excitation timing table. ,in At a minimum, it should include the acquisition cycle number, B / I / R segment labels, start and end times of each segment, applied voltage, shortwave on and off times, dose parameters, and repetition cycles. Simultaneously, it should output a gating excitation timing table. and collection condition set .

[0027] Step S2: Based on the gating excitation timing table and the acquisition condition set, synchronously acquire multi-source discharge signals and synchronization parameters, and output synchronous multi-source dataset after time alignment and the addition of acquisition stage labels.

[0028] As an optional implementation, the multi-source discharge signal includes the original waveforms of each partial discharge sensor channel, and the synchronization parameters include: applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp, and environmental parameters; Step S2 specifically includes: Step S21: Based on the gating excitation timing table and the acquisition condition set, synchronously acquire the applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp, environmental parameters, and the original waveforms of each partial discharge sensor channel; Step S22: Use a unified sampling clock, hardware trigger signal or timestamp to time-align the data collected by each partial discharge sensor channel, and use a pre-calibrated fixed channel delay to correct the arrival time deviation caused by each partial discharge sensor and the transmission link. Step S23: Based on the start and end times of each stage in the gating excitation timing table, add a collection cycle number and collection stage label to each sampling point, and associate the applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp and environmental parameters with the same time base, and output a synchronous multi-source dataset.

[0029] Specifically, a short-wavelength gated excitation is applied to the insulating component under test to introduce exogenous initial electrons into the defect region, thereby generating a stable and repeatable partial discharge response under the action of an external electric field. Simultaneously, partial discharge sensors are attached to the insulating component under test to collect the partial discharge signals generated by the partial discharge response. The partial discharge signals can be obtained from one or more of the following: high-frequency current, ultra-high frequency, coupling capacitor, or electroacoustic sensors.

[0030] Simultaneously acquire the raw waveforms of each partial discharge sensor. External voltage Power frequency phase Shortwave gated signal shortwave dose or flux Sampling timestamps and environmental parameters Then, a unified sampling clock, hardware trigger signal, or timestamp is used to align each channel, and a pre-calibrated channel delay is used to fix the delay. Corrects arrival time deviations caused by sensors and transmission links.

[0031] Output synchronized multi-source datasets ,in Tags for the data collection phase. To standardize sampling timestamps.

[0032] Step S3: Preprocess the synchronous multi-source dataset and extract candidate pulses for each channel.

[0033] As an optional implementation, step S3 specifically includes: Step S31: Perform bandpass filtering and baseline drift correction matching the effective frequency band of the corresponding sensor for each partial discharge sensor channel in the synchronous multi-source dataset to obtain the preprocessed waveform of each channel; Step S3,2: Based on the data of the dark field baseline segment of the synchronous multi-source dataset, estimate the noise mean, noise standard deviation and noise energy of each partial discharge sensor channel; Step S33: Determine the noise peak threshold for each channel based on the noise standard deviation and the first coefficient. Step S34: Determine the noise energy threshold for each channel based on the noise energy and the second coefficient of each channel; Step S35: When the peak value of the preprocessed waveform of any channel is greater than or equal to the peak value threshold of the corresponding channel, or when the energy within the preset pulse window of the preprocessed waveform of any channel is greater than or equal to the noise energy threshold of the corresponding channel, a local waveform segment of a preset length located on both sides of the trigger time on the preprocessed waveform of the corresponding channel is extracted as a candidate pulse of the corresponding channel.

[0034] Specifically, for the first Each partial discharge sensor channel performs bandpass filtering matching the effective frequency band of the sensor, uses the sliding median or a low-order trend term for baseline drift correction, and estimates the noise mean based on the dark field baseline segment. Noise Standard Deviation and noise energy The original waveform is obtained. Preprocessed waveform .

[0035] When preprocessing waveforms The peak value satisfies Or the energy within a preset pulse window satisfy At the trigger moment, local waveform segments of preset length are extracted on both sides as candidate pulses. The candidate pulse is a local waveform segment that has been preprocessed and triggered by a threshold, not the entire preprocessed waveform; and Determined by dark field noise statistics or calibration tests.

[0036] Step S4: Based on candidate pulses and cross-channel correlation conditions, filter discharge events and output a set of reliable discharge events.

[0037] As an optional implementation, cross-channel association conditions include cross-channel time consistency conditions, tag consistency during the acquisition phase, and waveform and phase similarity conditions; Step S4 specifically includes: Step S41: Calculate the morphological characteristics and power frequency phase of each candidate pulse, and based on the morphological characteristics and power frequency phase, perform preliminary screening on each candidate pulse according to preset criteria to remove random spikes and non-discharge noise that do not meet any preset criteria, and obtain the preliminary screened candidate pulses; the morphological characteristics include peak value, pulse energy, rise time, fall time, half peak width and waveform correlation coefficient, and the preset criteria include one or more of the preset amplitude lower limit, preset energy lower limit, preset pulse width range and preset rise edge range; Step S42: Remove duplicate reflected pulses and switching interference from the candidate pulses after initial screening to obtain valid candidate pulses; Step S43: Associate each valid candidate pulse that simultaneously meets the cross-channel time consistency condition, the acquisition stage tag consistency condition, and the waveform and phase similarity condition as the same discharge event, and determine all discharge events that meet the cross-channel association condition and are successfully associated as high-confidence discharge events, and determine the event quality coefficient of each high-confidence discharge event. Step S44: Valid candidate pulses that do not meet the cross-channel association condition, meet the preset repeatability condition, and are triggered by a single channel are judged as low-confidence discharge events, and the event quality coefficient of each low-confidence discharge event is determined; the event quality coefficient of the low-confidence discharge event is less than the event quality coefficient of the high-confidence discharge event, and the preset repeatability condition is that it recurs in multiple acquisition cycles at similar power frequency phases or after shortwave gating is turned on. Step S45: Based on all high-confidence discharge events and low-confidence discharge events, a set of credible discharge events is constructed; wherein, each discharge event carries a corresponding event quality coefficient.

[0038] Specifically, the morphological characteristics and power frequency phase of each candidate pulse are calculated, and preliminary screening is performed according to preset amplitude lower limit, energy lower limit, pulse width range, rise edge range and saturation criteria; candidate pulses that do not simultaneously meet at least one preset amplitude lower limit, preset energy lower limit and morphological characteristic criteria are judged as random spikes or non-discharge noise.

[0039] For subsequent pulses that appear within a preset reflection time window after the main pulse in the same channel, are highly correlated with the main pulse waveform, and have decreasing amplitude, they are marked as repeated reflection pulses and merged or removed from the main pulse; pulses that are fixedly synchronized with the gate switch but do not have statistical characteristics of the discharge phase are marked as switch interference.

[0040] A single candidate pulse is an observation of a transient signal by a particular sensor channel. Corresponding candidate pulses of the same physical discharge generated in different channels are correlated and merged into a single discharge event. For the channel and If the fixed propagation delay is corrected to satisfy If both have the same acquisition stage label and further satisfy the condition that the normalized waveform correlation coefficient is not lower than the corresponding preset threshold or the power frequency phase difference does not exceed the corresponding preset threshold, then they are considered to meet the cross-channel correlation condition.

[0041] If at least two independent channels meet the above correlation conditions, the correlated pulse group will be marked as a high-confidence discharge event; if only a single channel is triggered, the pulse will be retained only if its amplitude, energy and pulse width all meet the preset criteria, and it recurs in multiple acquisition cycles at similar power frequency phases or after shortwave gating is enabled, and its quality coefficient will be reduced; otherwise, it will not be retained.

[0042] The set of high-confidence discharge events includes high-confidence discharge events. and event quality coefficient Each high-confidence discharge event It should include at least the associated channel, occurrence time, power frequency phase, acquisition segment label, representative waveform, and quality coefficient; among which, the event quality coefficient... The calculation formula is: ; in, For cross-channel time consistency; For cross-channel waveform similarity; For power frequency phase similarity; The degree to which morphological feature criteria are met; For repeatability across acquisition cycles (i.e., the degree to which preset repeatability conditions are met); These are non-negative weighting coefficients.

[0043] Step S5: Perform multi-dimensional feature extraction on the set of credible discharge events to obtain a multi-domain feature set.

[0044] As an optional implementation, step S5 specifically includes: For each discharge event in the set of credible discharge events, time-domain morphological features, frequency-domain energy features, phase statistical features, excitation response features, and sequence dynamic features are extracted to form a multi-domain feature set; Among them, features in five dimensions were extracted, including temporal morphological features. This includes one or more of the following: pulse peak value, apparent discharge equivalent value, rise time, fall time, half width at half maximum (WWHM), pulse area, kurtosis, and skewness. Frequency domain energy characteristics This includes one or more of the following: dominant frequency, frequency band energy proportion, spectral centroid, spectral entropy, short-time energy peak, and wavelet packet energy distribution; Phase statistical characteristics This includes one or more of the following: discharge initiation phase, phase concentration, positive and negative half-cycle asymmetry, PRPD spectrum density distribution, and phase drift. Stimulus response characteristics This includes one or more of the following: discharge number gain, amplitude gain, time delay reduction rate, dose response slope, response saturation coefficient, and recovery decay time constant from dark field to excitation stage; Sequence dynamics characteristics This includes one or more of the following: adjacent discharge time interval, clustering degree, burst duration, repetition rate stability, inter-pulse energy correlation, and discharge memory factor.

[0045] Specifically, according to the acquisition cycle and dark field baseline segment Shortwave excitation segment and recovery observation segment High-confidence discharge events are grouped, and event-level features and acquisition window-level statistical features are calculated separately to form an unnormalized multi-domain feature set. .

[0046] Step S6: Construct a discharge response fingerprint vector based on the multi-domain feature set, and calculate the reliability weight of each dimension feature in the discharge response fingerprint vector.

[0047] As an optional implementation, step S6 specifically includes: Step S61: Perform physical quantity correction and statistical normalization on the features of each dimension in the multi-domain feature set to obtain the normalized features of each dimension, and construct the discharge response fingerprint vector based on the normalized features of each dimension. Step S62: Based on the event quality coefficient of each discharge event, calculate the normalized signal-to-noise ratio, cross-channel consistency coefficient, and repeatability coefficient of each dimension of the discharge response fingerprint vector. Step S63: Based on synchronous multi-source data, calculate the acquisition window integrity coefficient of each dimension of features in the discharge response fingerprint vector; Step S64: Based on the pre-calibrated sensor channel calibration parameters, calculate the calibration quality coefficients of each dimension of the features in the discharge response fingerprint vector; Step S65: Using the weighted geometric mean method, the reliability weights of each dimension of the discharge response fingerprint vector are determined based on the normalized signal-to-noise ratio, cross-channel consistency coefficient, repeatability coefficient, acquisition window integrity coefficient, and calibration quality coefficient.

[0048] Specifically, physical quantity correction and statistical normalization are performed on the characteristics of different dimensions, different sensors, and different acquisition segments.

[0049] Specifically, the physical quantity correction is as follows: the event rate is obtained by dividing the number of events by the corresponding sampling window duration; the amplitude and energy are first reduced by the background mean or median, and then divided by the sensor sensitivity or calibration coefficient; data under different applied voltages are compared within the same voltage group, or the voltage trend is removed using a reference model; the shortwave response is converted according to the measured dose or flux and irradiation duration; and the effective sampling duration is recorded at the same time to eliminate the difference in the acquisition window length.

[0050] After the above physical quantity correction, for the first Features in each dimension The reference sample mean was used. and standard deviation Standardization is performed to obtain normalized features for each dimension. : ; Each subvector is composed of normalized features within its corresponding feature domain. Arranged in a fixed order, among which... For time-domain morphological subvectors, For frequency domain energy subvectors, For phase statistics subvectors, For the excitation response subvector, These are the sequence dynamics sub-vectors; therefore, each sub-vector is a grouping result of normalized features. Thus, the obtained discharge response fingerprint vector can be expressed as: ; Excitation response subvector The differential response before and after short-wave gating and during the recovery phase is specifically preserved, so that the discharge response fingerprint can characterize the sensitivity of the defect to the initial external electron supply.

[0051] Preferably, the excitation response subvector is ,in, For the number of discharges or event rate gain, For amplitude gain, To calculate the time lag reduction rate, The dose response slope, This represents the response saturation coefficient in the high-dose range relative to the low-dose range. To restore the decay time constant. Under the same conditions of applied voltage, sensor settings, and acquisition duration, if the event rate or amplitude increases, the statistical lag shortens, the dose response is significant after the shortwave is turned on, or if persistent events persist after the shortwave is removed, then... The corresponding component in the equation increases, thus quantitatively reflecting the sensitivity of the defect to short-wavelength initial electron supply and its tendency for sustained activation. Therefore, the differential response before and after gating is incorporated. This can quantitatively reflect the sensitivity of defects to the initial short-wavelength electron supply.

[0052] For the i-th feature obtained from waveform amplitude or energy, first calculate the ratio of signal power to dark field noise power. and in accordance with = Obtain the normalized signal-to-noise ratio , This means limiting the result to the range of 0 to 1. The original signal-to-noise ratio. Minimum signal-to-noise ratio; This represents the maximum signal-to-noise ratio.

[0053] Cross-channel consistency coefficient The correlation coefficient is determined by the proportion of events that meet the correlation criteria to the total number of candidate events and their average waveform correlation coefficient; repeatability coefficient Press Calculation, where The coefficient of variation of this feature in repeated acquisition cycles; window integrity. The ratio of effective sampling points to the required sampling points; calibration quality coefficient. Press Calculation, where This represents the relative calibration error.

[0054] The reliability weights are calculated using the weighted geometric mean method. : ; in, , , , and All are limited to the range of 0 to 1. ~to Let be a non-negative weighting coefficient; when multi-channel acquisition is not performed, let =1, and simultaneously set the corresponding weight coefficient =0, so that the cross-channel consistency term does not participate in the reliability calculation, and finally S6 outputs the reliability weight vector. = Features with low signal-to-noise ratio, inconsistent across channels, poor repeatability, incomplete windows, or large calibration errors will receive lower weights.

[0055] Step S7: Based on the discharge response fingerprint vector and reliability weights, the defect risk index is obtained.

[0056] As an optional implementation, step S7 specifically includes: Step S71: Based on the information of the insulation component to be tested, the set of acquisition conditions and environmental parameters, a matching reference subset is selected from the preset risk model; the matching reference subset includes the safe endpoints and risk endpoints of each feature; Step S72: Compare the features of each dimension in the discharge response fingerprint vector with the safe endpoints and risk endpoints in the matching reference subset, and map each feature to a single risk score in the range of 0 to 1. Step S73: Multiply the individual risk scores of each dimension feature in the discharge response fingerprint vector with the corresponding reliability weights and the preset feature fusion weights to obtain the unnormalized contribution values ​​of each dimension feature. Step S74: The risk scores of each individual item corresponding to the excitation response sub-vector in the discharge response fingerprint vector are weighted and averaged to obtain the short-wave excitation comprehensive item. Step S75: Determine the memory term for the recovery segment based on the ratio of the event rates of the recovery observation segment to the shortwave excitation segment and the recovery time constant; Step S76: Summate and normalize the unnormalized contribution values ​​of each dimension feature, the shortwave excitation comprehensive term, and the recovery segment memory term to obtain the defect risk index.

[0057] Specifically, firstly, a reference subset matching the current detection conditions is selected from the preset risk model according to the equipment type, sensor configuration, applied voltage range, shortwave dose mode, and environmental conditions; then, each feature in the current discharge response fingerprint vector is compared with the safe endpoint and risk endpoint given by the reference subset.

[0058] For the i-th normalized feature The safe endpoint is determined by the reference subset. and risk endpoints This feature is then mapped to a single risk score. : ; When the eigenvalue decreases as risk increases, Less than The above formula still completes the mapping according to the corresponding risk direction.

[0059] Preferably, the defect risk index The calculation formula is: ; in, This is the normalized mapping function; This represents the individual risk score of the i-th feature relative to the reference model. Reliability weight; Let the fusion weight of the i-th feature satisfy... ≥0 and =1, which can be determined by confirmed sample training, cross-validation, expert weighting or a combination thereof; and These are the adjustment coefficients for the shortwave excitation synthesis term and the memory recovery term, respectively. (Shortwave excitation synthesis term) It can be obtained by weighted average of the individual risk scores within the stimulus response domain; the memory recovery term can be obtained by... = Calculation, where It is a normalized function that monotonically increases with the recovery time constant. and These represent the event rates for the recovery observation segment and the shortwave excitation segment, respectively. The event rate used to construct the shortwave excitation synthesis term is... and recovery segment memory items The characteristics of these risks will not be repeatedly included in the basic risk items. The unnormalized contribution of the i-th feature to the risk fusion result is: Normalized contribution is = The contribution of the k-th feature domain is = .

[0060] Step S8: Determine the risk level based on the defect risk index, and determine the defect type based on the discharge response fingerprint vector and reliability weight.

[0061] As an optional implementation, step S8 specifically includes: Step S81: When the defect risk index is greater than or equal to 0 and less than the first preset risk threshold, the risk level of the insulation component to be tested is determined to be low risk level. Step S82: When the defect risk index is greater than or equal to the first preset risk threshold and less than the second preset risk threshold, the risk level of the insulation component to be tested is determined to be a risk level of concern. Step S83: When the defect risk index is greater than or equal to the second preset risk threshold and less than the third preset risk threshold, the risk level of the insulation component to be tested is determined to be the warning risk level. Step S84: When the defect risk index is greater than or equal to the third preset risk threshold and less than 1, the risk level of the insulation component to be tested is determined to be a severe risk level. Step S85: Based on the reference defect category model, obtain the reference center vector corresponding to each preset defect category; Step S86: Calculate the reliability weighted distance between the discharge response fingerprint vector and the reference center vector of each preset defect category, and take the defect category with the smallest reliability weighted distance as the final defect type; the reliability weighted distance is obtained by weighting the distance based on the reliability weight of each dimension feature.

[0062] Specifically, three risk thresholds are determined based on a reference sample confirmed through retesting, quantile statistics, or receiver operating characteristic curves. The first pre-set risk threshold is... Second preset risk threshold and the third preset risk threshold .

[0063] Wherein, when DRI is located at [ When the DRI is within the specified range, the risk level of the insulation component under test is low risk; when the DRI is within the specified range... When the DRI is within the specified range, the risk level of the insulation component under test is the level of concern; when the DRI is within the specified range... When the DRI is within the specified range, the risk level of the insulation component under test is the warning risk level; when the DRI is within the specified range... When the range is defined, the risk level of the insulating component under test is classified as severe risk.

[0064] Defect type can be determined by the reliability-weighted distance between the current discharge response fingerprint vector F and the reference center vectors of each preset defect category. The category with the smallest distance or the highest similarity is determined as the final defect type; alternatively, a classification model trained on confirmed samples can be used for equivalent determination.

[0065] If the DRI is high and the excitation response domain contributes significantly, the defect is considered to have strong excitability; if the phase domain and sequence dynamics domain contribute significantly, the defect is considered to have a stable discharge development trend; if only the single-channel amplitude characteristic is high and... and If the output is low, it indicates noise or external interference.

[0066] As an optional implementation, the method further includes: step S9, wherein step S9 specifically includes: Step S91: Based on the defect risk index, single risk score vector, reliability weight, and preset feature fusion weight, determine the normalized contribution value of each dimension feature to obtain the feature contribution vector. Then, summarize the normalized contribution values ​​of each dimension feature according to the feature domain to obtain the contribution degree of each feature domain. The single risk score vector is obtained based on the single risk score of all dimension features. Step S92: Based on the defect risk index, individual risk score vector, feature contribution vector, contribution degree of each feature domain, and reference defect category model, determine the preliminary diagnosis results; the preliminary diagnosis results include candidate defect type, risk level, defect risk index, category similarity, key contribution features, and contribution degree of each feature domain. Step S93: Based on the preliminary diagnostic results, reliability weight vector, acquisition window integrity coefficient, repetition period judgment consistency rate, and defect category similarity, determine the diagnostic confidence and output it.

[0067] Specifically, the formula for calculating diagnostic confidence is: = ; in, , Average reliability of key features; To ensure average window integrity; To determine the consistency rate for repeated cycles; The normalized interval between the highest and second-highest category similarity; The acquisition conditions should include at least the identification of the device under test and the area, the sensor type and installation location, the sampling rate, bandwidth, range, gain, applied voltage amplitude and frequency, power frequency phase reference, ambient temperature, gas pressure, and calibration records; the gating excitation parameters should include at least the shortwave source type, energy or spectral range, dose or flux, pulse width, repetition interval, on / off time, irradiation location, and the geometric relationship between the source and the area under test.

[0068] The final output is a traceable diagnostic record. : For samples confirmed through retesting, disassembly inspection, or other independent means, their discharge response fingerprint vector, acquisition condition set, and final label are written into the reference fingerprint database, and the safe endpoint, risk endpoint, preset feature fusion weight, or reference defect category model are updated in a versioned manner; unconfirmed samples are not directly used as supervision labels to update the model.

[0069] Beneficial effects: This application addresses the problems of single feature set, sparse pulses from minute defects, misjudgments due to noise interference, and the difficulty in unifying and fusing data from multiple sources in traditional partial discharge diagnosis by employing shortwave gating excitation, cross-channel correlation screening, multi-domain feature fusion, and reliability weight calculation. Furthermore, it enables quantitative output of defect activation level, risk level, and diagnostic confidence without relying on a specific air gap sample structure. Specifically: (1) By setting a gated excitation timing table that includes a short-wave excitation segment, external initial electrons are provided to the defect region, so that early small defects can generate a stable and repeatable discharge response under a low electric field, which effectively solves the problem of sparse discharge pulses and difficulty in capturing them in conventional detection.

[0070] (2) By screening discharge events through cross-channel correlation conditions, the same discharge event must be detected by at least two independent channels at the same time to be identified. This effectively suppresses occasional noise and external interference in single channels and avoids the problem of misjudgment of a single indicator in complex field environments.

[0071] (3) By extracting multi-dimensional features from high-confidence discharge events, the discharge characteristics are comprehensively characterized from five dimensions: time domain morphology, frequency domain energy, phase statistics, excitation response and sequence dynamics, overcoming the limitations of traditional methods that rely on a single discharge quantity or a single PRPD spectrum for judgment.

[0072] (4) By constructing a unified discharge response fingerprint vector, multi-domain features are integrated into a unified representation that is calculable, comparable, and traceable, filling the gap in existing shortwave studies where there is a lack of systematic fusion schemes for discharge data before and after excitation.

[0073] (5) By calculating the reliability weight of each dimension of features, the contribution of features with low signal-to-noise ratio, poor cross-channel consistency and weak repeatability to the fusion result is reduced, and high-quality features dominate the diagnostic conclusion, which further improves the stability and anti-interference ability of the diagnostic result.

[0074] (6) By fusing the discharge response fingerprint vector with the preset risk model, the defect risk index is obtained, and the defect type and risk level are determined accordingly, providing a clear quantitative basis for operation and maintenance decisions.

[0075] The present application will be described below with reference to embodiments. These embodiments are for illustrative purposes only and do not limit the scope of protection of this application. Any equivalent substitutions made to the acquisition channel, shortwave source type, feature extraction algorithm, or fusion model under the concept of this application shall fall within the scope of protection of this application.

[0076] Example 1: Risk Identification of Minimal Insulation Defects Based on Shortwave Gated Excitation Connect the GIS basin-type insulator or epoxy insulator under test to a conventional partial discharge test circuit. With the external power frequency voltage maintained at the preset detection voltage, first acquire 30 seconds of dark field baseline data, then apply 10 seconds of shortwave gated excitation, followed by 60 seconds of recovery observation data. The acquired data includes high-frequency current pulses, ultra-high frequency signals, applied voltage, power frequency phase, and shortwave dose.

[0077] After preprocessing the synchronous multi-source dataset, the system identifies high-confidence discharge events and divides them into dark-field baseline segments, short-wave excitation segments, and recovery observation segments. If the number of discharges per unit time increases significantly after short-wave gating is activated, and the discharge phase is concentrated in the power frequency voltage rise range, while short-term discharge clusters still exist in the recovery observation segment, it indicates that the defect has a high excitability and a tendency to be continuously activated.

[0078] The system further extracts time-domain morphological features, frequency-domain energy features, phase statistical features, excitation response features, and sequence dynamics features to construct a discharge response fingerprint vector. By comparing this vector with a preset risk model and performing reliability weight fusion, a defect risk index is obtained. If the defect risk index is within the warning range, and the excitation response features and recovery segment memory features contribute significantly, a diagnostic suggestion of "suspected early-stage dangerous defect, retesting or scheduled power outage for detailed inspection" is output.

[0079] Example 2: Multi-channel consistency suppresses noise misjudgment During field testing, certain switching actions or external pulse interference may generate high-amplitude spikes in a single high-frequency current sensor channel. When using the traditional amplitude thresholding method, these spikes may be misjudged as severe partial discharge. This application performs cross-channel consistency judgment on these pulses: if the pulse does not generate a corresponding response in the ultra-high frequency sensor channel or phase statistics, and does not show increased repeatability with the opening of the shortwave gating, the system reduces the reliability weight of the corresponding feature of the pulse.

[0080] In this case, even if the amplitude of a single pulse is high, its contribution to the defect risk index is still limited, thereby avoiding high-risk misjudgments caused by sporadic noise.

[0081] Example 3: Defect Type Assisted Identification When the defect is a closed air gap or interface peeling, the short-wave excitation segment may show a significant increase in the number of discharges, a shortened statistical time delay, and a concentrated phase distribution; when the defect is metal particle adhesion, it may show an increase in the proportion of high-frequency energy and an enhanced asymmetry between the positive and negative half cycles; when the defect is surface contamination, it may show a longer duration of the recovery observation segment and an increased degree of sequence clustering.

[0082] This application does not rely on a single empirical spectrum, but instead achieves defect type-assisted identification through the combined contribution of multiple sub-features in the discharge response fingerprint vector. Therefore, this application, through the fusion of multi-dimensional, multi-source discharge response fingerprints under short-wavelength gating excitation, can effectively distinguish the differences in discharge characteristics among different types of defects, improving the accuracy and generalization ability of defect type identification.

[0083] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A risk identification method based on shortwave gating excitation and defect discharge response fingerprinting, characterized in that, The method includes: Acquire information about the insulation component under test and the initial test configuration, and generate a gating excitation timing table and a set of acquisition conditions; each acquisition cycle of the gating excitation timing table includes a shortwave excitation segment; Based on the gating excitation timing table and the acquisition condition set, multi-source discharge signals and synchronization parameters are acquired synchronously. After time alignment and the addition of acquisition stage labels, a synchronous multi-source dataset is output. The synchronous multi-source dataset is preprocessed, and candidate pulses for each channel are extracted; Discharge events are filtered based on candidate pulses and cross-channel correlation conditions, and a set of reliable discharge events is output. Multi-dimensional feature extraction is performed on the set of credible discharge events to obtain a multi-domain feature set; A discharge response fingerprint vector is constructed based on a multi-domain feature set, and the reliability weights of each dimension of the features in the discharge response fingerprint vector are calculated. The defect risk index is obtained based on the discharge response fingerprint vector and reliability weights; The risk level is determined based on the defect risk index, and the defect type is determined based on the discharge response fingerprint vector and reliability weight.

2. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, Information on the insulation components to be tested includes equipment type, area under test, and area of ​​concern for defects. Initial test configuration includes external voltage application method, sensor channel, sampling rate, shortwave source type, allowable dose range, and number of acquisition cycles. Acquire information about the insulation component under test and the initial test configuration, and generate a gating excitation timing table and a set of acquisition conditions, specifically including: Acquire information about the insulation component under test and the initial detection configuration. Based on the information about the insulation component under test and the initial detection configuration, set at least one acquisition cycle and divide each acquisition cycle into a dark field baseline segment, a shortwave excitation segment and a recovery observation segment in sequence, and set the start and end times of each stage respectively. Set the applied voltage, shortwave dose, gating pulse width, repetition interval, and number of repetitions for each acquisition cycle; The start and end times of each stage are arranged with the applied voltage, shortwave dose, gating pulse width, repetition interval and repetition number of each acquisition cycle according to a unified clock to form a gating excitation timing table. The applied voltage method, sensor channel, sampling rate, shortwave source type, allowable dose range and number of acquisition cycles are set as acquisition conditions and output.

3. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, The multi-source discharge signal includes the original waveforms of each partial discharge sensor channel, and the synchronization parameters include: applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp, and environmental parameters. Based on the gating excitation timing table and acquisition condition set, multi-source discharge signals and synchronization parameters are acquired synchronously. After time alignment and the addition of acquisition stage labels, a synchronous multi-source dataset is output, which specifically includes: Based on the gating excitation timing table and the acquisition condition set, the applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp, environmental parameters, and the original waveforms of each partial discharge sensor channel are synchronously acquired. The data collected by each partial discharge sensor channel is time-aligned using a unified sampling clock, hardware trigger signal or timestamp, and the arrival time deviation caused by each partial discharge sensor and transmission link is corrected by using a pre-calibrated fixed channel delay. Based on the start and end times of each stage in the gating excitation timing table, each sampling point is assigned a sampling cycle number and a sampling stage label. The applied voltage, power frequency phase, shortwave gating signal, shortwave dose, sampling timestamp, and environmental parameters are associated with the same time base, and a synchronous multi-source dataset is output.

4. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, The synchronous multi-source dataset is preprocessed, and candidate pulses for each channel are extracted, specifically including: Bandpass filtering and baseline drift correction matching the effective frequency band of the corresponding sensor are performed on each partial discharge sensor channel in the synchronous multi-source dataset to obtain the preprocessed waveform of each channel; Based on the data from the dark field baseline segment of the synchronous multi-source dataset, the noise mean, noise standard deviation, and noise energy of each partial discharge sensor channel are estimated. Based on the noise standard deviation and the first coefficient of each channel, the noise peak threshold corresponding to each channel is determined; Based on the noise energy and second coefficient of each channel, the noise energy threshold corresponding to each channel is determined. When the peak value of the preprocessed waveform of any channel is greater than or equal to the peak value threshold of the corresponding channel, or when the energy within the preset pulse window of the preprocessed waveform of any channel is greater than or equal to the noise energy threshold of the corresponding channel, a local waveform segment of preset length located on both sides of the trigger time on the preprocessed waveform of the corresponding channel is extracted as the candidate pulse of the corresponding channel.

5. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, Cross-channel correlation conditions include cross-channel time consistency conditions, tag consistency during the acquisition phase, and waveform and phase similarity conditions; Discharge events are filtered based on candidate pulses and cross-channel correlation conditions, and a set of reliable discharge events is output, specifically including: The morphological characteristics and power frequency phase of each candidate pulse are calculated. Based on the morphological characteristics and power frequency phase, each candidate pulse is initially screened according to preset criteria to remove random spikes and non-discharge noise that do not meet any preset criteria, thus obtaining the initially screened candidate pulses. The morphological characteristics include peak value, pulse energy, rise time, fall time, half peak width, and waveform correlation coefficient. The preset criteria include one or more of the following: preset amplitude lower limit, preset energy lower limit, preset pulse width range, and preset rise edge range. After removing duplicate reflected pulses and switching interference from the initial candidate pulses, valid candidate pulses are obtained. Each valid candidate pulse that simultaneously meets the cross-channel time consistency condition, the acquisition stage tag consistency condition, and the waveform and phase similarity condition is associated as the same discharge event. All discharge events that meet the cross-channel association condition and are successfully associated are judged as high-confidence discharge events, and the event quality coefficient of each high-confidence discharge event is determined. Valid candidate pulses that do not meet the cross-channel association condition, meet the preset repeatability condition, and are triggered by a single channel are judged as low-confidence discharge events, and the event quality coefficient of each low-confidence discharge event is determined. The event quality coefficient of a low-confidence discharge event is less than that of a high-confidence discharge event. The preset repeatability condition is that the event pulses recur in multiple acquisition cycles at similar power frequency phases or after shortwave gating is enabled. Based on all high-confidence and low-confidence discharge events, a set of credible discharge events is constructed; each discharge event carries a corresponding event quality coefficient.

6. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, Multi-dimensional feature extraction is performed on the set of credible discharge events to obtain a multi-domain feature set, specifically including: For each discharge event in the set of credible discharge events, time-domain morphological features, frequency-domain energy features, phase statistical features, excitation response features, and sequence dynamic features are extracted to form a multi-domain feature set; Among them, the temporal morphological characteristics include one or more of the following: pulse peak value, apparent discharge equivalent value, rise time, fall time, half width at half maximum (WWHM), pulse area, kurtosis, and skewness. Frequency domain energy characteristics include one or more of the following: dominant frequency, frequency band energy proportion, spectral centroid, spectral entropy, short-time energy peak, and wavelet packet energy distribution. Phase statistical characteristics include one or more of the following: discharge initiation phase, phase concentration, positive and negative half-cycle asymmetry, PRPD spectrum density distribution, and phase drift. Excitation response characteristics include one or more of the following: discharge number gain from dark field to excitation segment, amplitude gain, time delay reduction rate, dose response slope, response saturation coefficient, and recovery decay time constant; Sequence dynamics characteristics include one or more of the following: adjacent discharge time interval, clustering degree, burst duration, repetition rate stability, inter-pulse energy correlation, and discharge memory factor.

7. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, A discharge response fingerprint vector is constructed based on a multi-domain feature set, and the reliability weights of each dimension of the features in the discharge response fingerprint vector are calculated, specifically including: Physical quantity correction and statistical normalization are performed on the features of each dimension in the multi-domain feature set to obtain the normalized features of each dimension. Based on the normalized features of each dimension, the discharge response fingerprint vector is constructed. Based on the event quality coefficient of each discharge event, the normalized signal-to-noise ratio, cross-channel consistency coefficient and repeatability coefficient of each dimension of the discharge response fingerprint vector are calculated respectively. Based on synchronous multi-source data, the acquisition window integrity coefficient of each dimension of features in the discharge response fingerprint vector is calculated; Based on the pre-calibrated sensor channel calibration parameters, the calibration quality coefficients of each dimension of the discharge response fingerprint vector are calculated. The weighted geometric mean method is used to determine the reliability weights of each dimension of the discharge response fingerprint vector based on the normalized signal-to-noise ratio, cross-channel consistency coefficient, repeatability coefficient, acquisition window integrity coefficient, and calibration quality coefficient.

8. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, Based on the discharge response fingerprint vector and reliability weights, a defect risk index is obtained, which specifically includes: Based on the information of the insulation component under test, the set of acquisition conditions and environmental parameters, a matching reference subset is selected from the preset risk model; the matching reference subset includes the safe endpoints and risk endpoints of each feature; Each dimension of the discharge response fingerprint vector is compared with the safe endpoint and risk endpoint in the matching reference subset, and each dimension of the feature is mapped to a single risk score in the range of 0 to 1. The individual risk score of each feature in the discharge response fingerprint vector is multiplied by the corresponding reliability weight and the preset feature fusion weight to obtain the unnormalized contribution value of each feature. The shortwave excitation comprehensive term is obtained by weighting and averaging the individual risk scores corresponding to the excitation response sub-vectors in the discharge response fingerprint vector. The memory term for the recovery segment is determined based on the ratio of the event rates of the recovery observation segment to the shortwave excitation segment and the recovery time constant. The defect risk index is obtained by summing and normalizing the unnormalized contribution values ​​of each dimension's features, the shortwave excitation comprehensive term, and the recovery segment memory term.

9. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, Risk levels are determined based on a defect risk index, and defect types are determined based on discharge response fingerprint vectors and reliability weights, specifically including: When the defect risk index is greater than or equal to 0 and less than the first preset risk threshold, the risk level of the insulation component to be tested is determined to be low risk level. When the defect risk index is greater than or equal to the first preset risk threshold and less than the second preset risk threshold, the risk level of the insulation component to be tested is determined to be a risk level of concern. When the defect risk index is greater than or equal to the second preset risk threshold and less than the third preset risk threshold, the risk level of the insulation component to be tested is determined to be the warning risk level. When the defect risk index is greater than or equal to the third preset risk threshold and less than 1, the risk level of the insulation component to be tested is determined to be a severe risk level. Based on the reference defect category model, obtain the reference center vector corresponding to each preset defect category; The reliability-weighted distance between the discharge response fingerprint vector and the reference center vector of each preset defect category is calculated one by one, and the defect category with the smallest reliability-weighted distance is taken as the final defect type. The reliability-weighted distance is obtained by weighting the distance based on the reliability weight of each dimension feature.

10. The risk identification method based on shortwave gating excitation and defect discharge response fingerprinting according to claim 1, characterized in that, The method further includes: Based on the defect risk index, individual risk score vector, reliability weight, and preset feature fusion weight, the normalized contribution value of each dimension feature is determined, thereby obtaining the feature contribution vector. The normalized contribution values ​​of each dimension feature are then summarized according to the feature domain to obtain the contribution degree of each feature domain. The individual risk score vector is obtained based on the individual risk scores of all dimension features. Based on the defect risk index, individual risk score vector, feature contribution vector, contribution of each feature domain, and reference defect category model, preliminary diagnostic results are determined. The preliminary diagnostic results include candidate defect type, risk level, defect risk index, category similarity, key contribution features, and contribution of each feature domain. Based on the preliminary diagnostic results, reliability weight vector, acquisition window integrity coefficient, repetition period judgment consistency rate, and defect category similarity, the diagnostic confidence is determined and output.

Images