Partial discharge fusion detection method based on multiple sensors

Abstract

The invention discloses a partial discharge fusion detection method based on multiple sensors, and relates to the field of power system detection, and the method comprises the steps: S1, multi-modal sensor cooperative deployment: firstly, after the multi-modal sensor cooperative deployment in S1 is completed, S2, time-space synchronization data collection is carried out, and then S3, signal preprocessing and feature extraction are carried out after the time-space synchronization data collection in S2 is completed; and S3, after signal preprocessing and feature extraction are completed, performing S4, feature level fusion innovation, when the S4, feature level fusion innovation is completed, performing S5, a decision level fusion mechanism, and finally performing S6, three-dimensional visualization and early warning after the S5, the decision level fusion mechanism is completed. According to the partial discharge fusion detection method based on the multiple sensors, in the strong interference environment of the frequency converter, the false alarm rate is compressed from 68% to 11% through multi-sensor cooperative noise reduction and dynamic weight distribution, and the anti-interference capability of the system is remarkably improved; and centimeter-level positioning precision is realized, and powerful support is provided for accurate fault positioning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system detection, in particular to a partial discharge fusion detection method based on multiple sensors. BACKGROUND

[0002] The partial discharge fusion detection method based on multiple sensors is a method that integrates partial discharge (PD) signals from different types of sensors to improve detection accuracy, accuracy, and robustness. The core idea of this method is to use the complementary characteristics of multiple sources of information to analyze PD signals collected by different sensors, thereby more effectively identifying PD faults and reducing false positives.

[0003] In the prior art, during the operation of electrical equipment in the power grid system, PD and temperature are important parameters for indicating whether the primary equipment is operating normally. In particular, the connection parts of the primary equipment, due to equipment manufacturing reasons, environmental pollution of equipment, long-term operation of equipment, severe overload operation, contact oxidation, arc impact, partial discharge, and other reasons, the contact resistance increases, so that during power transmission, PD and heating phenomena often occur at the internal connections of the equipment. The PD and heating problem is a continuous accumulation process, and if not controlled, the degree of change will continue to intensify, and will have a great impact on the performance of the insulation and the service life of the equipment, thereby causing insulation damage, aging, and breakdown of the power equipment.

[0004] In order to overcome the above shortcomings, the prior art (publication number CN119738674A) discloses a partial discharge fusion detection method and system based on multiple sensors, which includes: a data sensing unit that collects generated transient voltage to ground signals, generated ultra-high frequency electromagnetic wave signals, and generated acoustic signals; a partial discharge collection unit that processes the received signals and selects whether to start a warning based on the results of the judgment; the partial discharge collection unit and the comprehensive online detection unit establish communication; the comprehensive online detection unit analyzes and processes and stores data, and performs centralized long-term detection of equipment in the power distribution network through human-computer interaction and mobile terminals, obtains multiple signals related to partial discharge based on multiple sensors, obtains the characteristics of partial discharge from different angles, and analyzes and determines in combination with the partial discharge collection unit, and interacts with the data of the comprehensive online detection unit, improves the accuracy of equipment detection in the power system, and improves the maintenance efficiency of the partial discharge phenomenon in the distribution network.

[0005] In order to overcome the above shortcomings, the prior art (publication number CN114113943B) discloses a transformer partial discharge detection system, method and equipment based on current and ultrasonic signals. The system realizes data fusion of current signals and ultrasonic signals collected by the collection module through a data analysis module, a model establishment module and a detection and recognition module. It does not need to rely on high-speed large data transmission and CPU operation in the rear, and directly realizes signal conditioning, data collection and multi-data source fusion on the edge of the transformer partial discharge detection system. It realizes a transformer partial discharge detection system with independent collection and edge computing, has high reliability and integration, and can effectively analyze the partial discharge fault state of the transformer. It can eliminate various adverse factors in the complex environment of dry-type transformer monitoring and improve the accuracy of identifying the partial discharge result of the transformer.

[0006] Although the prior art can overcome the above-mentioned shortcomings, there are still other problems in its operation process, such as: in the power grid system, the partial discharge fusion detection is easily disturbed by the external environment, the external disturbance easily affects the accuracy of detection, the detected result deviates, the anti-interference ability is poor, and it is not conducive to accurately capturing the required data. SUMMARY

[0007] The purpose of the present application is to provide a partial discharge fusion detection method based on multiple sensors to solve the problem of the above background technology that in the power grid system, the partial discharge fusion detection is easily disturbed by the external environment, the external disturbance easily affects the accuracy of detection, the detected result deviates, the anti-interference ability is poor, and it is not conducive to accurately capturing the required data.

[0008] To achieve the above purpose, the present application provides the following technical scheme: a partial discharge fusion detection method based on multiple sensors, comprising the following specific steps: S1: multi-modal sensor cooperative deployment, S2: time-space synchronous data collection, S3: signal preprocessing and feature extraction, S4: feature-level fusion innovation, S5: decision-level fusion mechanism and S6: three-dimensional visualization and early warning; Firstly, after S1: multi-modal sensor cooperative deployment is completed, S2: time-space synchronous data collection is performed, then when S2: time-space synchronous data collection is completed, S3: signal preprocessing and feature extraction is performed, and then after S3: signal preprocessing and feature extraction is completed, S4: feature-level fusion innovation is performed, when S4: feature-level fusion innovation is completed, S5: decision-level fusion mechanism is performed, and finally, after S5: decision-level fusion mechanism is completed, S6: three-dimensional visualization and early warning is performed.

[0009] Further, the S1: multi-modal sensor collaborative deployment installs ultra-high frequency sensors, ultrasonic sensors, high-frequency current transformers, and transient ground voltage sensors through collaborative deployment; innovates spatial topology calibration, uses a laser range finder to measure the three-dimensional coordinates of each sensor relative to the circuit breaker contact with millimeter-level precision, and constructs a detailed spatial position mapping table.

[0010] Further, the ultra-high frequency sensor in the S1: multi-modal sensor collaborative deployment is carefully installed inside the insulating observation window of the switch cabinet, ensuring unobstructed reception of electromagnetic waves, and adopting a spiral directional antenna design to optimize the reception of electromagnetic waves in the 300 MHz to 1.5 GHz frequency band, with an antenna gain not less than 5 decibels, greatly enhancing signal reception capability; The ultrasonic sensor is precisely attached to the stress concentration area of the cabinet outer wall, using a silicon grease layer with a thickness of only 0.2 millimeters to achieve acoustic impedance matching, ensuring efficient transmission of acoustic signals, and the resonant frequency is precisely controlled at 150 kilohertz to capture acoustic signals within a specific frequency range; The high-frequency current transformer is cleverly buckled on the cable grounding line, 30 centimeters away from the cabinet, and uses a Rogowski coil structure to achieve high-precision current measurement, with a transmission impedance of 100 millivolts / milliamperes at a frequency of 10 megahertz, accurately capturing current pulse signals; The transient ground voltage sensor is adsorbed at the hinge position of the metal cabinet door, with a built-in 1 to 100 megahertz bandpass filter, effectively filtering interference signals with a detection sensitivity of 1 millivolt / decibel microvolt, ensuring sensitive capture of ground voltage signals; According to the propagation characteristics of electromagnetic waves and sound waves in different propagation media, the signal delay is dynamically calculated to ensure the accuracy of data processing. The propagation speed of electromagnetic waves in the insulating gas in the cabinet is accurately calculated at 70% of the speed of light in a vacuum, and the propagation speed of sound waves in steel plates is real-time corrected according to the Young's modulus and density of the material, ensuring the accurate propagation time of sound wave signals.

[0011] Further, the S2: time-space synchronous data acquisition uses nanosecond-level synchronous control, the master control unit receives the second pulse signal of the GPS satellite through the optical fiber to ensure the accuracy of the time reference; an intelligent triggering strategy dynamically monitors background noise, calculates the noise median absolute deviation every 10 seconds, sets the triggering threshold to 3 times the standard deviation of the noise amplitude, and effectively avoids false triggering; A field programmable gate array is used to generate a synchronous triggering instruction to achieve precise synchronization of each acquisition channel. Each acquisition channel uses a high-precision clock chip with a clock jitter of less than 50 picoseconds, ensuring that the synchronization error of the entire system is controlled within ±1 nanosecond, achieving high-precision data acquisition. When any two sensors simultaneously exceed the threshold or a single sensor signal continuously exceeds the threshold for 200 milliseconds, the intelligent trigger triggers full-channel data acquisition, ensuring complete capture of important signals.

[0012] Further, the S3: signal preprocessing and feature extraction can adaptively reduce noise by wavelet, automatically select the most suitable wavelet basis for noise reduction processing according to the steepness of the signal waveform; pulse feature accurate extraction, using signal first-order difference zero-crossing point to detect pulse boundary, effectively exclude the interference of noise burr, ensure the accurate segmentation of pulse signal; For pulse sharp signal with kurtosis greater than 5, select db8 wavelet basis to effectively remove high-frequency noise, for smooth continuous signal, select sym6 wavelet basis to retain signal details, the noise reduction threshold is dynamically adjusted according to the signal entropy value, to ensure that the weak pulse characteristics are retained while reducing noise, and the signal quality is improved; A comprehensive feature library is constructed, including time domain features, frequency domain features and time-frequency features, to provide rich information basis for subsequent fault diagnosis.

[0013] Further, the S4: feature level fusion innovation dynamically allocates weights according to the signal-to-noise ratio and confidence of each sensor signal; The signal-to-noise ratio factor calculates the weight base value according to the ratio of signal amplitude to background noise, to ensure that the sensor with higher signal quality obtains higher weight, the confidence factor is based on the historical database to statistically analyze the detection accuracy of sensors under different working conditions, to provide reliable basis for weight allocation, the fusion rule is flexible and variable, the channel is automatically shielded when the sensor confidence is lower than 0.5, to effectively avoid the interference of false information, the signal-to-noise ratio and confidence are superimposed with activation coefficient to increase the correlation compensation term with typical discharge characteristics, to improve the recognition weight of similar discharge mode, and to ensure the accuracy of the diagnosis result.

[0014] Further, the S5: decision level fusion mechanism cooperates with multiple models for diagnosis, constructs a multiple model cooperative diagnosis system including SVM classifier, CNN network and HMM model, and fully utilizes the respective advantages for fault diagnosis; intelligent processing mechanism is automatically started when there is a major disagreement between the diagnosis results of different models.

[0015] Further, the SVM classifier in the S5: decision level fusion mechanism uses 20-dimensional time-frequency feature vector to classify discharge types, accurately identifies different types of discharge faults, CNN network processes wavelet time-frequency spectrum image to realize accurate prediction of discharge source three-dimensional coordinates, HMM model analyzes pulse interval time series to predict the probability of discharge development trend, and provides strong support for fault warning; Label the current data as conflict samples to provide valuable data support for subsequent model optimization, start the online learning module to update model parameters, improve the adaptability and accuracy of the model, reduce the confidence score of conflict sensors, and avoid the interference of false information on the diagnosis result.

[0016] Further, the S6: three-dimensional visualization and early warning is constructed by hyperbolic equation set through high-precision positioning technology, time difference positioning method, the space coordinates of various sensors and time delay difference; hierarchical early warning strategy, different early warning levels and response measures are set according to different ranges of discharge indexes.

[0017] Further, the particle swarm optimization algorithm is used for iterative solution of discharge source coordinates in the S6: three-dimensional visualization and early warning, so that the positioning accuracy of centimeter level is realized, and reliable basis is provided for fault positioning. Normal state is monitored regularly, the sampling rate is increased to 10 times per second in the attention state, the monitoring strength is strengthened, the circuit breaker interlocking protection is triggered in the dangerous state, the safety of the equipment is ensured, the DI calculation weight is automatically reduced in the high-noise working condition, the false alarm is avoided, and the stability and reliability of the early warning system are improved.

[0018] Compared with the prior art, the beneficial effects of the present application are: 1. In the strong interference environment of the frequency converter, the false alarm rate is compressed from 68% to 11% through multi-sensor cooperative noise reduction and dynamic weight distribution, and the anti-interference ability of the system is significantly improved. Further, through four-sensor time difference joint calculation and acoustic-electric wave speed real-time compensation algorithm, the positioning accuracy of centimeter level is realized, and strong support is provided for accurate fault positioning. Further, when the main sensor fails, the system can detect and adjust the weight distribution in real time, ensure that the detection rate remains at a high level, and realize self-healing type diagnosis closed loop.

[0019] 2. Different sensors have their own advantages for different types of discharge faults, through fusion processing, multi-source complementation is realized, and the accuracy and comprehensiveness of fault diagnosis are improved. BRIEF DESCRIPTION OF DRAWINGS

[0020] Fig. 1 It is the overall flowchart of the present application.

[0021] Fig. 2 It is the signal preprocessing and feature extraction flowchart of the present application.

[0022] Fig. 3 It is the decision-level fusion mechanism flowchart of the present application. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0024] Embodiment one: as shown in the technical solution, a local discharge fusion detection method based on multiple sensors, in order to solve the problem that local discharge is easily disturbed and the disturbance easily affects the detection accuracy, discloses the following specific steps: Figs. 1-3 S1: multi-modal sensor collaborative deployment, S2: time-space synchronous data acquisition, S3: signal preprocessing and feature extraction, S4: feature-level fusion innovation, S5: decision-level fusion mechanism, and S6: three-dimensional visualization and early warning; First, after S1: multi-modal sensor collaborative deployment is completed, S2: time-space synchronous data acquisition is performed, then when S2: time-space synchronous data acquisition is completed, S3: signal preprocessing and feature extraction is performed, and then after S3: signal preprocessing and feature extraction is completed, S4: feature-level fusion innovation is performed, when S4: feature-level fusion innovation is completed, S5: decision-level fusion mechanism is performed, and finally, after S5: decision-level fusion mechanism is completed, S6: three-dimensional visualization and early warning is performed.

[0025] S1: multi-modal sensor collaborative deployment installs ultra-high frequency sensors, ultrasonic sensors, high-frequency current transformers, and transient ground voltage sensors through collaborative deployment; the spatial topology is calibrated and innovated, a laser range finder is used to measure the three-dimensional coordinates of each sensor relative to the circuit breaker contact with millimeter-level accuracy, and a detailed spatial position mapping table is constructed.

[0026] The ultra-high frequency sensor in S1: multi-modal sensor collaborative deployment is carefully installed inside the insulation observation window of the switch cabinet, ensuring unobstructed reception of electromagnetic waves, and a spiral directional antenna design is adopted to optimize the reception of electromagnetic waves in the 300MHz to 1.5GHz frequency band, with an antenna gain not less than 5 decibels, greatly enhancing the signal reception capability; The ultrasonic sensor is precisely attached to the stress concentration area of the cabinet outer wall, and a silicon grease layer with a thickness of only 0.2 millimeters is used to achieve acoustic impedance matching, ensuring efficient transmission of acoustic signals, and the resonant frequency is accurately controlled at 150 kilohertz to capture acoustic signals within a specific frequency range; The high-frequency current transformer is cleverly buckled on the cable grounding line, 30 centimeters away from the cabinet, and uses a Rogowski coil structure to achieve high-precision current measurement, with a transmission impedance of 100 millivolts / milliamperes at a frequency of 10 megahertz, accurately capturing current pulse signals; The transient ground voltage sensor is adsorbed at the hinge position of the metal cabinet door, with a built-in 1 to 100 megahertz bandpass filter, effectively filtering interference signals with a detection sensitivity of 1 millivolt / decibel microvolt, ensuring sensitive capture of ground voltage signals; ​According to the propagation characteristics of electromagnetic waves and acoustic waves in different propagation media, the signal delay is dynamically calculated to ensure the accuracy of data processing. The propagation speed of electromagnetic waves in the insulating gas in the tank is accurately calculated as 70% of the speed of light in vacuum. The propagation speed of acoustic waves in steel plates is real-time corrected according to the Young's modulus and density of the material, ensuring the accurate propagation time of acoustic wave signals.

[0027] S2: Time-space synchronous data acquisition Through nanosecond-level synchronous control, the master control unit receives the second pulse signal of the GPS satellite through the optical fiber to ensure the accuracy of the time reference; intelligent triggering strategy, dynamically monitor the background noise, calculate the noise median absolute deviation every 10 seconds, set the trigger threshold to 3 times the standard deviation of the noise amplitude, effectively avoid false triggering; use field programmable gate array to generate synchronous trigger instructions to realize accurate synchronization of each acquisition channel; each acquisition channel uses a high-precision clock chip with a clock jitter of less than 50 picoseconds to ensure that the synchronization error of the whole system is controlled within ±1 nanosecond, realizing high-precision data acquisition; when any two sensors exceed the threshold at the same time, or a single sensor signal exceeds the threshold for 200 milliseconds, the intelligent trigger triggers the full-channel data acquisition, ensuring the complete capture of important signals.

[0028] S3: Signal preprocessing and feature extraction can adaptively reduce noise with wavelet, automatically select the most suitable wavelet basis for noise reduction processing according to the steepness of the signal waveform; accurate extraction of pulse characteristics, using signal first-order difference zero-crossing point detection to exclude noise burr interference and ensure accurate segmentation of pulse signals; for pulse sharp signals with kurtosis greater than 5, select db8 wavelet basis to effectively remove high-frequency noise, for smooth continuous signals, select sym6 wavelet basis to retain signal details, and dynamically adjust the noise reduction threshold according to the signal entropy value to ensure that weak pulse characteristics are retained while reducing noise, improving signal quality; build a comprehensive feature library including time domain features, frequency domain features and time-frequency features to provide a rich information base for subsequent fault diagnosis.

[0029] S4: Feature-level fusion innovation Through dynamic allocation of double-factor weights, the weight is dynamically allocated according to the signal-to-noise ratio and confidence of each sensor signal; signal-to-noise ratio factor: calculate the weight base value according to the ratio of signal amplitude to background noise to ensure that sensors with higher signal quality obtain higher weights; confidence factor: based on historical database statistics of sensor detection accuracy under different working conditions to provide reliable basis for weight allocation; flexible and variable fusion rules automatically shield the channel when the sensor confidence is less than 0.5 to effectively avoid the interference of false information; according to the signal-to-noise ratio and confidence, the activation coefficient is superimposed to increase the correlation compensation term with the typical discharge characteristics, improve the recognition weight of similar discharge mode, and ensure the accuracy of the diagnosis result.

[0030] S5: The decision-level fusion mechanism utilizes multi-model collaborative diagnosis to construct a multi-model collaborative diagnosis system, including an SVM classifier, a CNN network, and an HMM model, fully leveraging their respective advantages for fault diagnosis. Intelligent evidence conflict handling automatically activates when there are significant discrepancies in the diagnostic results of different models. The SVM classifier in the S5 decision-level fusion mechanism uses 20-dimensional time-frequency feature vectors to classify discharge types, accurately identifying different types of discharge faults. The CNN network processes wavelet time-spectrum images to accurately predict the three-dimensional coordinates of the discharge source. The HMM model analyzes pulse interval time series to predict the probability of discharge development trends, providing strong support for fault early warning. Labeling current data as conflict samples provides valuable data support for subsequent model optimization. An online learning module is activated to update model parameters, improving model adaptability and accuracy, reducing the confidence score of conflict sensors, and avoiding interference from erroneous information in the diagnostic results.

[0031] S6: 3D visualization and early warning utilizes high-precision positioning technology and time difference positioning method, combining the spatial coordinates and time delay difference of each sensor to construct a hyperboloid equation system; a graded early warning strategy sets different early warning levels and response measures according to different ranges of discharge index; S6: 3D visualization and early warning uses particle swarm optimization algorithm to iteratively calculate the coordinates of the discharge power source, achieving centimeter-level positioning accuracy and providing a reliable basis for fault location; Under normal conditions, routine monitoring is performed. Under alert conditions, the sampling rate is increased to 10 times per second to enhance monitoring efforts. Under dangerous conditions, the circuit breaker interlock protection is triggered to ensure equipment safety. Under high-noise conditions, the DI calculation weight is automatically reduced to avoid false alarms and improve the stability and reliability of the early warning system.

[0032] Multi-sensor weight dynamic allocation mechanism: The weight allocation is based on both signal-to-noise ratio and confidence level to ensure the accuracy and reliability of information fusion. At the same time, the automatic shielding mechanism for low-confidence sensors effectively avoids interference from erroneous information.

[0033] Decision-making conflict resolution mechanism: When there are conflicts in the diagnostic results of different models, the intelligent processing mechanism is automatically activated, including measures such as labeling conflicting samples, updating model parameters online, and reducing the confidence of conflicting sensors, to ensure the accuracy and consistency of diagnostic results.

[0034] Environmental adaptive early warning strategy: By dynamically adjusting the DI calculation weight and switching the directional monitoring mode according to the noise conditions, the stability and adaptability of the early warning system are improved, ensuring that early warning signals can be accurately issued in different environments.

[0035] Example 2: Figs. 1-3 The technical solution shown, based on Embodiment 1, also discloses: Fault setting: 3mm floating discharge defect is set on phase A busbar, superimposed with frequency converter interference (58dB 1-20MHz), simulating the complex situation in actual working environment.

[0036] Detection process: HFCT first detects the 42mV pulse train signal with a repetition frequency of 120Hz. Subsequently, the feature fusion module dynamically allocates weights, with HFCT obtaining the highest weight (52%), UHF and AE obtaining weights of 31% and 17% respectively. After the CNN network preliminarily locates the coordinates, the accurate position is obtained after optimization and correction. The D-S evidence fusion determines the floating discharge probability to be 91%, providing a strong basis for fault confirmation.

[0037] Early warning response: the discharge index DI is calculated to be 71.6, triggering an orange early warning. The system automatically increases the sampling rate to 10Hz and activates the UHF directional scanning mode, focusing on the suspicious area for further monitoring and analysis.

[0038] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A partial discharge fusion detection method based on multiple sensors, characterized in that, The specific steps include the following: S1: Multimodal sensor collaborative deployment; S2: Spatiotemporal synchronous data acquisition; S3: Signal preprocessing and feature extraction; S4: Feature-level fusion innovation; S5: Decision-level fusion mechanism; and S6: 3D visualization and early warning. First, after S1: multimodal sensor collaborative deployment is completed, S2: spatiotemporal synchronous data acquisition is performed. Then, after S2: spatiotemporal synchronous data acquisition is completed, S3: signal preprocessing and feature extraction is performed. After S3: signal preprocessing and feature extraction is completed, S4: feature-level fusion innovation is performed. After S4: feature-level fusion innovation is completed, S5: decision-level fusion mechanism is performed. Finally, after S5: decision-level fusion mechanism is completed, S6: 3D visualization and early warning is performed.

2. The partial discharge fusion detection method based on multiple sensors according to claim 1, characterized in that: S1: Multimodal sensor collaborative deployment involves the collaborative deployment of ultra-high frequency sensors, ultrasonic sensors, high frequency current transformers, and transient ground voltage sensors; and innovative spatial topology calibration utilizes a laser rangefinder to measure the three-dimensional coordinates of each sensor relative to the circuit breaker contacts with millimeter-level precision, constructing a detailed spatial position mapping table.

3. The partial discharge fusion detection method based on multiple sensors according to claim 2, characterized in that: The S1: The ultra-high frequency sensor in the multi-modal sensor collaborative deployment is carefully installed inside the insulating observation window of the switch cabinet to ensure unobstructed reception of electromagnetic waves. It adopts a spiral directional antenna design to optimize the reception of electromagnetic waves in the 300MHz to 1.5GHz frequency band. The antenna gain is not less than 5 dB, which greatly enhances the signal reception capability. The ultrasonic sensor is precisely attached to the stress concentration area on the outer wall of the cabinet. The acoustic impedance is matched by a silicone grease layer with a thickness of only 0.2 mm, which ensures the efficient transmission of sound wave signals. The resonant frequency is precisely controlled at 150 kHz to capture sound wave signals in a specific frequency range. The high-frequency current transformer is cleverly clipped onto the cable grounding wire, 30 centimeters away from the cabinet. It uses a Rogowski coil structure to achieve high-precision current measurement. At a frequency of 10 MHz, the transmission impedance reaches 100 mV / mA, accurately capturing current pulse signals. The transient ground voltage sensor is attached to the hinge of the metal cabinet door and has a built-in bandpass filter from 1 to 100 MHz to effectively filter interference signals. The detection sensitivity is up to 1 mV / dB / µV, ensuring sensitive capture of ground voltage signals. Based on the propagation characteristics of electromagnetic waves and sound waves in different propagation media, the signal delay is dynamically calculated to ensure the accuracy of data processing. The propagation speed of electromagnetic waves in the insulating gas inside the cabinet is accurately calculated as 70% of the speed of light in a vacuum, while the propagation speed of sound waves in the steel plate is corrected in real time according to the Young's modulus and density of the material to ensure the accurate propagation time of the sound wave signal.

4. The partial discharge fusion detection method based on multiple sensors according to claim 1, characterized in that: The S2: Spatiotemporal synchronization data acquisition is achieved through nanosecond-level synchronization control. The main control unit receives the second pulse signal from GPS satellites through optical fiber to ensure the accuracy of the time reference. The intelligent triggering strategy dynamically monitors background noise, calculates the median absolute deviation of the noise every 10 seconds, and sets the trigger threshold to 3 times the standard deviation of the noise amplitude to effectively avoid false triggering. Synchronization trigger commands are generated using a field-programmable gate array (FPGA) to achieve precise synchronization of each acquisition channel. Each acquisition channel uses a high-precision clock chip with a clock jitter of less than 50 picoseconds, ensuring that the synchronization error of the entire system is controlled within ±1 nanosecond, thus achieving high-precision data acquisition. When any two sensors exceed the threshold simultaneously, or when the signal of a single sensor exceeds the limit for 200 milliseconds, the full-channel data acquisition is intelligently triggered to ensure the complete capture of important signals.

5. The partial discharge fusion detection method based on multiple sensors according to claim 1, characterized in that: The S3: Signal preprocessing and feature extraction can adaptively perform wavelet denoising, automatically selecting the most suitable wavelet basis for denoising based on the steepness of the signal waveform; Accurate pulse feature extraction utilizes the zero-crossing point of the first-order difference of the signal to detect the pulse boundary, effectively eliminating interference from noise spikes and ensuring accurate segmentation of the pulse signal. For sharp pulse signals with a kurtosis greater than 5, the db8 wavelet basis is selected to effectively remove high-frequency noise. For smooth and continuous signals, the sym6 wavelet basis is selected to preserve signal details. The noise reduction threshold is dynamically adjusted according to the signal entropy value to ensure that weak pulse characteristics are preserved while reducing noise, thereby improving signal quality. A comprehensive feature library is constructed, including time-domain features, frequency-domain features, and time-frequency features, to provide a rich information foundation for subsequent fault diagnosis.

6. The partial discharge fusion detection method based on multiple sensors according to claim 1, characterized in that: The S4: Feature-level fusion innovation uses a two-factor weight dynamic allocation method to dynamically allocate weights based on the signal-to-noise ratio and confidence level of each sensor signal; Signal-to-noise ratio (SNR) factor: Calculates the base weight value based on the ratio of signal amplitude to background noise, ensuring that sensors with higher signal quality receive higher weights. Confidence factor: Based on historical database statistics of sensor detection accuracy under different operating conditions, it provides a reliable basis for weight allocation. The fusion rules are flexible and varied. When the sensor confidence level is below 0.5, the channel is automatically blocked to effectively avoid interference from erroneous information. The activation coefficient is superimposed on the SNR and confidence level to add a correlation compensation term with typical discharge characteristics, thereby improving the recognition weight of similar discharge patterns and ensuring the accuracy of diagnostic results.

7. The partial discharge fusion detection method based on multiple sensors according to claim 1, characterized in that: The S5: Decision-level fusion mechanism constructs a multi-model collaborative diagnosis system, including SVM classifiers, CNN networks, and HMM models, through multi-model collaborative diagnosis, making full use of their respective advantages for fault diagnosis; intelligent handling of evidence conflicts: when there are significant discrepancies in the diagnostic results of different models, the intelligent handling mechanism is automatically activated.

8. The partial discharge fusion detection method based on multiple sensors according to claim 7, characterized in that: The S5: In the decision-level fusion mechanism, the SVM classifier uses a 20-dimensional time-frequency feature vector to classify discharge types and accurately identify different types of discharge faults; the CNN network processes wavelet time-spectrum images to achieve accurate prediction of the three-dimensional coordinates of the discharge source; and the HMM model analyzes the pulse interval time series to predict the probability of discharge development trends, providing strong support for fault early warning. Labeling the current data as conflict samples provides valuable data support for subsequent model optimization. The online learning module is then activated to update model parameters, improve the model's adaptability and accuracy, reduce the confidence score of the conflict sensor, and avoid interference from erroneous information in the diagnostic results.

9. The partial discharge fusion detection method based on multiple sensors according to claim 1, characterized in that: The S6: 3D visualization and early warning uses high-precision positioning technology and time difference positioning method to construct a hyperboloid equation system by combining the spatial coordinates and time delay difference of each sensor; A tiered early warning strategy is implemented, setting different early warning levels and response measures based on different ranges of the discharge index.

10. The partial discharge fusion detection method based on multiple sensors according to claim 9, characterized in that: The S6: In the three-dimensional visualization and early warning, the particle swarm optimization algorithm is used to iteratively calculate the coordinates of the power supply, achieving centimeter-level positioning accuracy and providing a reliable basis for fault location; Under normal conditions, routine monitoring is performed. Under alert conditions, the sampling rate is increased to 10 times per second to enhance monitoring efforts. Under dangerous conditions, the circuit breaker interlock protection is triggered to ensure equipment safety. Under high-noise conditions, the DI calculation weight is automatically reduced to avoid false alarms and improve the stability and reliability of the early warning system.

Images