A partial discharge comprehensive diagnosis system and method suitable for a substation environment

Abstract

The application belongs to the technical field of power equipment insulation monitoring, and discloses a partial discharge comprehensive diagnosis system and method suitable for a substation environment, wherein the system comprises an acoustic sensor array module, a TEV sensor module, a synchronous data acquisition unit, a main control processing unit and a man-machine interaction and communication module. The system provides a unified global sampling clock for multiple signals through the synchronous data acquisition unit with FPGA as the core, and realizes the hardware-level microsecond-to-nanosecond synchronous acquisition. The method is based on the synchronous acquisition of acoustic waves and TEV signals, extracts time domain, frequency domain and phase domain features in parallel, carries out intelligent diagnosis and event type discrimination through a two-stage cascaded hybrid classification model, and realizes high-precision three-dimensional spatial positioning of the discharge source by using a hyperbolic equation set for solving after fusing acoustic TDOA positioning and TEV absolute time reference. The application has the advantages of edge intelligence, strong anti-interference capability and accurate positioning.

Description

Technical Field

[0001] This invention belongs to the field of power equipment insulation condition monitoring technology, and particularly relates to a comprehensive diagnostic system and method for partial discharge suitable for substation environments. Background Technology

[0002] Partial discharge is an early sign and major manifestation of insulation degradation in high-voltage electrical equipment. Effective detection and precise location of partial discharge are crucial for ensuring the safe operation of the power grid. Currently, commonly used partial discharge detection technologies mainly include acoustic methods and TEV methods.

[0003] Traditional acoustic imaging or positioning systems typically employ a single ultrasonic sensor array, whose spatial resolution is limited by the array aperture and acoustic wavelength. In complex substation environments, these systems are susceptible to background noise, acoustic wave reflection, and refraction, resulting in limited positioning accuracy and reliability. Furthermore, conventional acoustic detection usually operates in the 10kHz-60kHz frequency band, which may not be able to capture discharge details at higher frequencies.

[0004] The TEV detection method identifies discharge activity by detecting the transient rise in ground potential induced on the metal casing by internal discharges in the equipment. However, in the strong electromagnetic interference environment of substations, this method is susceptible to external electromagnetic noise, which may lead to false alarms. Furthermore, the installation location and contact quality of the TEV sensor significantly affect signal coupling efficiency, and it typically only provides a determination of the presence of the discharge, making spatial location difficult.

[0005] Existing joint detection solutions often use software post-processing to perform time alignment and fusion analysis on separately acquired acoustic and electrical signals. The synchronization accuracy is often limited to the millisecond level, which is insufficient to meet the microsecond or even nanosecond time synchronization requirements for centimeter-level precise positioning.

[0006] Therefore, there is an urgent need for a comprehensive diagnostic system for partial discharge in substations that can achieve high-precision hardware synchronization, has strong anti-interference capabilities, and can integrate multi-mode signals for rapid, accurate positioning and intelligent diagnosis. Summary of the Invention

[0007] The purpose of this invention is to provide a comprehensive partial discharge diagnostic system and method suitable for substation environments, in order to solve the above-mentioned technical problems.

[0008] To solve the above-mentioned technical problems, the specific technical solution of the partial discharge comprehensive diagnostic system and method applicable to substation environments of the present invention is as follows:

[0009] A comprehensive diagnostic system for partial discharge in substation environments includes: an acoustic sensor array module for acquiring ultrasonic signals generated by partial discharge; a TEV sensor module for acquiring transient voltage signals to ground caused by partial discharge; a synchronous data acquisition unit for providing a unified global sampling clock for the acoustic sensor array module and the TEV sensor module, and for realizing hardware-level synchronous acquisition of multiple signals; a main control processing unit for receiving synchronously acquired data, performing signal processing, fusion analysis, and intelligent diagnosis; and a human-machine interaction and communication module for status display, parameter setting, and data transmission.

[0010] Furthermore, the acoustic sensor array module includes an acoustic sensor array formed by arranging multiple MEMS ultrasonic sensors in a preset geometric shape, with an array diameter of no more than 20 cm and an element spacing that is an integer multiple of half a wavelength.

[0011] Furthermore, the acoustic sensor array has no fewer than 128 sensors, the array center frequency is set at around 40kHz, and the element spacing is set at 11mm.

[0012] Furthermore, the TEV sensor module adopts a planar capacitor structure or a high-frequency current transformer structure, with a bandwidth covering 1MHz-150MHz. The sensor housing integrates a Faraday cage structure and has a built-in hardware filtering network.

[0013] Furthermore, the signal processing circuit of the TEV sensor module includes a five-stage LC filter network and a low-noise preamplifier circuit, which are used to suppress out-of-band interference and improve signal detection sensitivity.

[0014] Furthermore, the synchronous data acquisition unit is based on a field-programmable gate array (FPGA) to provide a unified global sampling clock for all acquisition channels and uses a TEV signal as a trigger reference to start the synchronous acquisition of the acoustic sensor array.

[0015] Furthermore, the main control processing unit adopts a dual-core architecture of FPGA and embedded processor ARM, where the FPGA is responsible for high-speed data acquisition and signal processing algorithm execution, and the ARM is responsible for system control, data integration and communication.

[0016] This invention also discloses a comprehensive diagnostic method for partial discharge in substation environments, applied to the aforementioned system. The method includes: Step S1: Parallel feature extraction of synchronously acquired acoustic and TEV signals, with extraction dimensions including time-domain, frequency-domain, and phase-domain features; Step S2: Inputting the extracted features into a preset two-layer cascaded hybrid classification model for event type discrimination, the model including a first-level initial judgment based on support vector machines and a second-level deep analysis based on lightweight neural networks; Step S3: Calculating the high-precision time difference (TDOA) between signals of each channel of the acoustic array using the generalized cross-correlation function method based on the hardware synchronously acquired data; Step S4: Fusing acoustic TDOA information with the absolute time reference of the TEV signal, and establishing and solving a hyperbolic equation system to achieve three-dimensional spatial positioning of the discharge source.

[0017] Furthermore, step S2 also includes dynamic reliability assessment and adaptive optimization of the diagnostic results, specifically including:

[0018] Based on the accuracy feedback of historical diagnostic results, a dynamic evaluation mechanism for model prediction performance is established.

[0019] Based on changes in the on-site environment and the evolution of interference patterns, the hybrid classification model is updated through incremental learning.

[0020] When the confidence level of a diagnosis falls below a preset threshold multiple times in a row, the parameter adjustment or model retraining process will be automatically initiated.

[0021] Furthermore, in step 4, when the confidence level of the positioning result is lower than a preset threshold, a verification mechanism is triggered. The verification mechanism includes at least one of the following: performing time-frequency analysis on the original signal to identify multipath effects, verifying the time-domain correlation of acoustic and electrical pulses, querying the historical event database for pattern matching, and activating an enhanced monitoring mode for abnormal areas.

[0022] The partial discharge comprehensive diagnostic system and method of the present invention, applicable to substation environments, has the following advantages:

[0023] Hardware-level synchronization at the nanosecond to microsecond level has been achieved: by driving the signal with a unified clock via FPGA, the problem of high-precision time alignment of multi-source signals has been fundamentally solved, laying a solid foundation for subsequent precise positioning.

[0024] It has edge intelligent diagnostic capabilities: the core signal processing and diagnostic algorithms are embedded in the hardware and embedded firmware, and the system can run independently without the host computer, realizing millisecond-level real-time on-site diagnosis and early warning, with fast response speed.

[0025] Strong anti-interference capability: From physical shielding to hardware filtering, and then to interference identification and filtering at the algorithm layer, a three-dimensional anti-interference system combining software and hardware has been formed, which significantly improves the detection reliability in complex substation environments.

[0026] High integration and easy deployment: It adopts an integrated and modular design with a compact structure, adapts to the harsh environment of substations, and is easy to install and operate stably for a long time.

[0027] High positioning accuracy: By combining the advantages of both acoustic and electrical signals and utilizing the TDOA algorithm and joint solution model, the shortcomings of acoustic positioning being susceptible to multipath effects are effectively overcome, achieving a positioning accuracy far superior to traditional single detection methods. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall hardware architecture of the system of the present invention.

[0029] Figure 2 An example diagram of an acoustic spiral array arrangement.

[0030] Figure 3 This is an example diagram of the signal processing flow for a TEV sensor.

[0031] Figure 4 This is a schematic diagram of a five-stage LC filter in a TEV processing circuit.

[0032] Figure 5 This is a flowchart of an intelligent diagnosis process based on a two-level classification model.

[0033] Figure 6 This is a schematic diagram of the system signal synchronization and processing flow.

[0034] Figure 7 This is a schematic diagram of the localization results of surface discharge in the experiment. Detailed Implementation

[0035] To better understand the purpose, structure, and function of this invention, the following detailed description, in conjunction with the accompanying drawings, provides an integrated partial discharge diagnostic system and method applicable to substation environments.

[0036] like Figure 1As shown, the present invention discloses a comprehensive diagnostic system for partial discharge in substations based on an acoustic array and a TEV sensor, mainly comprising: an acoustic sensor array module, a TEV sensor module, a synchronous data acquisition unit, a main control processing unit, and a human-machine interaction and communication module. Each module and unit is designed with plug-and-play interfaces for easy upgrades and maintenance. The device can be equipped with either the acoustic sensor array module or the TEV sensor module individually for data acquisition, or both modules can be installed simultaneously to detect internal insulation faults and external physical leaks or discharges, thus addressing both internal and external factors.

[0037] Acoustic sensor array module: This module consists of M (e.g., no fewer than 128) MEMS ultrasonic sensors arranged in a preset geometry (e.g., spiral or circular) to capture ultrasonic signals generated by partial discharge; preferably, the array diameter is 14 cm. Data processing of the array uses a dedicated decoding chip to avoid signal attenuation caused by excessively long lines. The output uses a MIPI interface, and the acoustic sensor array has a built-in electromagnetic shielding layer to prevent TEV signal crosstalk.

[0038] like Figure 2 As shown, preferably, the acoustic sensor array module adopts a design concept that complements acoustic and electrical characteristics. Based on MEMS microphone array technology, the acoustic sensor array module consists of no fewer than 128 ultrasonic sensors distributed on a circular substrate according to a specific geometric pattern, forming a circular array layout. This layout allows the array to effectively cover a 360° detection range, and each microphone is responsible for receiving ultrasonic signals from a specific direction, thereby achieving spatial orientation identification of the partial discharge source. Crucially, the center frequency of the acoustic sensor array is set around 40kHz. This frequency band selection avoids audible interference while ensuring a specific response to the ultrasonic signals of partial discharge.

[0039] The structural parameters of the acoustic sensor array underwent rigorous theoretical calculations and experimental verification. The element spacing was set to an integer multiple of the half-wavelength of the ultrasound. For an ultrasound with a center frequency of 40kHz, the half-wavelength in air is approximately 4.3mm. In the actual design, considering the balance between array aperture and resolution, the element spacing was set to 11mm. This value was optimized using a virtual array expansion algorithm (an existing algorithm). Through this optimization, the acoustic sensor array achieved higher direction-finding accuracy without significantly increasing its physical size.

[0040] The performance indicators of the acoustic sensor array have undergone rigorous laboratory testing. Under standard conditions, the array achieves a detection sensitivity of 1 pC for partial discharge ultrasonic signals, with an azimuth estimation error of no more than ±3° and an elevation estimation error of no more than ±5%. Within a 3-meter distance, the positioning accuracy reaches ±5 cm, far exceeding the accuracy of traditional array sensor detection methods. These superior performance indicators enable the system to meet the requirements for precise partial discharge localization in substations.

[0041] like Figure 3 As shown, the TEV sensor module adopts a high-frequency current transformer or a parallel-plate capacitor structure, with a bandwidth covering the main frequency band of partial discharge electromagnetic pulses (preferably, bandwidth 1MHz-150MHz, sensitivity 0.1mV / A). It is directly mounted on the equipment grounding bolt and is used to detect transient voltage signals to ground caused by partial discharge. The sensor housing integrates a Faraday cage structure (stainless steel mesh <1mm) and has a built-in hardware filter network to filter out 50Hz power frequency interference. The output terminal uses a USB connector for easy connection to the main system.

[0042] Preferably, the TEV sensor module adopts a planar capacitor structure design and is directly mounted against the switchgear wall. The TEV sensor uses a circular copper foil as the upper electrode and the switchgear wall itself as the lower electrode. This design enables the TEV sensor to efficiently couple the high-frequency electromagnetic signals generated during partial discharge. Experiments show that this sensor has the best response characteristics to signals in the 10MHz-60MHz frequency band, which precisely covers the main energy distribution range of partial discharge electromagnetic pulses.

[0043] like Figure 4 As shown, the TEV sensor module employs a five-stage LC filter, which, through the series and parallel connection of LC filters, provides better out-of-band rejection. The TEV sensor module utilizes a detector with extremely high sensitivity and fast response speed to suppress noise, resulting in better signal reproduction in complex electromagnetic environments. It incorporates nanoscale technology and a low-noise preamplifier circuit for optimization, preferentially using the AD8138 chip, which features a high bandwidth of 320MHz and a 2.4 nV / With 1.5 pA / Low noise improves the accuracy and sensitivity of signal detection.

[0044] like Figure 5 As shown, the synchronous data acquisition unit, with a field-programmable gate array (FPGA) as its core, provides a unified global sampling clock for all acoustic sensor array channels and TEV sensor channels, realizing hardware-level synchronous acquisition of multiple signals and ensuring consistent time reference.

[0045] The synchronous data acquisition unit includes a data acquisition board with a highly integrated design, featuring an ultra-high frequency signal input interface, multiple acoustic signal input interfaces, and signal output interfaces. The front-end analog-to-digital converter (ADC) performs detection, amplification, and differential switching, employing a 12-bit precision AD9235 digital-to-analog converter chip with a sampling frequency of up to 65MHz, ensuring high-fidelity signal acquisition. The core function of the data acquisition board is to achieve synchronous acquisition of acoustic and electrical signals. When the TEV sensor detects a transient voltage signal to ground generated by partial discharge, it immediately triggers the acoustic sensor array to begin acquiring ultrasonic signals. This triggering mechanism ensures precise synchronization of the two signals in time. The acoustic sensor array then performs PDM decoding on the acquired multi-channel microphone data. The resulting PCM signal undergoes spectrum and covariance calculations, and is multiplied and accumulated with the steering matrix to obtain the localization result. Based on the maximum localization point, the speech enhancement result is calculated, while also providing single-channel spectrum analysis.

[0046] The main control processing unit, comprising an embedded processor (ARM) and a field-programmable gate array (FPGA), is responsible for system control, data integration, human-machine interaction, and communication. The unit employs a dual-core architecture of FPGA and ARM. The FPGA handles high-speed data acquisition and decoding, and executes complex signal processing algorithms, while the ARM runs an embedded Linux system, handling the integration, transmission, and display of FPGA data. This architecture ensures both the real-time performance requirements of data acquisition and the computational resource demands of complex algorithm processing. The host also includes a large-capacity memory for storing historical detection data, providing data support for subsequent trend analysis and fault prediction.

[0047] Existing technologies mostly employ software or network time synchronization, with synchronization errors typically in the millisecond range, failing to meet the microsecond-level accuracy requirements for centimeter-level positioning. The signal synchronization acquisition mechanism of this system is one of the core technologies of this invention. It employs a working mode using TEV signals as the trigger reference. When partial discharge occurs in substation equipment, the TEV sensor first detects the electromagnetic signal propagating at the speed of light. After recognizing this signal, the TEV sensor module immediately sends a trigger command to the acoustic sensor array, initiating the synchronous acquisition of multi-channel acoustic signals. This design cleverly solves the problem of synchronizing acoustic and electrical signals. Because the propagation speed of electromagnetic waves is much higher than that of sound waves, the distance between the partial discharge source and the sensor can be accurately estimated by calculating the time difference between the arrival of the TEV signal and the ultrasonic signal.

[0048] The human-machine interface and communication module includes an LCD screen and physical buttons for displaying on-site status and setting parameters. It provides Ethernet and wireless communication interfaces for uploading diagnostic results and early warning information to the backend monitoring system.

[0049] The system adopts a multi-level shielding and hardware filtering design from the sensor layer, signal conditioning layer, data acquisition layer to the chassis, forming an anti-interference system throughout the entire process.

[0050] The present invention provides a comprehensive diagnostic method for partial discharge in substations based on an acoustic array and a TEV sensor, comprising the following steps:

[0051] Step 1: Multi-feature fusion diagnosis:

[0052] The system processes data segments (containing multi-channel acoustic signals) for each candidate event. and TEV signal Parallel feature extraction is performed. Specifically, this includes:

[0053] 1) Time-domain characteristics: pulse amplitude, rise time, and decay characteristics.

[0054] Pulse amplitude: Represents the maximum instantaneous absolute value of the signal waveform deviating from the baseline (zero or DC component) within the time window of a single discharge event.

[0055]

[0056] in For sampling point index, For event time windows, Indicates baseline calibration (to)

[0057] The discrete signal sequence after DC offset.

[0058] Rise time: The time required for the pulse front to rise from 10% to 90% of its peak value, reflecting the speed of discharge.

[0059] Attenuation characteristics: The time required for the signal amplitude to decay to a certain proportion (such as 1 / e or 50%) after the pulse peak, or the fitting parameter of the attenuation process.

[0060] 2) Frequency domain characteristics: Perform FFT transformation on the acoustic signal to analyze its spectral centroid, bandwidth and other characteristics.

[0061] Spectral centroid: The "average frequency" of the spectrum, reflecting the central location of the signal energy distribution in the frequency domain.

[0062]

[0063] in, Represents discrete frequency values. This represents the power spectral density at frequency index k. and Indicates the lower and upper frequency indices of the effective frequency band. Indicates the centroid of the spectrum.

[0064] Spectral bandwidth: characterizes the degree of concentration of signal spectral energy, usually defined as the frequency range that contains a certain proportion (such as 90%) of the total signal energy.

[0065] 1) Phase domain characteristics: Extract the phase distribution spectrum of the discharge pulse over the power frequency cycle. Different types of discharges (internal, surface, corona) and interferences have different phase aggregation characteristics.

[0066] Step 2: Signal Processing and Enhancement

[0067] The extracted multi-dimensional feature vectors are input into a pre-trained two-layer cascaded hybrid classification model. The core of the two-layer cascaded hybrid classification model lies in achieving efficient and accurate classification through a progressive decision-making process.

[0068] The first stage uses support vector machines to quickly make initial judgments on highly discriminative feature subsets. Its decision function can be expressed as: ,in It is the input feature vector. It is a kernel function mapping. It is a weight vector. It is the output value of the decision function. When the output value is lower than the set threshold, the event will be directly identified as a typical disturbance and the process will be terminated.

[0069] Events that pass the initial assessment will proceed to the second-level deep analysis stage, processed by a lightweight fully connected neural network. This network calculates the class probability through a forward propagation structure containing two hidden layers, and the final output layer uses the Softmax function to generate a probability distribution vector. ,in This indicates that the currently input partial discharge event belongs to the first... The predicted probability of each category.

[0070] The system makes a comprehensive decision by fusing the confidence levels of the two-level outputs. The comprehensive confidence level calculation model is as follows: Where α and β are weighting coefficients, and σ is the Sigmoid function. Based on the confidence threshold rule, the system ultimately categorizes events into high-confidence discharge types, pending events requiring manual verification, or unknown events. This architecture, by first filtering out a large number of simple interferences and then performing in-depth analysis on suspicious events, significantly improves the overall robustness and accuracy of the diagnostic system while ensuring real-time performance.

[0071] This model is trained based on laboratory simulations and diverse field measurement data, which are precisely labeled through physical experiments and expert spectrum analysis. The model is trained using an adaptive gradient descent algorithm, effectively distinguishing between real partial discharge types and preset typical non-discharge interference modes. The system has a built-in, initializeable and dynamically updated interference feature library and possesses self-optimization capabilities based on incremental learning, continuously adapting to the field environment. For events classified as interference, the system will automatically discard or store them in the interference library without generating a warning.

[0072] An FPGA is used to denoise and beamform the acoustic array signal, estimate the spatial spectrum of the sound source, generate an initial acoustic image, and determine the location of the discharge source. Simultaneously, features such as the amplitude and pulse count of the TEV signal are extracted. Synchronous processing and fusion analysis are performed, utilizing the time difference between the acoustic and TEV electrical signals arriving at the sensor, combined with the sensor's spatial location, to accurately calculate the three-dimensional coordinates of the discharge source using a time-difference positioning algorithm. A deep learning network incorporating a physical model of sound wave propagation (delay and attenuation constraints) is employed to enhance the initial acoustic image.

[0073] like Figure 5 As shown, the system employs FIR filtering and MVDR algorithms. FIR filtering is primarily responsible for decoding and frequency compensation of the PDM signal, ensuring the flatness of the output PCM signal through downsampling, compensation, and post-processing. The MVDR algorithm, through adaptive beamforming, enhances the target speech and suppresses interference through covariance matrix processing, steering vector calculation, optimal weight vector calculation, beamforming, and output. It is the core step in array signal processing.

[0074] Step 3: High-precision latency acquisition with hardware synchronization guarantee:

[0075] The basis of precise positioning lies in accurately measuring the time difference between the arrival of ultrasonic signals at each sensor in the acoustic array.

[0076] The core hardware mechanism of this solution employs a pure hardware architecture of "FPGA global clock driving + synchronous ADC sampling + microcontroller data transmission". A single FPGA directly drives all channels (each channel of the acoustic array), while the microcontroller performs TEV data acquisition and integrates the data on the ARM module. This ensures that all sensor data are on the same absolute time base from the "sampling start moment", eliminating inherent sampling delay jitter between channels and achieving hardware-level synchronization at the nanosecond to microsecond level.

[0077] like Figure 6As shown, based on the original waveform data of multiple channels obtained through hardware synchronization and strictly aligned, a unified reference clock source is used to ensure the frequency consistency of the clock signal. Then, the clock counter is reset or adjusted by a hardware signal to ensure phase consistency. The generalized cross-correlation function method (an existing algorithm) is used to calculate the time delay. This method preprocesses and weights the signal, which can effectively suppress the effects of noise and reverberation. It finds the maximum correlation point between the two signals in the time domain, thereby calculating the high-precision time difference (TDOA) at the sub-sampling interval level. By retaining phase information and discarding amplitude information, the accuracy of time delay estimation is improved.

[0078] Step 4: Spatial localization using combined multimodal signals:

[0079] Acoustic positioning alone is susceptible to sound wave refraction and reflection, and may fail when the discharge is weak. This invention introduces TEV signals as crucial auxiliary and verification information.

[0080] Assuming the acoustic sensor array is composed of It consists of several sensors, whose spatial coordinates are known, denoted as . Let the unknown coordinates of the power source point be... The speed of sound is The moment the TEV sensor detects the discharge pulse. As an absolute time reference for the occurrence of the discharge.

[0081] 1. Establishing the equation of a hyperbola

[0082] Based on the signal arrival time difference, the following hyperbolic equation system can be established:

[0083]

[0084] in, From the source point to the 1st The distance of the sensor, These are the unknown spatial coordinates of the power source. It is the first The known spatial coordinates of the acoustic sensor For sensors The high-precision TDOA value between the reference sensor 1 and the sensor, where c is the speed of sound in air. This set of equations defines a series of hyperboloids in three-dimensional space with the sensor pair as foci, and their intersection points are the discharge points.

[0085] Using the weighted least squares-Newton iteration method, first linearize to obtain Find the initial value, where It is the augmented state vector to be solved. Represents the three-dimensional coordinates of the discharge source. Let A represent the distance from the power source to the reference sensor 1, where A is the coefficient matrix and b is the constant term vector. Then, iteratively optimize the sum of squared residuals.

[0086] 2. Low confidence level determination and verification mechanism

[0087] The quantitative criteria for "low confidence" are: 1) the distance between the location point and the center of the TEV-associated electrical area exceeds 1.5 meters; or 2) the overlap volume between the 95% confidence ellipsoid of the location result and the target area is less than 30%. Meeting either condition will result in a flag. Specific measures to trigger further diagnosis include: performing time-frequency analysis on the original signal to identify multipath effects; rigorously verifying the time-domain correlation of acoustic and electrical pulses; querying the historical event database for pattern matching; and activating enhanced monitoring modes for abnormal areas.

[0088] Experimental results: such as Figure 7 As shown, in a simulated substation environment, the system successfully detected partial discharge on the surface of the switchgear and accurately located the discharge location in the three-dimensional spatial model with a positioning error within ±5cm, verifying the effectiveness of the invention.

[0089] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A comprehensive diagnostic system for partial discharge suitable for substation environments, characterized in that, include: Acoustic sensor array module, used to collect ultrasonic signals generated by partial discharge; The TEV sensor module is used to collect transient voltage-to-ground signals caused by partial discharge. The synchronous data acquisition unit is used to provide a unified global sampling clock for the acoustic sensor array module and the TEV sensor module, and to realize hardware-level synchronous acquisition of multiple signals. The main control processing unit is used to receive synchronously acquired data, perform signal processing, fusion analysis, and intelligent diagnosis; including parallel feature extraction of synchronously acquired ultrasonic signals and transient voltage signals, the extraction dimensions including time domain features, frequency domain features, and phase domain features; inputting the extracted features into a preset two-layer cascaded hybrid classification model for event type discrimination, the model including a first-level initial judgment based on support vector machines and a second-level deep analysis based on lightweight neural networks; Based on data acquired synchronously by hardware, the high-precision time difference TDOA between signals of each channel of the acoustic array is calculated using the generalized cross-correlation function method. By integrating acoustic TDOA information with the absolute time reference of TEV signal, the three-dimensional spatial positioning of the discharge source is achieved by establishing and solving the hyperbolic equation system. The human-computer interaction and communication module is used for status display, parameter setting, and data transmission.

2. The system according to claim 1, characterized in that, The acoustic sensor array module includes an acoustic sensor array formed by arranging multiple MEMS ultrasonic sensors in a preset geometric shape. The array diameter is no more than 20cm, and the spacing between array elements is an integer multiple of half a wavelength.

3. The system according to claim 2, characterized in that, The acoustic sensor array has no fewer than 128 sensors, the array center frequency is set at around 40kHz, and the element spacing is set at 11mm.

4. The system according to claim 1, characterized in that, The TEV sensor module adopts a planar capacitor structure or a high-frequency current transformer structure, with a bandwidth covering 1MHz-150MHz. The sensor housing integrates a Faraday cage structure and has a built-in hardware filtering network.

5. The system according to claim 4, characterized in that, The signal processing circuit of the TEV sensor module includes a five-stage LC filter network and a low-noise preamplifier circuit, which are used to suppress out-of-band interference and improve signal detection sensitivity.

6. The system according to claim 1, characterized in that, The synchronous data acquisition unit is based on a field-programmable gate array (FPGA), which provides a unified global sampling clock for all acquisition channels and uses the TEV signal as a trigger reference to start the synchronous acquisition of the acoustic sensor array.

7. The system according to claim 1, characterized in that, The main control processing unit adopts a dual-core architecture of FPGA and embedded processor ARM, where the FPGA is responsible for high-speed data acquisition and signal processing algorithm execution, and the ARM is responsible for system control, data integration and communication.

8. A comprehensive diagnostic method for partial discharge applicable to substation environments, characterized in that, Applied to the system as described in any one of claims 1-7, the method comprises: Step S1: Parallel feature extraction of synchronously acquired acoustic signals and TEV signals, the extracted dimensions including time-domain features, frequency-domain features, and phase-domain features; Step S2: Inputting the extracted features into a preset two-layer cascaded hybrid classification model for event type discrimination, the model including a first-level preliminary judgment based on support vector machines and a second-level deep analysis based on lightweight neural networks; Step S3: Calculating the high-precision time difference (TDOA) between signals of each channel of the acoustic array using the generalized cross-correlation function method based on the hardware synchronously acquired data; Step S4: Fusing acoustic TDOA information with the absolute time reference of the TEV signal, and realizing the three-dimensional spatial positioning of the discharge source by establishing and solving a hyperbolic equation system.

9. The method according to claim 8, characterized in that, Step S2 further includes dynamic reliability assessment and adaptive optimization of the diagnostic results, specifically including: Based on the accuracy feedback of historical diagnostic results, a dynamic evaluation mechanism for model prediction performance is established. Based on changes in the on-site environment and the evolution of interference patterns, the hybrid classification model is updated through incremental learning. When the confidence level of a diagnosis falls below a preset threshold multiple times in a row, the parameter adjustment or model retraining process will be automatically initiated.

10. The method according to claim 8, characterized in that, In step S4, when the confidence level of the positioning result is lower than a preset threshold, a verification mechanism is triggered. The verification mechanism includes at least one of the following: performing time-frequency analysis on the original signal to identify multipath effects, verifying the time-domain correlation of acoustic and electrical pulses, querying the historical event database for pattern matching, and activating an enhanced monitoring mode for abnormal areas.

Images