Fusion type partial discharge monitoring method based on optics

Abstract

The invention discloses a fusion type partial discharge monitoring method based on optics, and the method comprises the steps: S1, enabling a GaN ultraviolet sensor and a TEV sensor to work synchronously, and enabling a system to immediately capture the complete waveforms of the two sensors as long as the signal of any sensor exceeds a threshold value; s2, accurately extracting key parameters from each captured waveform, and combining the key parameters into a comprehensive feature vector; s3, associating the extracted parameters with the discharge intensity and the development speed of partial discharge, estimating the severity of the discharge through a model, tracking the variation trend of the parameters along with time, judging which stage the partial discharge is in, and predicting the future development of the partial discharge; s4, adopting a D-S evidence theory to carry out mathematical fusion on two independent diagnosis results, and only when evidences of the two sensors support each other, confirming the fault as a high-confidence partial discharge fault; the method has the advantages of high-sensitivity capture and early warning, greatly reduced false alarm rate and predictable evaluation.

Description

Technical Field

[0001] This invention belongs to the field of partial discharge monitoring technology for power distribution switchgear, and specifically relates to an optical-based fusion partial discharge monitoring method. Background Technology

[0002] Traditional maintenance methods for power distribution switchgear mainly rely on periodic manual inspections and preventative tests, which can easily miss subtle early signs of faults. Furthermore, preventative tests are typically conducted when the equipment is de-energized, failing to accurately reflect the insulation condition under operating conditions. Partial discharge monitoring is a technology used to detect potential partial discharge phenomena within the equipment of power distribution switchgear in power systems. Its main purpose is to monitor the partial discharge status of the switchgear in real time to promptly identify potential insulation defects and fault hazards.

[0003] Currently, the partial discharge monitoring of power distribution switchgear has the following problems:

[0004] 1. Traditional partial discharge monitoring methods may not be able to effectively detect weak partial discharge signals, especially in the complex electromagnetic environment of power distribution switchgear;

[0005] 2. The partial discharge process is divided into three stages: pre-discharge, mid-discharge, and post-discharge. The sensor response characteristics are different in each stage. Existing technologies cannot accurately extract these characteristics (such as waveform changes, frequency components, etc.) and distinguish between partial discharge signals and normal operating signals.

[0006] 3. Existing technologies cannot effectively correlate parameters such as amplitude, response time, and pulse count extracted from the sensor output waveform with the partial discharge process (such as discharge intensity, propagation speed, and repetition frequency);

[0007] 4. When using multiple sensors (ultraviolet and transient voltage to ground), the fusion and integration of multi-source data is not ideal, resulting in low accuracy of partial discharge identification and location.

[0008] To address the aforementioned issues, it is essential to develop an optical-based fusion method for partial discharge monitoring. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide an optically based fusion partial discharge monitoring method that can achieve high sensitivity, early warning, significantly reduce false alarm rate, and predictability assessment. It can achieve the goals of high sensitivity, high accuracy and strong anti-interference, and improve the safe operation level of power distribution switchgear.

[0010] The objective of this invention is achieved as follows: an optically based fusion-type partial discharge monitoring method, comprising the following steps:

[0011] S1, Synchronous Acquisition and Intelligent Trigger: The gallium nitride (GaN) ultraviolet sensor and transient voltage to ground (TEV) sensor are set to work synchronously, and a dynamic threshold is set for each sensor. As long as the signal of any sensor exceeds the threshold, the system immediately captures the complete waveforms of both sensors.

[0012] S2, Multi-dimensional Feature Extraction: Accurately extract key parameters from each captured waveform, including amplitude, rise time, pulse width, oscillation frequency, and number of pulses, and combine them into a comprehensive feature vector to provide a data foundation for subsequent analysis;

[0013] S3, Correlation Analysis and Trend Prediction: The extracted parameters are correlated with the discharge intensity and development speed of partial discharge. The severity of the discharge is estimated through the model. At the same time, the trend of parameter changes over time is tracked to determine whether the partial discharge is in the early, middle or late stage, and its future development is predicted.

[0014] S4, Evidence Fusion and Comprehensive Diagnosis: Using the DS evidence theory, the diagnostic results of the GaN ultraviolet sensor and the TEV sensor are mathematically fused. Only when the evidence from the two sensors supports each other is it confirmed as a high-confidence partial discharge fault.

[0015] Preferably, step S1 includes the following steps:

[0016] S11, Sensor Deployment and Initialization: The GaN ultraviolet sensor is aligned with the area to be measured inside the switch cabinet, and the TEV sensor is attached to the grounded metal panel outside the switch cabinet. The central processing unit controls the two sensors to work synchronously.

[0017] S12, Background Noise Learning and Dynamic Threshold Setting: Under conditions of no partial discharge activity, background signals are collected for a period of time, and the mean background noise of the two sensor signal channels is calculated respectively. and standard deviation Set the dynamic trigger threshold for each channel:

[0018] ;

[0019] ;

[0020] In the formula, k is the sensitivity coefficient;

[0021] S13, or logic trigger and data buffer: continuously monitors the signals of two channels. When the signal amplitude of either channel exceeds its own dynamic threshold, it is immediately determined as a potential partial discharge event and triggers synchronous acquisition.

[0022] Preferably, in step S13, after the synchronous acquisition is triggered, the timestamp of the event is recorded, and the complete waveform data of the two sensors for a period of time before and after the trigger time is cached.

[0023] Preferably, step S2 includes the following steps:

[0024] S21, Waveform parameterization: For each trigger event, extract the following parameters from the buffered waveforms of the two sensors respectively:

[0025] Pulse amplitude: (Peak voltage, unit: V or dBmV);

[0026] Ascent time: (Time required to rise from 10% to 90% of the peak value, in ns).

[0027] Pulse width: (Pulse full width at half maximum duration, in ns);

[0028] Oscillation frequency: Perform FFT analysis on TEV signals with obvious oscillation characteristics to extract their dominant oscillation frequency. (Unit: MHz);

[0029] S22, Power frequency phase correlation and pulse counting: Synchronize the timestamp of each event with the power frequency phase of the power grid, and count the total number of triggered events N within the time window of each power frequency cycle;

[0030] S23, Constructing a comprehensive feature vector: Combine all the above parameters into a multi-dimensional feature vector. This is used to comprehensively describe the event:

[0031] .

[0032] Preferably, step S3 includes the following steps:

[0033] S31, Discharge Intensity and Activity Assessment:

[0034] S311, Relative Discharge Quantity Assessment: Using TEV Signal Amplitude Correlation with discharge quantity, calculate relative discharge quantity : In the formula, K is the system calibration coefficient;

[0035] S312, Discharge Energy / Intensity Assessment: Estimate the energy characteristics of the discharge by combining the amplitude and frequency of the ultraviolet signal. : In the formula These are coefficients calibrated experimentally.

[0036] S32, Development Stage Identification and Trend Prediction:

[0037] S321, Stage Identification: Based on Feature Vectors Based on its position in the preset clustering model, determine whether the current event is in the early, middle, or late stage;

[0038] S322, Trend Forecasting: Tracking trends on an hourly or daily basis. The changes in (average amplitude) and N (pulse rate) are modeled using an exponential model. Perform fitting and calculate the growth coefficient. , The sustained positive value and increase of [amount] are important indicators of risk escalation.

[0039] Preferably, in step S321, the stage judgment principle is as follows:

[0040] Previous stage: low Low N, random phase distribution;

[0041] Middle section: Medium As N increases, phase focusing occurs;

[0042] Later stage: High High and dense N-type electrons result in significant phase focusing, which may increase the pulse width.

[0043] Preferably, step S4 includes the following steps:

[0044] S41, Independent Evidence Generation: Transforming Feature Vectors Decomposed into UV-related features Features related to TEV ,Will and Two pre-trained classifiers are input respectively, and each classifier outputs a base probability assignment (BPA), which is the confidence level that the event belongs to different propositions. Let... For GaN ultraviolet sensors, the proposition The reliability, let For TEV sensor to the proposition Reliability;

[0045] S42, Evidence Combination:

[0046] S421, using the DS evidence theory synthesis rules, fuses the credibility of two independent evidence sources to obtain a joint basic probability assignment. ;

[0047] S422, for any proposition A, its fused reliability is:

[0048] ;

[0049] In the formula, the conflict coefficient K is calculated as follows: A high K value indicates a significant discrepancy in the judgments of the two sensors, which may be due to interference.

[0050] S43, Final Decision and Alarm:

[0051] S431, Select the reliability after fusion The highest proposition serves as the final diagnostic result;

[0052] S432, anti-interference logic: Only when both sensors capture relevant signals within an extreme time period, and the fused diagnostic result is not "normal", will this event be identified as a sequential high-confidence partial discharge event, and the corresponding level of early warning or alarm will be activated accordingly.

[0053] Preferably, in step S41, the classifier is a rule-based expert system or a lightweight neural network.

[0054] Preferably, in step S41, the propositions include normal, corona discharge, surface discharge, and internal discharge.

[0055] Due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0056] (1) This invention uses GaN ultraviolet and TEV sensors for synchronous monitoring and sets an adaptive dynamic threshold. Recording is triggered when either sensor exceeds the threshold, thereby significantly improving the detection limit, discovering potential hazards in the early stages of partial discharge, and fundamentally avoiding missed detection.

[0057] (2) This invention mathematically fuses the independent diagnostic results of two sensors and only confirms the fault when the evidence is consistent, thereby effectively suppressing false alarms caused by switching operation, ambient light, etc., and improving the reliability and credibility of the system.

[0058] (3) This invention constructs an identification and prediction model by integrating characteristic parameters such as amplitude, frequency, and phase, thereby upgrading from a simple "alarm" to a precise "diagnosis" and "trend prediction", providing a basis for decision-making for predictive maintenance;

[0059] In summary, this invention has the advantages of high sensitivity in detection and early warning, significantly reduced false alarm rate, and predictability assessment. It can achieve the goals of high sensitivity, high accuracy, and strong anti-interference, and improve the safe operation level of power distribution switchgear. Attached Figure Description

[0060] Figure 1 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0061] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0062] like Figure 1 As shown, this invention provides an optically based fusion-type partial discharge monitoring method, comprising the following steps:

[0063] S1, Synchronous Acquisition and Intelligent Triggering: The GaN UV sensor and TEV sensor are set to work synchronously, and a dynamic threshold is set for each sensor. As long as the signal of either sensor exceeds the threshold, the system immediately captures the complete waveform of both sensors, ensuring that even weak signals can be recorded. The purpose is to achieve high-sensitivity capture and avoid missed detections.

[0064] S11, Sensor Deployment and Initialization: Point the GaN UV sensor at the area to be measured inside the switchgear (such as insulators or busbar connections) to ensure that the optical path is unobstructed; attach the TEV sensor to the grounded metal panel outside the switchgear, power on the system, and the two sensors start data acquisition synchronously. The central processing unit ensures that the data stream has a unified high-precision timestamp.

[0065] S12, Background Noise Learning and Dynamic Threshold Setting: Under conditions of no partial discharge activity, acquire background signals for a period of time (e.g., 60 seconds), and calculate the average background noise of the two sensor signal channels respectively. and standard deviation Set the dynamic trigger threshold for each channel:

[0066] ;

[0067] ;

[0068] In the formula, k is the sensitivity coefficient, which is usually taken as 3 to 5.

[0069] S13, or logic trigger and data buffer: continuously monitor the signals of two channels. When the signal amplitude of either channel exceeds its own dynamic threshold, it is immediately determined as a potential partial discharge event and triggers synchronous acquisition. After synchronous acquisition is triggered, the timestamp of the event is recorded and the complete waveform data of the two sensors for a period of time before and after the trigger time (e.g., 5us before the start and 50us after the trigger) is buffered.

[0070] S2, Multi-dimensional Feature Extraction: Accurately extract key parameters such as amplitude, rise time, pulse width, oscillation frequency, and number of pulses from each captured waveform, and combine them into a comprehensive feature vector to provide a data foundation for subsequent analysis. The purpose is to transform the original waveform into analyzable feature data.

[0071] S21, Waveform parameterization: For each trigger event, extract the following parameters from the buffered waveforms of the two sensors respectively:

[0072] Pulse amplitude: (Peak voltage, unit: V or dBmV);

[0073] Ascent time: (Time required to rise from 10% to 90% of the peak value, in ns).

[0074] Pulse width: (Pulse full width at half maximum duration, in ns);

[0075] Oscillation frequency: Perform FFT analysis on TEV signals with obvious oscillation characteristics to extract their dominant oscillation frequency. (Unit: MHz)

[0076] S22, Power Frequency Phase Correlation and Pulse Counting: Synchronize the timestamp of each event with the power frequency phase of the power grid, and count the total number of triggered events N within the time window of each power frequency cycle (e.g., 20ms corresponding to 50Hz).

[0077] S23, Constructing a comprehensive feature vector: Combine all the above parameters into a multi-dimensional feature vector. This is used to comprehensively describe the event:

[0078] .

[0079] S3, Correlation Analysis and Trend Prediction: The extracted parameters are correlated with the discharge intensity and development speed of partial discharge. The severity of the discharge is estimated through the model. At the same time, the trend of parameter changes over time is tracked to determine whether the partial discharge is in the early, middle or late stage, and its future development is predicted. The purpose is to understand the physical meaning and development trend behind the characteristic parameters.

[0080] S31, Discharge Intensity and Activity Assessment:

[0081] S311, Relative Discharge Quantity Assessment: Using TEV Signal Amplitude Correlation with discharge quantity, calculate relative discharge quantity : In the formula, K is the system calibration coefficient;

[0082] S312, Discharge Energy / Intensity Assessment: Estimate the energy characteristics of the discharge by combining the amplitude and frequency of the ultraviolet signal. : In the formula These are the coefficients calibrated experimentally.

[0083] S32, Development Stage Identification and Trend Prediction:

[0084] S321, Stage Identification: Based on Feature Vectors Based on the position in the preset clustering model, determine whether the current event is in the pre-, middle, or late stage. The stage determination principle is as follows:

[0085] Previous stage: low Low N, random phase distribution;

[0086] Middle section: Medium As N increases, phase focusing occurs;

[0087] Later stage: High High and dense N-type electrons result in significant phase focusing, which may increase the pulse width.

[0088] S322, Trend Forecasting: Tracking trends on an hourly or daily basis. The changes in (average amplitude) and N (pulse rate) are modeled using an exponential model. Perform fitting and calculate the growth coefficient. , The sustained positive value and increase of [amount] are important indicators of risk escalation.

[0089] S4, Evidence Fusion and Comprehensive Diagnosis: Using the DS evidence theory, the diagnostic results (evidence) of the GaN ultraviolet sensor and the TEV sensor are mathematically fused. Only when the evidence from the two sensors supports each other is it confirmed as a high-confidence partial discharge fault, thereby effectively eliminating false alarms from a single sensor. The goal is to make a final judgment with strong anti-interference ability and high accuracy.

[0090] S41, Independent Evidence Generation:

[0091] eigenvectors Decomposed into UV-related features Features related to TEV ,Will and Two pre-trained classifiers (such as rule-based expert systems or lightweight neural networks) are input respectively. Each classifier outputs a basic probability assignment (BPA), which is the confidence level that the event belongs to different propositions (such as normal, corona discharge, surface discharge, internal discharge). Let... For GaN ultraviolet sensors, the proposition The reliability, let For TEV sensor to the proposition The reliability.

[0092] S42, Evidence Combination:

[0093] S421, using the DS evidence theory synthesis rules, fuses the credibility of two independent evidence sources to obtain a joint basic probability assignment. ;

[0094] S422, for any proposition A, its fused reliability is:

[0095] ;

[0096] In the formula, the conflict coefficient K is calculated as follows: A high K value indicates a significant discrepancy in the judgments of the two sensors, which may be due to interference.

[0097] S43, Final Decision and Alarm:

[0098] S431, Select the reliability after fusion The highest proposition serves as the final diagnostic result;

[0099] S432, Anti-interference logic: Only when both sensors capture relevant signals within an extreme time (e.g., 1µs) and the fused diagnostic result is not "normal", will this event be identified as a sequential high-confidence partial discharge event, and the corresponding level of early warning or alarm will be activated accordingly.

[0100] The workflow of this invention is as follows:

[0101] Step 1: The process begins with the synchronous acquisition of signals from two sensors. The system continuously compares the signals with dynamic thresholds. Once either sensor is triggered, the complete waveform data of both channels is buffered simultaneously. This is the basis for high-sensitivity detection.

[0102] The second step is to extract a series of key feature parameters such as amplitude, time, and frequency from the cached raw waveform data using an algorithm, and combine them into a comprehensive feature vector, thus transforming the raw signal into analyzable information.

[0103] The third step is to use eigenvectors to assess the intensity and activity of the discharge, identify the development stage (early, middle, and late) of the partial discharge, and predict its development trend based on historical data, thus achieving a leap from "current situation perception" to "risk warning".

[0104] Step 4: This is the core of achieving high accuracy and strong anti-interference capability. The system forms a preliminary diagnosis (evidence) based on ultraviolet and TEV data respectively, and then uses DS evidence theory to fuse them. Only when the two pieces of evidence support each other and have high confidence can they be finally confirmed as partial discharge and an alarm be triggered, thereby effectively eliminating false alarms from a single sensor.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. An optically based fusion-type partial discharge monitoring method, characterized in that, Includes the following steps: S1, Synchronous Acquisition and Intelligent Trigger: The GaN UV sensor and TEV sensor are set to work synchronously, and a dynamic threshold is set for each sensor. As long as the signal of any sensor exceeds the threshold, the system will immediately capture the complete waveform of both sensors. S2, Multi-dimensional Feature Extraction: Accurately extract key parameters from each captured waveform, including amplitude, rise time, pulse width, oscillation frequency, and number of pulses, and combine them into a comprehensive feature vector to provide a data foundation for subsequent analysis; S3, Correlation Analysis and Trend Prediction: The extracted parameters are correlated with the discharge intensity and development speed of partial discharge. The severity of the discharge is estimated through the model. At the same time, the trend of parameter changes over time is tracked to determine whether the partial discharge is in the early, middle or late stage, and its future development is predicted. S4, Evidence Fusion and Comprehensive Diagnosis: Using the DS evidence theory, the diagnostic results of the GaN ultraviolet sensor and the TEV sensor are mathematically fused. Only when the evidence from the two sensors supports each other is it confirmed as a high-confidence partial discharge fault.

2. The optical-based fusion partial discharge monitoring method according to claim 1, characterized in that, Step S1 includes the following steps: S11, Sensor Deployment and Initialization: The GaN ultraviolet sensor is aligned with the area to be measured inside the switch cabinet, and the TEV sensor is attached to the grounded metal panel outside the switch cabinet. The central processing unit controls the two sensors to work synchronously. S12, Background Noise Learning and Dynamic Threshold Setting: Under conditions of no partial discharge activity, background signals are collected for a period of time, and the mean background noise of the two sensor signal channels is calculated respectively. and standard deviation Set the dynamic trigger threshold for each channel: ； ； In the formula, k is the sensitivity coefficient; S13, or logic trigger and data buffer: continuously monitors the signals of two channels. When the signal amplitude of either channel exceeds its own dynamic threshold, it is immediately determined as a potential partial discharge event and triggers synchronous acquisition.

3. The optical-based fusion partial discharge monitoring method according to claim 2, characterized in that: In step S13, after synchronous acquisition is triggered, the timestamp of the event is recorded, and the complete waveform data of the two sensors for a period of time before and after the trigger time is cached.

4. The optical-based fusion partial discharge monitoring method according to claim 1, characterized in that, Step S2 includes the following steps: S21, Waveform parameterization: For each trigger event, extract the following parameters from the buffered waveforms of the two sensors respectively: Pulse amplitude: (Peak voltage, unit: V or dBmV); Ascent time: (Time required to rise from 10% to 90% of the peak value, in ns). Pulse width: (Pulse full width at half maximum duration, in ns); Oscillation frequency: Perform FFT analysis on TEV signals with obvious oscillation characteristics to extract their dominant oscillation frequency. (Unit: MHz); S22, Power frequency phase correlation and pulse counting: Synchronize the timestamp of each event with the power frequency phase of the power grid, and count the total number of triggered events N within the time window of each power frequency cycle; S23, Constructing a comprehensive feature vector: Combine all the above parameters into a multi-dimensional feature vector. This is used to comprehensively describe the event: 。 5. The optical-based fusion partial discharge monitoring method according to claim 1, characterized in that, Step S3 includes the following steps: S31, Discharge Intensity and Activity Assessment: S311, Relative Discharge Quantity Assessment: Using TEV Signal Amplitude Correlation with discharge quantity, calculate relative discharge quantity : In the formula, K is the system calibration coefficient; S312, Discharge Energy / Intensity Assessment: Estimate the energy characteristics of the discharge by combining the amplitude and frequency of the ultraviolet signal. : In the formula These are coefficients calibrated experimentally. S32, Development Stage Identification and Trend Prediction: S321, Stage Identification: Based on Feature Vectors Based on its position in the preset clustering model, determine whether the current event is in the early, middle, or late stage; S322, Trend Forecasting: Tracking trends on an hourly or daily basis. The changes in (average amplitude) and N (pulse rate) are modeled using an exponential model. Perform fitting and calculate the growth coefficient. , The sustained positive value and increase of [amount] are important indicators of risk escalation.

6. The optical-based fusion partial discharge monitoring method according to claim 5, characterized in that, In step S321, the stage judgment principle is as follows: Previous stage: low Low N, random phase distribution; Middle section: Medium As N increases, phase focusing occurs; Later stage: High High and dense N-type electrons result in significant phase focusing, which may increase the pulse width.

7. The optical-based fusion partial discharge monitoring method according to claim 1, characterized in that, Step S4 includes the following steps: S41, Independent Evidence Generation: Transforming Feature Vectors Decomposed into UV-related features Features related to TEV ,Will and Two pre-trained classifiers are input respectively, and each classifier outputs a base probability assignment (BPA), which is the confidence level that the event belongs to different propositions. Let... For GaN ultraviolet sensors, the proposition The reliability, let For TEV sensor to the proposition Reliability; S42, Evidence Combination: S421, using the DS evidence theory synthesis rules, fuses the credibility of two independent evidence sources to obtain a joint basic probability assignment. ; S422, for any proposition A, its fused reliability is: ； In the formula, the conflict coefficient K is calculated as follows: A high K value indicates a significant discrepancy in the judgments of the two sensors, which may be due to interference. S43, Final Decision and Alarm: S431, Select the reliability after fusion The highest proposition serves as the final diagnostic result; S432, anti-interference logic: Only when both sensors capture relevant signals within an extreme time period, and the fused diagnostic result is not "normal", will this event be identified as a sequential high-confidence partial discharge event, and the corresponding level of early warning or alarm will be activated accordingly.

8. The optical-based fusion partial discharge monitoring method according to claim 1, characterized in that: In step S41, the classifier is a rule-based expert system or a lightweight neural network.

9. The optical-based fusion partial discharge monitoring method according to claim 1, characterized in that: In step S41, the propositions include normal, corona discharge, surface discharge, and internal discharge.

Images