Partial discharge real-time monitoring and early warning system based on multi-source data fusion

Abstract

The invention relates to the technical field of discharge monitoring and early warning, in particular to a partial discharge real-time monitoring and early warning system based on multi-source data fusion, which is characterized in that decision tree classification is carried out on a mixed transient signal flow through a signal physical separation module, and pure electromagnetic wave speed type and mechanical wave speed type signal sources are separated; the feature marking module extracts feature points with a unified time reference and marks physical attribute categories; the double-path credibility fusion module executes time difference logic verification on the two types of feature points, a credible discharge event is marked only when an electromagnetic wave verification result and a mechanical wave verification result are logic values 1 at the same time, event confidence parameters are generated through dynamic confidence fusion, and physical interlocking verification and uniqueness locking are achieved; the anti-interference response module dynamically adjusts a response threshold value based on the event confidence parameter, and only triggers a targeted response to the threat which is physically and essentially confirmed to be real discharge and has confidence reaching the standard; the problem of distinguishing a real discharge event from multi-source interference is solved, and the real discharge event is reliably identified.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of discharge monitoring and early warning, in particular to a local discharge real-time monitoring and early warning system based on multi-source data fusion. BACKGROUND

[0002] The field of local discharge monitoring of power equipment has long been plagued by the bottleneck problem of multi-source interference coupling. Traditional monitoring systems rely on a single type of sensor (such as ultra-high frequency or ultrasonic wave) to capture transient signals, but in actual working conditions, electromagnetic noise, mechanical vibration and environmental interference will mix with real discharge signals, forming a mixed transient signal stream with similar time-frequency characteristics. Existing technologies cannot effectively distinguish between electromagnetic wave signals (such as ultra-high frequency radiation) and mechanical wave signals (such as ultrasonic wave propagation), resulting in a risk of physical nature misjudgment in the feature extraction process.

[0003] In addition, the single-path verification mechanism (such as relying only on spatial time difference or structure propagation model) cannot verify the physical consistency of the signal propagation path, and is easily affected by multi-path reflection, medium attenuation distortion and other factors, resulting in false alarms. More importantly, a fixed response threshold cannot adapt to complex working conditions such as changes in device load, fluctuations in environmental temperature and humidity, and often requires sacrificing monitoring sensitivity to suppress false positives, ultimately leading to an increased rate of missed detection of early weak discharge. SUMMARY

[0004] The purpose of the present application is to provide a local discharge real-time monitoring and early warning system based on multi-source data fusion to solve the problems raised in the background art. The specific technical problems include how to realize double-path cross-verification and dynamic confidence fusion of electromagnetic wave signals and mechanical wave signals to solve the physical nature distinction problem between real discharge events and multi-source interference in the mixed transient signal stream.

[0005] To achieve the above purpose, the present application provides the following technical solutions: The local discharge real-time monitoring and early warning system based on multi-source data fusion includes a signal physical separation module, a feature marking module, a double-path confidence fusion module and an anti-interference response module, wherein: The signal physical separation module obtains the original waveform of the mixed transient signal stream, uses a classification engine to extract key feature attributes of the mixed transient signal stream, including signal initial rising edge time, main frequency band concentration area and relative arrival time difference between multiple sensors; a multi-level decision node is established according to the key feature attributes through a decision tree algorithm, and three-level physical feature verification is performed in turn using the signal initial rising edge time, the main frequency band concentration area and the relative arrival time difference between multiple sensors, and classification is completed when the three-dimensional features meet the physical law of the same type of wave; wherein the decision process of the decision tree algorithm specifically includes: The first level node is based on the comparison of the initial rising edge time of the signal and the preset time threshold, wherein if the initial rising edge time of the signal is lower than the preset time threshold, the electromagnetic wave speed type branch is entered, otherwise the mechanical wave speed type branch is entered; The second level node is based on the energy distribution proportion of the signal energy in the preset frequency band threshold in the main frequency band concentrated area, wherein if the signal energy exceeds the preset frequency band threshold, the electromagnetic wave speed type candidate signal is taken, otherwise the reinforced mechanical wave speed type candidate signal is taken; The third level node verifies the physical consistency of the determination conclusions of the previous two levels based on the deviation tolerance of the relative arrival time difference between the multiple sensors and the theoretical wave speed propagation model.

[0006] The signal physical separation module is used for separating the physical attributes of the electromagnetic wave speed and the mechanical wave speed signals, eliminating the aliasing of the interference signals in a single feature dimension; through three-level physical law verification (rising edge speed, spectrum energy, multi-source time difference correlation), it is ensured that the two types of signals input into the subsequent module meet the respective propagation nature, and pure electromagnetic wave speed type and mechanical wave speed type signal sources are provided for double-path cross verification.

[0007] The feature marking module is configured to extract feature points with a unified time reference from the original waveforms of the electromagnetic wave speed type signal and the mechanical wave speed type signal, and mark the electromagnetic wave speed type and the mechanical wave speed type for each feature point. Specifically, a unified time reference is established by using a high-precision time unified signal source, three types of feature points, i.e. pulse starting point, amplitude extreme point and oscillation decay inflection point, are extracted by using a method combining adaptive threshold detection and multi-scale morphological analysis, and a time stamp is assigned to each type of feature point.

[0008] The feature marking module constructs a time-space reference system across sensors, and the unified time reference ensures that the time stamps of the feature points of the electromagnetic wave speed and the mechanical wave speed signals are strictly comparable; the three types of feature points form standardized time-space markers (the pulse starting point locates the propagation starting point, the amplitude extreme point quantifies the energy release, and the oscillation decay inflection point reveals the path characteristics), which provide synchronized and correlatable input data for the time difference logic verification of the double paths, and support the accurate time difference comparison of the electromagnetic wave spatial propagation and the mechanical wave structure propagation.

[0009] The verification unit in the double-path credibility fusion module performs spatial propagation time difference logic verification on the electromagnetic wave speed type feature points, calculates the multi-sensor measured time difference based on the device three-dimensional coordinate model and the electromagnetic wave theoretical propagation model, verifies the consistency of the measured time difference and the theoretical time difference to generate an electromagnetic wave verification result (logical value 1 / 0), and generates an electromagnetic credibility value by comprehensively considering the amplitude spatial distribution of the amplitude extreme point and the frequency damping characteristics of the oscillation decay inflection point.

[0010] The verification unit performs structural travel-time logic verification on the mechanical wave velocity type feature point, calculates the predicted travel-time through the device entity material acoustic model, verifies the consistency between the measured travel-time and the predicted value to generate the mechanical wave verification result (logical value 1 / 0); and generates the mechanical credible value by combining the amplitude gradient distribution of the amplitude extreme point and the oscillation decay inflection point boundary reflection effect.

[0011] The verification unit is used to realize the physical interlocking verification of the electromagnetic wave and mechanical wave propagation path. When the electromagnetic wave verification result is 1, it indicates that the signal conforms to the law of spatial light speed propagation, excluding slow conduction interference; when the mechanical wave verification result is 1, it indicates that the signal fits the structure acoustic propagation model, suppressing non-discharge vibration noise; only when the double-path verification result is 1 at the same time, mark the credible discharge event, from the essence of physical propagation, lock the uniqueness of the real discharge (such as only the real discharge event can produce electromagnetic radiation conforming to the law of spatial light speed propagation and mechanical vibration conforming to the law of solid sound speed propagation at the same time).

[0012] The fusion unit marks the current signal as a credible discharge event when the electromagnetic wave verification result is logical value 1 and the mechanical wave verification result is logical value 1; and assigns the electromagnetic credible value and the mechanical credible value to the reference weight of the final parameter based on the device type preset basic weight; introduces a feature correction mechanism on the reference weight, dynamically adjusts the double-path weight distribution using the energy release intensity ratio represented by the amplitude extreme point, and simultaneously performs reverse compensation adjustment on the corresponding path weight according to the propagation path complexity reflected by the oscillation decay inflection point, and outputs the event confidence parameter of continuous scale after normalization processing.

[0013] The fusion unit is used to dynamically fuse the confidence contribution of the double path. The energy intensity ratio adjustment gives higher weight to the strong energy release path (reflecting the main factor of discharge hazard); the path complexity compensation suppresses the interference of medium distortion on the credibility of a single propagation path (such as the influence of sound wave attenuation in oil-immersed equipment); makes the confidence parameter objectively quantify the joint credibility of the real discharge event in the electromagnetic and mechanical double paths, and solves the misjudgment caused by local interference (such as sensor failure, medium unevenness) in traditional single-path verification.

[0014] The anti-interference response module dynamically adjusts the response threshold based on the event confidence parameter: first match it to the preset three-level static reference interval, dynamically adjust the response threshold proportion coefficient according to the confidence interval to generate a dynamic response threshold; only when the credible discharge event is marked as true and the event confidence parameter exceeds the dynamic threshold, the alarm response is triggered, the process includes dividing multiple alarm levels according to the event confidence parameter value, generating early warning escalation decision combining historical confidence trend, recording confidence parameter change trajectory and response time sequence.

[0015] The anti-interference response module is used for driving a dynamic response based on a confidence parameter of double-path fusion. A dynamic threshold mechanism automatically increases a response threshold in a strong interference environment (such as a lightning electromagnetic pulse), to avoid false alarms. For early discharge (such as a corona on the surface of an insulator) that is weak but has a slowly rising confidence, a historical trend analysis is used to give an early warning to reduce a missed alarm. Finally, only a threat that is physically confirmed as real discharge and has a confidence up to standard triggers a targeted response, to distinguish a multi-source interference from a real discharge event.

[0016] Compared with the prior art, the present application has the following beneficial effects: The double-path credibility fusion module is used for physically interlocking verification of electromagnetic wave and mechanical wave propagation paths. When the double-path verification result is simultaneously a logic value 1, a credible discharge event is marked, to lock the uniqueness of real discharge from a physical propagation nature. Meanwhile, event confidence parameters are dynamically fused based on the energy release intensity ratio and the propagation path complexity, and finally, only a threat that is physically confirmed as real discharge and has a confidence up to standard triggers a targeted response by the anti-interference response module, to solve the problem of distinguishing a real discharge event from a multi-source interference in a mixed transient signal stream, and to reliably identify a real discharge event. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The figure is a schematic diagram of the overall module of the present application; Figure 2 The figure is a schematic diagram of the double-path credibility fusion module unit of the present application; Figure 3 The figure is a schematic diagram of the verification unit process of the present application; Figure 4 The figure is a schematic diagram of the fusion unit process of the present application.

[0018] In the figure: 100, signal physical separation module; 200, feature marking module; 300, double-path credibility fusion module; 301, verification unit; 302, fusion unit; 400, anti-interference response module. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] Next, please refer to Figure 1 The present application provides a technical solution: a partial discharge real-time monitoring and early warning system based on multi-source data fusion, which comprises a signal physical separation module 100, a feature marking module 200, a double-path credibility fusion module 300 and an anti-interference response module 400.

[0021] The signal physical separation module 100 is the physical basis for the entire system to realize preliminary information processing, and its core task is to classify and synchronously record the original mixed signals collected from multiple measuring points of the electrical equipment based on the differences in different propagation media and wave speed characteristics excited by partial discharge phenomena; The specific technical process is as follows: Receive the original waveform of each mixed transient signal stream detected by the sensor array (such as ultra-high frequency sensor, ultrasonic sensor, high-frequency current transformer, etc.); Scan each mixed transient signal stream using a classification engine (which may include methods such as wavelet packet analysis, multi-resolution spectrum analysis, or other feature extraction methods) to extract its key feature attributes, including signal initial rising edge time, main frequency band concentration area, and relative arrival time difference between multiple sensors; Based on the key feature attributes corresponding to each mixed transient signal stream, use the decision tree algorithm to classify the mixed transient signal stream into electromagnetic wave speed class signals and mechanical wave speed class signals, wherein the classification process specifically includes: First, establish multiple levels of decision nodes, each node corresponding to a key physical feature attribute. The first level of nodes focuses on the signal initial rising edge time. Since electromagnetic waves propagate at a speed close to the speed of light, the transient pulses they induce usually have a steep front with a sub-microsecond or even shorter duration. Mechanical waves (such as ultrasonic waves) have a propagation speed that is constrained by the density and elastic modulus of the medium, with a typical value of about kilometers per second, corresponding to a pulse front that is broadened to more than a microsecond. Therefore, by extracting the actual duration of the pulse rising edge through high-precision time domain analysis, when the value is below the preset time threshold, the signal is determined to enter the electromagnetic wave speed class branch, otherwise it enters the mechanical wave speed class branch; The second level of nodes analyzes the main frequency band concentration area. Electromagnetic wave signals have their energy mainly distributed in the hundreds of megahertz to gigahertz frequency band due to the high-frequency characteristics of the discharge source and electromagnetic coupling effects. Mechanical wave signals are constrained by the frequency response of the transducer and the propagation attenuation, with the main frequency band concentrated in the tens of kilohertz to hundreds of kilohertz interval. Therefore, using real-time spectrum analysis techniques, compare the signal energy in the main frequency band concentration area with the preset frequency band threshold. If the signal energy exceeds the preset frequency band threshold, it is considered as an electromagnetic wave speed class candidate signal, otherwise it is considered as a reinforced mechanical wave speed class candidate signal; The third level of nodes verifies the relative arrival time difference between multiple sensors. Based on the known sensor spatial coordinates, the algorithm constructs a theoretical wave speed propagation model. For electromagnetic wave speed class candidate signals, check whether the arrival time difference of each sensor signal conforms to the light speed level propagation delay. For mechanical wave speed class candidate signals, check whether the time difference conforms to the sound speed level propagation delay. This node performs strict physical consistency verification. If the measured time difference deviates from the theoretical model by more than the tolerance, the previous two levels of judgment are overturned and a reclassification mechanism is triggered. The process of the reclassification mechanism specifically includes: Returning to the second level node, the candidate signal class is reclassified according to the energy distribution proportion of the main frequency band concentrated region; if the reclassification still cannot pass the third level verification, the auxiliary judgment node is activated; the oscillation modal damping ratio of the signal waveform is analyzed to further distinguish the electromagnetic oscillation decay and mechanical resonance decay characteristics, and finally the physical nature classification is completed.

[0022] Finally, the decision tree algorithm forms a classification conclusion through three layers of physical feature verification, and only when the signal initial rising edge time, the main frequency band concentrated region, and the relative arrival time difference among multiple sensors meet the physical law of the same class of wave, it is determined as an electromagnetic wave speed class signal or a mechanical wave speed class signal; the whole process is executed in real time in the embedded system in the pipeline architecture, ensuring that each transient event completes the physical nature classification within milliseconds; among them, for the transition state signal that cannot meet the three-dimensional physical law at the same time, it is marked as a to-be-processed signal and executes a buffering mechanism, temporarily buffers it to an independent queue, and dynamically reclassifies it by combining the feature change trend of the subsequent continuous sampling period (such as the shortening or lengthening of the rising edge time); if it still cannot be classified, the diagnostic support mechanism is triggered and the original waveform data is retained; the process of the diagnostic support mechanism specifically includes: Call the key feature attributes (signal initial rising edge time, main frequency band concentrated region, relative arrival time difference among multiple sensors) extracted in the signal physical separation module 100, combine the oscillation decay inflection point features in the dual-path credibility fusion module, and generate a multi-dimensional feature comparison atlas; compare the atlas with the waveform features marked as typical interference in the historical cache (based on the Euclidean distance to calculate the difference degree of the key feature vector), if the difference degree is lower than the preset threshold, it is classified as interference; otherwise, the original waveform data is retained and the artificial review interface is activated.

[0023] The feature marking module 200 is the direct successor processing link of the signal physical separation module 100, and its core function is to extract key points with physical representation significance from the original waveforms of the two classified signals, and give strict class identification and time reference. The specific implementation process is as follows: Receive two parallel data streams output from the signal physical separation module 100, i.e. the original waveform of the electromagnetic wave speed class signal and the original waveform of the mechanical wave speed class signal; for each class of signal, first perform feature point extraction with a unified time reference; this process relies on the global clock system established by a high-precision time-unified signal source (such as GPS timing or crystal oscillator synchronization) to ensure that the waveforms collected by different sensors have strict time synchronization through a trigger alignment mechanism; the process of the trigger alignment mechanism is as follows: The synchronization pulse generated by a high-precision time system signal source is used to control the synchronization start of all sensor acquisition channels in a hardware triggering manner; in view of the transmission delay, the timestamp of the pulse starting point in the feature marking module 200 is used to back-propagate the relative delay amount of each channel, and the timestamp is linearly compensated and calibrated in the data preprocessing stage.

[0024] The feature extraction engine adopts a method combining adaptive threshold detection and multi-scale morphological analysis to identify feature points from the time domain sequence of each original waveform, including the pulse starting point (corresponding to the discharge initiation time), the amplitude extreme point (representing the energy release intensity) and the oscillation decay inflection point (reflecting the propagation attenuation characteristic); each extracted feature point is assigned a timestamp accurate to the nanosecond level, which is derived from a clock reference source with a unified time reference, thereby realizing absolute time alignment across sensors and across signal types; After completing the feature point extraction, the generic identification labeling is performed; this step automatically inherits the classification conclusion of the original waveform by the signal physical separation module 100, and automatically labels the electromagnetic wave speed class identification for the feature points extracted from the electromagnetic wave speed class signal waveform; for the feature points extracted from the mechanical wave speed class signal waveform, the mechanical wave speed class identification is labeled; such identification is bound to the timestamp and amplitude feature vector of each feature point as metadata, forming a structured feature data set; in this process, the data flow pipeline architecture is used to maintain the processing timeliness, ensuring that the feature point set with a unified time reference and clear physical generic identification is transmitted in real time to the subsequent dual-path credibility fusion module 300 under the premise of preserving the complete details of the original waveform, providing standardized input for the spatial / structural propagation time difference logic verification.

[0025] Please refer to Figure 2 The dual-path credibility fusion module 300 is the core link for system criterion generation, which implements physical propagation consistency verification on the feature point set input from the previous stage through a parallel dual-channel verification mechanism, and generates a quantitative credibility evaluation parameter; the specific execution process is as follows: The verification unit 301 in the dual-path credibility fusion module 300 performs spatial propagation time difference logic verification on the feature points labeled with electromagnetic wave speed class identification, extracts the pulse starting point timestamp (corresponding to the discharge initiation time) in the feature points to calculate the multi-sensor absolute arrival time difference, verifies the consistency of the measured time difference and the model theoretical time difference, and if the consistency is satisfied, the electromagnetic wave verification result is set to the logical value 1, otherwise it is 0; the amplitude spatial distribution of the amplitude extreme point (representing the energy release intensity) is simultaneously used to verify the electromagnetic wave attenuation law, and the oscillation mode of the oscillation decay inflection point (reflecting the propagation attenuation characteristic) is used to supplement the verification of the propagation path characteristics; the electromagnetic credibility value is generated by integrating the verification results of the three-dimensional feature points, specifically including: Firstly, the credibility basis value is calculated by the time difference matching degree of the pulse starting point timestamp, and the smaller the time difference deviation is, the higher the basis value is; secondly, the amplitude space distribution of the energy release intensity is represented by the amplitude extreme point to verify whether it conforms to the electromagnetic wave square inverse ratio decay law to generate the decay compliance coefficient; finally, the frequency damping characteristics of the oscillation decay inflection point reflecting the propagation attenuation characteristics are analyzed to analyze the uniformity of the propagation path medium; the quantization results of the above three types of feature points are fused according to the preset weight, wherein the time difference matching contributes mainly to the basis value, and the attenuation gradient and the oscillation mode characteristics are respectively taken as the core supplement and auxiliary correction term, and finally an electromagnetic credibility value of 0 to 100% is generated; wherein the quantization results of the above three types of feature points are fused according to the preset weight, the time difference matching degree weight a (for example, the reference value is 0.6), the amplitude extreme point attenuation gradient weight β (for example, the reference value is 0.3), and the oscillation decay inflection point damping feature weight γ (for example, the reference value is 0.1) are set, which satisfies the normalization condition a+β+γ=1; the fusion formula is: The electromagnetic credibility value = a x time difference matching degree + β x attenuation compliance coefficient + γ x oscillation mode score, wherein the time difference matching degree is determined by the deviation degree of the measured time difference and the theoretical time difference, the attenuation compliance coefficient is obtained by logarithmic operation of the measured and theoretical attenuation ratio, and the oscillation mode score is calculated according to the chi-square test result of the frequency damping ratio and the standard value.

[0026] The verification unit 301 performs structural propagation time difference logical verification on the feature points marked with mechanical wave velocity type identification, calculates the structural propagation time difference through the feature point pulse starting point timestamp according to the equipment entity material acoustic model, verifies the consistency of the time difference with the predicted value in the equipment entity material acoustic model, and if the consistency is met, the mechanical wave verification result is set to logical value 1, otherwise 0; at the same time, the amplitude gradient distribution of the amplitude extreme point is extracted to verify the energy attenuation characteristics of the mechanical wave in the structure, and the damping oscillation characteristics of the oscillation decay inflection point (reflecting the structure boundary reflection effect) are analyzed to enhance the credibility of the propagation path; the mechanical credibility value is generated by fusing multiple features, specifically including: The measured time difference of the pulse starting point timestamp is converted into a basic credibility value according to the compliance with the theoretical value; the amplitude change gradient of the energy release intensity is represented by the amplitude extreme point to verify its exponential mechanical vibration attenuation characteristics in the equipment entity structure; further, the echo interval and amplitude attenuation rate reflecting the propagation attenuation characteristics of the oscillation decay inflection point are extracted to analyze the structure boundary reflection effect and material impedance characteristics; the verification data of the three types of feature points are given different weights according to the equipment structure characteristics, the time difference compliance rate constitutes the core reference, the attenuation characteristics and the boundary effect are respectively taken as the key supplement and special correction factor, and finally a mechanical credibility value of 0 to 100% is synthesized.

[0027] Please refer to Figure 3 , the verification unit 301 performs spatial propagation time difference logical verification and structural propagation time difference logical verification in a parallel manner, specifically including: First, for the electromagnetic wave speed type feature point, based on the device three-dimensional coordinate model and the electromagnetic wave theory propagation model, the multi-sensor measured time difference (obtained through the pulse starting point timestamp) is calculated and compared with the theoretical time difference for consistency; if the time difference deviation is within the tolerance, the electromagnetic wave verification result is set to logical value 1, otherwise 0; At the same time, the amplitude space distribution of the amplitude extreme point (verifying the electromagnetic wave attenuation law) and the frequency damping characteristics of the oscillation attenuation inflection point (analyzing the uniformity of the propagation path) are integrated to generate an electromagnetic credibility value of 0 to 100% (the basic value is derived from the time difference matching degree, and the attenuation gradient and oscillation mode are used as supplements); For the mechanical wave speed type feature point, the process uses the device entity material acoustic model to calculate the structure propagation measured time difference, verifies its consistency with the predicted value to generate a mechanical wave verification result (logical value 1 / 0), and combines the amplitude gradient distribution of the amplitude extreme point (verifying the mechanical wave exponential attenuation characteristic) and the boundary reflection effect of the oscillation attenuation inflection point (analyzing the structure impedance) to generate a mechanical credibility value (the time difference coincidence rate is the core, and the attenuation characteristics and boundary effect are correction factors); The whole process ensures the mutual verification of the nature of the double-path physical propagation, and only when both types of verification meet the preset law can the valid result be output, providing reliable input for the subsequent fusion unit 302.

[0028] The fusion unit 302 in the double-path credibility fusion module 300 performs a double-physical-path joint decision mechanism, specifically including: Only when the electromagnetic wave verification result and the mechanical wave verification result are both logical value 1, that is, the pulse starting point time difference verification of the double-path passes, and the energy attenuation characteristics of the amplitude extreme point and the propagation characteristics of the oscillation attenuation inflection point all meet the preset law, is marked as a credible discharge event; The two types of credibility values (electromagnetic credibility value and mechanical credibility value) are weighted and fused to generate an event confidence parameter. The event confidence parameter generation adopts a dynamic weighted fusion algorithm. Based on the device type, the electromagnetic credibility value and the mechanical credibility value are assigned a reference weight in the final parameter. For example, gas insulated equipment focuses on the weight of the electromagnetic wave path, while oil immersed equipment focuses on the mechanical wave path. A feature correction mechanism is introduced on the basis of the weight. The energy release intensity ratio represented by the amplitude extreme point is used to dynamically adjust the weight distribution of the double-path, and the propagation path complexity reflected by the oscillation attenuation inflection point is used to adjust the weight of the corresponding path in the opposite direction. After normalization, an event confidence parameter with a continuous scale of 0 to 100% is output. This parameter comprehensively represents the joint credibility level of the discharge event in the electromagnetic propagation space characteristic and mechanical propagation structure characteristic dimensions. Among them: The process of dynamically adjusting the weight distribution of the double-path is as follows: According to the amplitude extreme point amplitude value extracted by the characteristic marking module 200, the amplitude ratio of the electromagnetic path and the mechanical path is calculated, and the ratio is taken as the adjustment factor. According to the preset device type reference weight in the fusion unit 302 (for example, the gas insulated device emphasizes the electromagnetic path), the weight proportion of the high amplitude value path is amplified in proportion, wherein the high amplitude value path is determined by comparing the preset amplitude ratio threshold value, and the value greater than the preset amplitude ratio threshold value is taken as the high amplitude value path; The process of reverse compensation adjustment is specifically: Extract the oscillation decay inflection point parameters (damping oscillation period number, amplitude decay rate) in the characteristic marking module 200, calculate the ratio of the damping oscillation period number and the amplitude decay rate, and take the ratio as the path complexity factor; According to the preset complexity and weight mapping relationship in the device entity material acoustic model, the weight of the high complexity path is reduced and compensated, wherein the high complexity path is determined by comparing the preset complexity ratio threshold value, and the value greater than the preset complexity ratio threshold value is taken as the high complexity path.

[0029] Please refer to Figure 4 The fusion unit 302 executes a double-physical-path joint decision mechanism, specifically including: First, the electromagnetic wave check result and the mechanical wave check result output by the verification unit 301 both need to be logical value 1 (indicating that the space propagation and the structure propagation both conform to the physical law), at this time, it is marked as a credible discharge event (only locking the real discharge nature); Subsequently, the event confidence parameter is generated, the baseline weight of the electromagnetic credible value and the mechanical credible value is allocated based on the preset basis weight of the device type (for example, the gas insulated device emphasizes the electromagnetic path weight, and the oil immersed device emphasizes the mechanical path weight); Introducing a dynamic feature correction mechanism, using the energy release intensity ratio represented by the amplitude extreme point (for example, the weight of the strong energy release path is increased) to dynamically adjust the double-path weight distribution, and combining the propagation path complexity reflected by the oscillation decay inflection point (for example, the influence of sound wave attenuation in oil immersed medium) to perform reverse compensation adjustment (the weight of the high complexity path is reduced to suppress distortion interference); Finally, through normalization processing, the event confidence parameter of 0 to 100% continuous scale is output, which quantifies the joint credibility of the electromagnetic and mechanical double paths, and provides a dynamic threshold adjustment basis for the anti-interference response module 400, solving the misjudgment problem caused by local interference of a single path.

[0030] The anti-interference response module 400 dynamically adjusts the response threshold based on the event confidence parameter, and only triggers an alarm response for a credible discharge event, specifically including: The three-level static benchmark interval of the preset event confidence parameter includes a high confidence interval (for example, preset to 80% to 100%), a medium confidence interval (50% to 80%), and a low confidence interval (0% to 50%). When the event confidence parameter falls into the high confidence interval, the response threshold is adjusted according to the high confidence coefficient. Dynamically adjust down (where ), used to enhance the alarm sensitivity of high-confidence events; when the parameter falls into the medium confidence interval, the response threshold is adjusted according to the neutral proportional coefficient Maintaining baseline levels (where ); When the parameter falls into the low confidence interval, the response threshold is proportional to the low coefficient Dynamically adjust (where ), used to suppress the false alarm risk of low-confidence signals; Based on dynamic control of the response threshold, the alarm response process is activated only when the input event simultaneously satisfies the requirement for a credible discharge event to be marked as true and its event confidence parameter exceeds the current response threshold. During the response process, multiple alarm levels are assigned based on the event confidence parameter values ​​(e.g., high-confidence emergency alarm, medium-confidence warning, etc.), and an alarm escalation decision is generated based on historical confidence trend analysis (e.g., three consecutive detections of credible events with rising confidence parameters). All response actions are transmitted to the monitoring system via standard industrial protocols, and the complete trajectory of event confidence parameter changes and response time series are automatically recorded, providing a data foundation for subsequent diagnosis. The entire process operates under real-time closed-loop control, effectively suppressing false alarms while ensuring the immediate capture of real insulation defects, ultimately achieving credible perception and risk warning of the partial discharge status of power equipment. The specific process for generating the alarm escalation decision is as follows: Within the sliding time window, the event confidence parameter sequence of credible discharge events is statistically analyzed. If the confidence parameters of consecutive events conform to the monotonically increasing law, or the proportion of high-confidence events in the window (confidence parameters in the high confidence interval) exceeds the preset proportion, the alarm level is automatically raised.

[0031] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion is characterized by: It comprises a signal physical separation module (100), a feature marking module (200), a dual-path credibility fusion module (300) and an anti-interference response module (400), wherein: The signal physical separation module (100) obtains the original waveform of the mixed transient signal flow, and classifies the mixed transient signal flow into electromagnetic wave velocity type signals and mechanical wave velocity type signals using a decision tree algorithm; The feature marking module (200) is configured to extract feature points with a unified time reference from the original waveforms of the electromagnetic wave velocity signal and the mechanical wave velocity signal, and mark each feature point with an identifier of the electromagnetic wave velocity class and the mechanical wave velocity class; The dual-path credibility fusion module (300) is configured to perform a spatial propagation time difference logic check on feature points labeled with electromagnetic wave velocity, and generate an electromagnetic wave verification result and an electromagnetic credibility value; perform a structural propagation time difference logic check on feature points labeled with mechanical wave velocity, and generate a mechanical wave verification result and a mechanical credibility value; only when both types of verification results simultaneously meet a preset law are they marked as a credible discharge event, and the two types of credibility values ​​are weightedly fused to generate an event confidence parameter; The anti-interference response module (400) is configured to dynamically adjust a response threshold based on an event confidence parameter, and trigger an alarm response only for credible discharge events.

2. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 1 is characterized in that: The classification process of the signal physical separation module (100) includes: The classification engine is used to extract key characteristic attributes of mixed transient signal streams, including the initial rising edge time of the signal, the main frequency band concentration area, and the relative arrival time difference between multiple sensors; Based on the key feature attributes, a multi-level judgment node is established through the decision tree algorithm. The three-level physical feature verification is carried out in sequence using the initial rising edge time of the signal, the main frequency band concentration area, and the relative arrival time difference between multiple sensors. When the three-dimensional features meet the physical laws of the same type of wave, the classification is completed.

3. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 2 is characterized in that: The decision tree algorithm's determination process specifically includes: The first-level node is based on the comparison between the initial rising edge time of the signal and the preset time threshold. If the initial rising edge time of the signal is lower than the preset time threshold, it enters the electromagnetic wave velocity branch; otherwise, it enters the mechanical wave velocity branch. The second-level node is based on the energy distribution ratio of the signal energy in the main frequency band concentration area within the preset frequency band threshold. If the signal energy exceeds the preset frequency band threshold, it is regarded as a candidate signal of the electromagnetic wave velocity type; otherwise, it is regarded as a candidate signal of the enhanced mechanical wave velocity type. The third-level node verifies the physical consistency of the conclusions of the first two levels based on the deviation tolerance of the relative arrival time difference between multiple sensors and the theoretical wave speed propagation model.

4. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 1 is characterized in that: The feature marking module (200) establishes a unified time base through a high-precision time signal source, extracts three types of feature points, namely pulse starting point, amplitude extreme point and oscillation attenuation inflection point, by using a method combining adaptive threshold detection and multi-scale morphological analysis, and assigns a timestamp to each type of feature point.

5. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 1 is characterized in that: The dual-path credibility fusion module (300) comprises a verification unit (301), and the verification unit (301) is used for logical verification of spatial propagation time difference, specifically comprising: Based on the three-dimensional coordinate model of the equipment and the theoretical electromagnetic wave propagation model, the multi-sensor measured time difference is calculated by the pulse starting point timestamp, and the consistency of this time difference with the theoretical time difference in the electromagnetic wave theoretical propagation model is verified to generate an electromagnetic wave verification result. If the consistency is met, the electromagnetic wave verification result is set to a logical value of 1, otherwise it is 0; the electromagnetic credibility value is generated by comprehensively considering the amplitude spatial distribution of the amplitude extreme point and the frequency damping characteristics of the oscillation attenuation inflection point.

6. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 5 is characterized in that: The process of the verification unit (301) performing the structure propagation time difference logic verification specifically includes: Through the acoustic model of the device's physical material, the pulse starting point timestamp is used to calculate the measured time difference of structural propagation, and the consistency of this time difference with the predicted value in the acoustic model of the device's physical material is verified to generate a mechanical wave verification result. If the consistency is met, the mechanical wave verification result is set to a logical value of 1, otherwise it is 0; and the mechanical credibility value is generated by combining the amplitude gradient distribution of the amplitude extreme point and the boundary reflection effect of the oscillation attenuation inflection point.

7. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 1 is characterized in that: The dual-path credibility fusion module (300) comprises a fusion unit (302), and when the electromagnetic wave verification result obtained by the fusion unit (302) is a logic value 1 and the mechanical wave verification result is a logic value 1, the current signal is marked as a credible discharge event.

8. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 7 is characterized in that: The process of generating event confidence parameters by the fusion unit (302) specifically includes: The benchmark weights of electromagnetic credibility value and mechanical credibility value in the final parameters are allocated based on the preset basic weights of the equipment type; a feature correction mechanism is introduced on the benchmark weights, and the energy release intensity ratio represented by the amplitude extreme point is used to dynamically adjust the dual-path weight distribution. At the same time, the corresponding path weights are reversely compensated and adjusted according to the complexity of the propagation path reflected by the oscillation attenuation inflection point, and the continuously scaled event confidence parameters are output after normalization processing.

9. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 1, characterized in that: The process of triggering an alarm response by the anti-interference response module (400) specifically includes: Matching event confidence parameters to the preset three-level static reference interval; Dynamically adjust the response threshold proportional coefficient according to the confidence interval and obtain the adjusted response threshold; The alarm response is triggered only when the credible discharge event flag is true and the event confidence parameter exceeds the adjusted response threshold.

10. The real-time monitoring and early warning system for partial discharge based on multi-source data fusion according to claim 9, characterized in that: The execution process of the alarm response specifically includes: Divide the alarm level into multiple levels according to the event confidence parameter value; Combine historical confidence trend analysis to generate early warning upgrade decisions; Record the event confidence parameter change trajectory and response time series.

Images