Accurate monitoring system for abnormal discharge hidden danger of self-adaptive intelligent distribution box

Abstract

The invention discloses a self-adaptive intelligent power distribution box abnormal discharge hidden danger accurate monitoring system, and relates to the technical field of power equipment monitoring. The system comprises a sensor array used for collecting multi-physical field signals in a distribution box; the decoupling module is used for performing interference decoupling and feature extraction of time, frequency and space dimensions on the multi-physical field signal; and the edge-cloud collaborative computing architecture comprises an edge computing node and a cloud platform, the edge computing node is used for executing signal real-time processing and abnormal early warning, and the cloud platform is used for constructing an equipment health model and outputting long-term trend prediction. According to the invention, through the sensor array, the time-frequency-space three-dimensional signal decoupling technology and the edge-cloud cooperative computing architecture, high-precision acquisition and interference decoupling of multi-physical field signals are realized, and the monitoring precision and reliability of abnormal discharge hidden dangers are fundamentally improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment monitoring, and specifically to an adaptive intelligent distribution box abnormal discharge hidden danger precise monitoring system. Background Technique

[0002] As a key node in the power system, the abnormal discharge inside the distribution box is the main inducement for insulation failure. Existing monitoring technologies adopt a multi-sensor fusion scheme (such as ultra-high frequency UHF, ultrasonic AE, transient earth voltage TEV sensors), but there are problems in decoupling the interference of multi-sensor signals: signal coupling in the time domain, frequency domain, and spatial domain (such as the cross-interference between UHF signals and AE signals in the 20 kHz - 300 kHz frequency band), resulting in the discharge positioning error expanding to more than 15%, and the false alarm rate being as high as 30%. In a certain subway project, due to the same-frequency interference of adjacent sensors, environmental noise was misidentified as partial discharge, leading to unnecessary outage maintenance. The traditional threshold alarm mechanism and fixed baseline calibration method cannot effectively identify the signal coupling interference in a complex electromagnetic environment, resulting in missed detection or misjudgment of early discharge hidden dangers, seriously affecting the operation reliability of the distribution box. Summary of the Invention

[0003] The purpose of the present invention is to provide an adaptive intelligent distribution box abnormal discharge hidden danger precise monitoring system to solve the problems raised in the above background technique.

[0004] To achieve the above purpose, the present invention provides the following technical solution: An adaptive intelligent distribution box abnormal discharge hidden danger precise monitoring system, including:

[0005] A sensor array for collecting multi-physical field signals inside the distribution box;

[0006] A decoupling module for performing interference decoupling and feature extraction on the multi-physical field signals in the time, frequency, and spatial dimensions;

[0007] An edge-cloud collaborative computing architecture, including an edge computing node and a cloud platform, where the edge computing node is used to perform real-time signal processing and abnormal early warning, and the cloud platform is used to construct a device health model and output long-term trend prediction.

[0008] Preferably, the sensor array includes an ultra-high frequency UHF sensor, an ultrasonic AE sensor, a transient earth voltage TEV sensor, an ultraviolet sensor, and an infrared sensor, and quantum dot sensitized thin films are deposited on the surfaces of the sensitive elements of each sensor;

[0009] Among them, the quantum dot sensitized thin film of the ultra-high frequency UHF sensor uses CdSe material, and the quantum dot sensitized thin film of the AE sensor uses a ZnO quantum dot and piezoelectric ceramic heterostructure.

[0010] Preferably, the time - dimension decoupling of the decoupling module adopts the dual - time - slice multiplexing protocol DTMP, and each monitoring period is divided into a high - frequency time - slice and a low - frequency time - slice;

[0011] The high - frequency time - slice is used to activate the ultra - high - frequency UHF sensor and the ultraviolet sensor, and the duration is 50μs - 200μs;

[0012] The low - frequency time - slice is used to activate the ultrasonic AE sensor, the transient earth voltage TEV sensor and the infrared sensor, and the duration is 200μs - 2ms;

[0013] Among them, the decoupling module avoids the time - domain overlap of different sensor signals through a time - sharing activation mechanism.

[0014] Preferably, the frequency - dimension decoupling of the decoupling module includes:

[0015] A wavelet packet decomposition sub - unit, which is used to perform 8 - layer wavelet packet decomposition on the ultra - high - frequency UHF sensor signal and extract a 128 - dimensional time - frequency feature vector in the 200MHz - 300MHz frequency band;

[0016] A compressive sensing reconstruction sub - unit, which sparsely represents the ultrasonic AE sensor signal and reconstructs the effective signal in the 20kHz - 100kHz frequency band to suppress the power - frequency interference below 20kHz;

[0017] The decoupling module improves the signal - to - noise ratio to more than 25dB through a frequency - domain feature separation algorithm.

[0018] Preferably, the spatial - dimension decoupling unit of the decoupling module adopts the ultra - wideband time - difference - of - arrival positioning method. A three - dimensional array is formed by at least 3 ultra - high - frequency UHF sensors and 2 ultrasonic AE sensors, and the discharge point coordinates are calculated based on the time - difference - of - arrival TDoA of the signals; the mathematical model of the positioning algorithm is:

[0019]

[0020] Among them, (x i ,y i ,z i ) is the three - dimensional coordinate of the i - th sensor, Δt i-j is the time - difference between the signal arriving at the i - th and the j - th sensors, and c is the propagation speed of electromagnetic waves in the air.

[0021] Preferably, the edge computing node adopts a heterogeneous computing platform of an FPGA unit + a DSP unit;

[0022] The FPGA unit realizes signal pre - processing, including baseline subtraction, digital filtering and analog - to - digital conversion;

[0023] The DSP unit runs a fusion model of the lightweight convolutional neural network MobileNetV3 and the long short-term memory network LSTM to perform pattern recognition on the discharge signal.

[0024] Preferably, the FPGA unit preprocesses the signal through a dynamic baseline calibration algorithm, specifically including:

[0025] Construct a Gaussian mixture model based on the historical sensor data as the signal baseline, and calculate the Mahalanobis distance between the current signal and the baseline in real time;

[0026] When the Mahalanobis distance exceeds 3 times the standard deviation, an abnormal warning is triggered, and at the same time, the warning threshold is dynamically adjusted through the particle swarm optimization algorithm PSO.

[0027] Preferably, the cloud platform is also used to construct a graph neural network GNN device health model, and the graph neural network GNN device health model includes:

[0028] The input layer fuses the discharge signal characteristics, temperature field data, load current curve, and device ledger information;

[0029] The graph convolutional layer captures the non-linear associations between multi-source data through an attention mechanism to generate a three-dimensional fault tree map;

[0030] The output layer predicts the remaining life of the insulating material.

[0031] Preferably, the sensor array is built-in with a self-calibration module;

[0032] The self-calibration module includes a temperature compensation circuit and a quantum dot fluorescence intensity monitoring unit;

[0033] The quantum dot fluorescence spectrum monitoring unit collects the emission wavelength offset of the quantum dot sensitized film in real time and establishes a temperature drift error correction model:

[0034] S(T) = S0·(1 + k·ΔT)

[0035] where S(T) is the sensor sensitivity at temperature T, S0 is the reference sensitivity at 25°C, k is the quantum dot temperature drift coefficient (k ≤ 5×10 -4 / °C), and ΔT is the temperature deviation;

[0036] The temperature compensation circuit stabilizes the operating temperature of the sensor sensitive element within the range of 25°C ± 0.1°C through a feedback control loop composed of a thermoelectric cooler and a temperature sensor to compensate for the temperature drift effect of the quantum dot material.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] High-precision acquisition of multi-physical field signals through a sensor array significantly improves the ability to capture weak discharge signals and the long-term stability of sensors, reduces the frequency of manual calibration. The decoupling module can decouple the interference of multi-sensor signals in the time, frequency, and space dimensions, effectively separate the discharge signals from the noise, achieve precise positioning and pattern recognition of abnormal discharges, significantly reduce the false alarm rate. The edge-cloud collaborative computing architecture can perform real-time signal processing and intelligent analysis. The edge node realizes millisecond-level abnormal early warning, and the cloud platform constructs a device health model by combining multi-source data to provide long-term trend prediction, supporting predictive maintenance strategies and improving the operation reliability and maintenance efficiency of the distribution box. Brief Description of the Drawings

[0039] Figure 1 FIG. is a schematic structural diagram of an adaptive intelligent distribution box abnormal discharge hidden danger precise monitoring system provided by an embodiment of the present invention;

[0040] Figure 2 FIG. is a specific operation step diagram of an adaptive intelligent distribution box abnormal discharge hidden danger precise monitoring system provided by an embodiment of the present invention. Detailed Embodiments

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0042] Embodiment 1

[0043] Please refer to Figure 1 , the present invention provides an adaptive intelligent distribution box abnormal discharge hidden danger precise monitoring system, and the system architecture includes:

[0044] A sensor array 11 for collecting multi-physical field signals in the distribution box;

[0045] A decoupling module 12 for decoupling interference and feature extraction of the multi-physical field signals in the time, frequency, and space dimensions;

[0046] An edge-cloud collaborative computing architecture 13, including an edge computing node and a cloud platform. The edge computing node is used to perform real-time signal processing and abnormal early warning, and the cloud platform is used to construct a device health model and output long-term trend prediction.

[0047] In an alternative embodiment, the sensor array 11 includes an ultra-high frequency (UHF) sensor, an ultrasonic acoustic emission (AE) sensor, a transient earth voltage (TEV) sensor, an ultraviolet (UV) sensor, and an infrared (IR) sensor. Quantum dot sensitized films are deposited on the surfaces of the sensitive elements of each sensor.

[0048] Among them, the quantum dot sensitized film of the UHF sensor uses CdSe material, and the quantum dot sensitized film of the AE sensor uses a heterostructure of ZnO quantum dots and piezoelectric ceramics.

[0049] Specifically, the sensor array 11 integrates five types of sensors, covering the multi-physical field characteristics of the discharge process.

[0050] Among them, the UHF sensor: is used to capture the electromagnetic pulse signals of 300 MHz - 3 GHz generated by the discharge. The sensitive element is a microstrip antenna structure, on the surface of which a 50 nm thick CdSe quantum dot film is deposited. The exciton binding energy (40 meV) of the CdSe quantum dots increases the response sensitivity of the sensor to weak discharge signals of 0.01 pC level by 10 times, achieving an order of magnitude breakthrough compared with traditional UHF sensors (0.1 pC).

[0051] The ultrasonic AE sensor: uses a heterostructure composed of ZnO quantum dots and PZT piezoelectric ceramics. The tunneling effect of the quantum dots increases the mechanical vibration - electrical signal conversion efficiency from 60% to 84%, and can detect acoustic wave signals as low as 20 μV (the threshold of traditional AE sensors is 50 μV).

[0052] The transient earth voltage (TEV) sensor: based on the principle of capacitive coupling, detects the transient voltage signals on the surface of the cabinet, and the effective frequency band is 10 kHz - 100 MHz.

[0053] The UV sensor: uses a solar-blind ultraviolet photomultiplier tube to detect the ultraviolet radiation of 240 nm - 280 nm generated by the discharge, and the resolution reaches 1 μW / cm 2 ;

[0054] The IR sensor: an uncooled focal plane array, with a temperature measurement accuracy of ±0.5 °C, is used to capture the local temperature rise accompanied by the discharge.

[0055] Among them, the nano-size effect (particle size 2 - 10 nm) of the quantum dot material endows it with quantum confinement characteristics. When the electromagnetic / acoustic wave signals generated by the discharge act on the quantum dot film, the exciton absorption cross-section increases, and the carrier mobility is increased by 3 times (the electron mobility of CdSe quantum dots reaches 10 4 cm 2 / (V·s)) at 300 K. Taking the UHF sensor as an example, the quantum dot film expands the equivalent noise bandwidth of the antenna from 200 MHz to 5 GHz, significantly enhancing the signal capture ability of the sensor in the high frequency band.

[0056] In an alternative embodiment, the sensor array 11 is built-in with a self-calibration module;

[0057] The self-calibration module includes a temperature compensation circuit and a quantum dot fluorescence intensity monitoring unit;

[0058] The quantum dot fluorescence spectrum monitoring unit collects the emission wavelength offset of the quantum dot sensitized film in real time and establishes a temperature drift error correction model:

[0059] S(T) = S0·(1 + k·ΔT

[0060] where S(T) is the sensor sensitivity at temperature T, S0 is the initial sensitivity at the reference temperature of 25°C, k is the quantum dot temperature drift coefficient (k ≤ 5×10 -4 / °C), ΔT is the temperature deviation, that is, the deviation value between the actual working temperature and the reference temperature. The temperature drift coefficient of CdSe quantum dot material is k ≤ 4.8×10 -4 / °C, and for ZnO quantum dots it is k ≤ 3.5×10 -4 / °C. Through this model, the continuous operation time of the sensor without manual calibration can reach 18 months, which is 6 times that of traditional sensors (with a calibration cycle of 3 months);

[0061] The temperature compensation circuit stabilizes the working temperature of the sensor sensitive element within the range of 25°C ± 0.1°C through a feedback control loop composed of a thermoelectric cooler and a temperature sensor to compensate for the temperature drift effect of the quantum dot material.

[0062] Specifically, the built-in temperature compensation circuit and the quantum dot fluorescence spectrum monitoring unit of the sensor form a double closed-loop self-calibration system. Among them, the quantum dot fluorescence spectrum monitoring unit uses CdSe / ZnO core-shell structure quantum dots as the sensing medium, and captures the emission wavelength offset of the quantum dots in real time through a high-resolution spectrometer. This offset has a strong correlation with temperature changes. The supporting temperature compensation circuit uses a Peltier effect thermoelectric cooler and a platinum resistance temperature sensor to form a feedback control loop, which can stabilize the working environment temperature of the sensor within the range of ±0.1°C.

[0063] Among them, the spectrum monitoring unit collects quantum dot spectrum data every 5 minutes, extracts the wavelength offset after being processed by fast Fourier transform (FFT), and inputs the measured temperature value of the temperature compensation circuit into the temperature drift correction model to automatically adjust the sensor output compensation coefficient. In the extreme temperature change (-20°C to 60°C) environment test, this self-calibration module controls the measurement error within ±0.3%, which is a significant improvement compared to the ±2% error accuracy of traditional sensors. Practical engineering applications show that the continuous operation time of the sensor without manual calibration can reach 18 months. Compared with traditional sensors with a 3-month calibration cycle, the maintenance efficiency is increased by 6 times, and the operation and maintenance cost is greatly reduced.

[0064] In an alternative embodiment, the time - dimension decoupling of the decoupling module 12 adopts a dual - time - slice multiplexing protocol DTMP, which divides each monitoring period into a high - frequency time - slice and a low - frequency time - slice;

[0065] The high - frequency time - slice is used to activate the ultra - high - frequency UHF sensor and the ultraviolet sensor, and the duration is 50 μs - 200 μs;

[0066] The low - frequency time - slice is used to activate the ultrasonic AE sensor, the transient earth voltage TEV sensor and the infrared sensor, and the duration is 200 μs - 2 ms;

[0067] Among them, the decoupling module 12 avoids the time - domain overlap of different sensor signals through a time - division activation mechanism.

[0068] It can be understood that the dual - time - slice multiplexing protocol DTMP divides each monitoring period (10 ms) into a high - frequency time - slice (T1) and a low - frequency time - slice (T2):

[0069] High - frequency time - slice (T1 = 50 μs - 200 μs): Activate the UHF sensor and the ultraviolet sensor, which are used to capture nanosecond - level electromagnetic pulse and ultraviolet radiation signals. A time - division multiplexing control circuit can be used to generate mutually exclusive activation signals through a timing controller to ensure that the two sensors do not work simultaneously and avoid electromagnetic coupling interference;

[0070] Low - frequency time - slice (T2 = 200 μs - 2 ms): Activate the AE sensor, the TEV sensor and the infrared sensor to collect acoustic wave (20 kHz - 100 kHz), earth voltage (10 kHz - 100 MHz) and temperature signals. The duration of T2 is dynamically adjusted according to the maximum propagation distance of the acoustic wave signal (about 2 meters in the distribution box, and the acoustic wave propagation time is about 6 ms) to ensure complete signal acquisition.

[0071] In an alternative embodiment, the frequency - dimension decoupling of the decoupling module 12 includes:

[0072] A wavelet packet decomposition sub - unit, which is used to perform 8 - layer wavelet packet decomposition on the ultra - high - frequency UHF sensor signal and extract a 128 - dimensional time - frequency feature vector in the 200 MHz - 300 MHz frequency band;

[0073] A compressive sensing reconstruction sub - unit, which sparsely represents the ultrasonic AE sensor signal and reconstructs the effective signal in the 20 kHz - 100 kHz frequency band to suppress the power - frequency interference below 20 kHz;

[0074] The decoupling module 12 improves the signal - to - noise ratio to more than 25 dB through a frequency - domain feature separation algorithm.

[0075] Specifically, the frequency dimension decoupling adopts a joint algorithm of wavelet packet and compressive sensing:

[0076] UHF signal processing: Through 8-layer wavelet packet decomposition (selecting the db4 wavelet basis), the 200 MHz - 300 MHz frequency band is divided into 256 sub-bands, and a 128-dimensional time-frequency feature vector with the top 50% energy ratio is extracted. The frequency resolution of this method is 4 times higher than that of the traditional FFT algorithm (reaching 1.25 MHz), effectively separating the discharge signal from the 2.4 GHz narrowband interference generated by equipment such as frequency converters;

[0077] AE signal processing: Using the compressive sensing theory, a sparse observation matrix (Gaussian random matrix) is constructed in the 20 kHz - 100 kHz frequency band, and the effective signal is reconstructed through the orthogonal matching pursuit algorithm (OMP). Experiments show that this method can improve the signal-to-noise ratio from 15 dB to 28 dB, and the ability to suppress 50 Hz power frequency interference is 30% higher than that of the traditional band-pass filter.

[0078] In the actual application scenario, the TEV signal processing also uses the joint algorithm of wavelet packet and compressive sensing. First, the 10 kHz - 100 MHz frequency band is finely divided through wavelet packet decomposition, and the key time-frequency features are extracted. Then, the compressive sensing is used to reconstruct the signal, effectively removing the background noise interference generated by the complex electromagnetic environment on-site, increasing the signal-to-noise ratio of the TEV signal to 30 dB, and ensuring the accuracy and reliability of the transient earth voltage signal analysis.

[0079] In an optional embodiment, the spatial dimension decoupling unit of the decoupling module 12 adopts the ultra-wideband time difference of arrival positioning method. A three-dimensional array is formed by at least 3 ultra-high frequency UHF sensors and 2 ultrasonic AE sensors, and the discharge point coordinates are calculated based on the time difference of arrival TDoA of the signals.

[0080] It can be understood that in a complex electromagnetic environment, it is difficult for traditional positioning methods to accurately capture the spatial position of abnormal discharge points. Therefore, this embodiment adopts the ultra-wideband time difference of arrival positioning method, and a three-dimensional array layout is formed by at least 3 UHF (ultra-high frequency) sensors and 2 AE (acoustic emission) sensors. To ensure the positioning accuracy, the sensor spacing is strictly controlled at ≥0.5 meters. This spacing can not only avoid signal interference between sensors but also meet the requirement of synchronous acquisition of multi-source signals. The positioning principle is based on the time difference of arrival (TDoA) technology, and the three-dimensional coordinates of the discharge point are calculated by establishing a positioning equation system:

[0081]

[0082] Among them, (x i , y i , z i ) is the three-dimensional coordinate of the i-th sensor, and its accuracy is calibrated at the millimeter level by a laser rangefinder, and the error is controlled at ≤1 mm; Δti-j is the time difference for the signal to reach the i-th and j-th sensors, c is the propagation speed of electromagnetic waves in air, with a value of 2.998×10 8 m / s. The Newton iteration method is used to solve the non-linear equations, with the initial value set as the geometric center of the distribution box, and the iteration termination condition is that the Euclidean distance between two adjacent solutions < 0.1 mm.

[0083] It can be understood that the Newton iteration method is adopted in this embodiment. To improve the iteration efficiency and convergence, the geometric center of the distribution box is set as the initial iteration value, and the Euclidean distance between two adjacent solutions < 0.1 mm is used as the iteration termination condition. In the on-site measurement of the 10 kV ring main unit, this method successfully improves the positioning accuracy to ±0.25 m, achieving a six-fold improvement in accuracy compared with the traditional TDOA algorithm (±1.5 m). Compared with the traditional algorithm, this embodiment effectively reduces the influence of multipath effects and signal attenuation on the positioning result by optimizing the sensor array configuration and iteration strategy, providing technical support for the accurate traceability of potential abnormal discharge hazards.

[0084] In an optional embodiment, the edge computing node adopts a heterogeneous computing platform of an FPGA unit + a DSP unit;

[0085] The FPGA unit realizes signal preprocessing, including baseline subtraction, digital filtering, and analog-to-digital conversion;

[0086] The DSP unit runs a fusion model of the lightweight convolutional neural network MobileNetV3 and the long short-term memory network LSTM to perform pattern recognition on the discharge signal.

[0087] Among them, the FPGA unit (optional XC7Z020 chip) realizes 12-bit ADC analog-to-digital conversion (sampling rate 1 GS / s), supports 8-channel synchronous acquisition; performs baseline subtraction (based on the sliding window median method, window size 1024 points) and digital filtering (50th-order FIR low-pass filter, cut-off frequency 200 MHz), with a processing delay ≤ 15 ms; transmits the preprocessed data to the DSP unit through the AXI high-speed bus;

[0088] The DSP unit (optional TMS320C6678 chip) runs the MobileNetV3-LSTM fusion model: MobileNetV3 extracts the spatial features of the discharge signal (convolution kernel size 3×3, 16 layers of depthwise separable convolution), and the LSTM network processes the time series features (number of neurons in the hidden layer 128, time step 32); The model quantization technology (16-bit fixed-point number operation) is adopted, the single-sample inference time ≤ 15ms, and the recognition accuracy reaches 96.2% (the test set contains 100,000 discharge samples); Through the ping-pong buffer mechanism (alternate reading and writing of double buffers) and the pipeline architecture (three-level pipeline of data preprocessing, feature extraction, and model inference), the overall response time of the edge node is stabilized at 75ms, meeting the requirements of the IEC61850 standard for real-time early warning (≤ 100ms).

[0089] In an optional embodiment, the FPGA unit preprocesses the signal through a dynamic baseline calibration algorithm, specifically including:

[0090] Construct a Gaussian mixture model based on the historical data of the sensor as the signal baseline, and calculate the Mahalanobis distance between the current signal and the baseline in real time;

[0091] When the Mahalanobis distance exceeds 3 times the standard deviation, an abnormal early warning is triggered, and at the same time, the early warning threshold is dynamically adjusted through the particle swarm optimization algorithm PSO.

[0092] Specifically, based on the historical data of the sensor in the past 7 days, a mixture model containing 3 Gaussian components is constructed as the signal baseline:

[0093]

[0094] Among them, π k is the weight coefficient of the k-th Gaussian component, satisfying and π k ≥0, and its numerical size directly reflects the relative importance of this component in the mixture model; μ k is the mean parameter of the k-th Gaussian component, representing the central position of this Gaussian distribution in the feature space; σ k is the standard deviation parameter of the k-th Gaussian component, used to measure the degree of dispersion of the data around the mean. The smaller the σ k value, the more concentrated the data distribution, and vice versa, the more dispersed the distribution;

[0095] Calculate the Mahalanobis distance between the current signal x t and the baseline in real time:

[0096] d 2 (x t )=(x t -μ) T Σ -1 (xt -μ)

[0097] When 2 (x t )>3 2 The warning threshold is adjusted dynamically through the particle swarm optimization algorithm (PSO, number of particles 30, number of iterations 50), which reduces the false alarm rate from 20% of the traditional fixed threshold method to 4.3%;

[0098] When the Mahalanobis distance exceeds the preset dynamic threshold upper limit, an early warning is triggered. At the same time, the early warning threshold is dynamically adjusted through the particle swarm optimization algorithm (PSO, number of particles 30, number of iterations 50), so that the false alarm rate is reduced from 20% of the traditional fixed threshold method to 4.3%. Through this optimization mechanism, the system can adaptively adjust the early warning threshold according to the changing characteristics of real-time monitoring data. After actual test verification, compared with the false alarm rate of 20% of the traditional fixed threshold method, after adopting the dynamic baseline calibration algorithm of this embodiment, the system false alarm rate is significantly reduced to 4.3%, which effectively improves the accuracy and reliability of abnormal discharge hidden danger monitoring.

[0099] In an optional embodiment, the cloud platform is also used to construct a graph neural network GNN device health model, and the graph neural network GNN device health model includes:

[0100] The input layer integrates discharge signal characteristics, temperature field data, load current curves and equipment inventory information;

[0101] The graph convolution layer captures the nonlinear correlation between multi-source data through the attention mechanism and generates a three-dimensional fault tree map;

[0102] The output layer predicts the remaining life of the insulation material.

[0103] Specifically, the input layer integrates four types of data: discharge signal characteristics (128-dimensional time-frequency vector), temperature field data (16-point infrared temperature measurement matrix), load current curve (sampling frequency 1kHz, duration 10s), and equipment ledger information (operating years, insulation material type, historical maintenance records);

[0104] The graph convolution layer constructs a device-sensor-defect association graph, where the nodes include a distribution box (1), sensors (5 types × N), and defect types (6 main discharge modes). The association weights between nodes are calculated through the attention mechanism. For example, the weights of the cable joint node and the adjacent UHF sensor node are determined by the signal transmission distance and attenuation coefficient, generating a three-dimensional fault tree map containing spatial association information.

[0105] The output layer uses a fully connected neural network to predict the remaining life (RUL) of insulation materials, and the loss function is the mean square error (MSE).

[0106] In an optional embodiment, the sensor array 11 adopts a magnetic adsorption type quick installation structure, the deployment time of a single sensor is ≤5 minutes, and electrical isolation is achieved through a three-stage signal isolation circuit:

[0107] The first stage uses a DCDC high isolation power module (such as B0505S-1W), with an isolation voltage of 2000Vrms, to suppress common mode interference at the power supply end;

[0108] The second stage uses an optocoupler chip (such as 6N137) to achieve signal transmission isolation, with an isolation bandwidth of up to 10MHz;

[0109] The third stage suppresses common mode interference through an impedance matching circuit (such as a 50Ω terminal resistor), making the crosstalk noise between sensors ≤10μV, meeting the electromagnetic compatibility standard of GB / T17626.3.

[0110] Among them, the DCDC high isolation power module selects the B0505S-1W model. This module adopts advanced isolation technology, has a high isolation voltage of 2500Vrms, can effectively cut off the common mode interference path at the power supply end, ensure the purity of the power supply system, provide reliable power guarantee for the stable operation of the backend circuit, and realize the physical isolation between power input and output through the built-in magnetic isolation transformer and filter circuit, reducing the electrical risks brought by power grid fluctuations and surge impacts;

[0111] The high-speed optocoupler chip adopts the 6N137 model. Its core advantage is to achieve electrical isolation during signal transmission, with an isolation bandwidth as high as 10MHz, which can meet the distortion-free transmission requirements of high-speed digital signals. This chip converts electrical signals into optical signals for transmission through an optoelectronic conversion mechanism, and then restores them to electrical signals through photosensitive elements, completely eliminating the electrical connection between the input and output ends, effectively resisting the impact of electromagnetic interference on signal integrity, and ensuring the accuracy and stability of data transmission;

[0112] The impedance matching circuit adopts a 50Ω terminal resistor design. Based on the transmission line theory, this resistor can effectively match the characteristic impedance of the sensor and the transmission line, suppress the crosstalk noise between sensors to an extremely low level of 8μV. After detection and verification, this index fully meets the strict requirements of the electromagnetic harassment limit in the GB / T17626.3 electromagnetic compatibility standard, ensuring that the system can still maintain high-precision and low-noise monitoring performance in a complex electromagnetic environment.

[0113] In addition, it should be noted that the combination methods of the technical features in this case are not limited to the combination methods recorded in the claims of this case or the combination methods recorded in the specific embodiments. All the technical features recorded in this case can be freely combined or combined in any way, unless contradictions occur between them.

[0114] In this embodiment, high-precision acquisition of multi-physical field signals is performed through the sensor array 11, significantly improving the ability to capture weak discharge signals and the long-term stability of the sensors, reducing the frequency of manual calibration. Through the decoupling module 12, interference decoupling of multi-sensor signals in the time, frequency, and space dimensions can be performed, effectively separating the discharge signal from the noise, achieving precise positioning and pattern recognition of abnormal discharges, and significantly reducing the false alarm rate. Through the edge-cloud collaborative computing architecture 13, real-time processing and intelligent analysis of signals can be performed. The edge node realizes millisecond-level abnormal early warning, and the cloud platform constructs a device health model by combining multi-source data, provides long-term trend prediction, supports predictive maintenance strategies, and improves the operation reliability and operation and maintenance efficiency of the distribution box.

[0115] Embodiment 2

[0116] For a better understanding of the above Embodiment 1, please refer to Figure 2 , the present invention provides specific operation steps for an adaptive intelligent distribution box abnormal discharge hidden danger precise monitoring system, and the steps include:

[0117] S1, Sensor deployment: Install 3 groups of UHF sensors (location: 1 in the top bus compartment, 2 in the cable compartment), 2 groups of AE sensors (near the cable joints), 1 group of TEV sensors (on the cabinet surface), 1 ultraviolet sensor (inside the observation window), and 2 infrared sensors (on both sides of the insulator) inside the ring main unit (size 1.2m×0.8m×1.5m). All sensors are fixed through magnetic adsorption brackets, and the total deployment time is 15 minutes;

[0118] S2, Signal acquisition and decoupling: Within the high-frequency time slice (T1 = 100 μs), the UHF sensor collects an electromagnetic pulse signal with a center frequency of 250 MHz and an amplitude of 800 mV, and the ultraviolet sensor detects a radiation signal with a wavelength of 260 nm; within the low-frequency time slice (T2 = 1 ms), the AE sensor collects a sound wave signal of 50 μV, and the infrared sensor detects a local temperature rise of 1.2°C. Through the processing of the decoupling module 12, it is determined that the discharge point is located at the left joint of the cable compartment (coordinates: x = 0.3 m, y = 0.5 m, z = 0.2 m), with an error of 0.15 meters from the actual discharge position;

[0119] S3, Edge node processing: The FPGA unit completes signal analog-to-digital conversion and baseline subtraction. The MobileNetV3-LSTM model of the DSP unit identifies it as "discharge due to poor contact of cable joints". The edge node triggers a local audible and visual alarm within 72 ms and sends a warning message (including discharge type, positioning coordinates, signal characteristics) to the operation and maintenance platform through the 4G network;

[0120] S4, Cloud Analysis and Prediction: The cloud GNN model combines the historical data of this ring main unit (it has been in operation for 5 years and has had 3 bolt loosening failures), predicts its remaining insulation life to be 87 days, guides the operation and maintenance department to formulate a targeted maintenance plan, and avoids short-circuit accidents caused by poor contact.

[0121] In this embodiment, high-precision acquisition of multi-physical field signals is performed through the sensor array 11, significantly improving the ability to capture weak discharge signals and the long-term stability of the sensors, reducing the frequency of manual calibration. Through the decoupling module 12, the interference of multi-sensor signals in the time, frequency, and space dimensions can be decoupled, effectively separating the discharge signal from the noise, realizing the accurate positioning and pattern recognition of abnormal discharges, significantly reducing the false alarm rate. Through the edge-cloud collaborative computing architecture 13, real-time processing and intelligent analysis of signals can be carried out. The edge node realizes millisecond-level abnormal early warning, and the cloud platform constructs a device health model by combining multi-source data, provides long-term trend prediction, supports predictive maintenance strategies, and improves the operation reliability and operation and maintenance efficiency of the distribution box.

[0122] It should be noted that the above-listed are only specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and there are many similar variations. All deformations directly derived or associated by those skilled in the art from the content disclosed in the present invention shall fall within the protection scope of the present invention.

[0123] The above is only a preferred embodiment of the present invention and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An accurate monitoring system for hidden dangers of abnormal discharge in an adaptive intelligent distribution box, characterized in that, Including: A sensor array for collecting multi - physical - field signals inside the distribution box; A decoupling module for performing interference decoupling and feature extraction on the multi - physical - field signals in the time, frequency, and space dimensions; An edge - cloud collaborative computing architecture, including an edge computing node and a cloud platform. The edge computing node is used to perform real - time signal processing and abnormal warning, and the cloud platform is used to construct a device health model and output long - term trend prediction.

2. The precise monitoring system for hidden dangers of abnormal discharge of the adaptive intelligent distribution box according to claim 1, wherein The sensor array includes a very high frequency (VHF) UHF sensor, an ultrasonic acoustic emission (AE) sensor, a transient earth voltage (TEV) sensor, an ultraviolet sensor, and an infrared sensor. Quantum dot sensitized thin films are deposited on the surfaces of the sensitive elements of each sensor; Among them, the quantum dot sensitized thin film of the VHF UHF sensor uses CdSe material, and the quantum dot sensitized thin film of the AE sensor uses a ZnO quantum dot and piezoelectric ceramic heterostructure.

3. The adaptive intelligent power distribution box abnormal discharge hidden danger precise monitoring system according to claim 2, wherein, The time - dimension decoupling of the decoupling module adopts a dual - time - slice multiplexing protocol (DTMP), which divides each monitoring period into a high - frequency time - slice and a low - frequency time - slice; The high - frequency time - slice is used to activate the VHF UHF sensor and the ultraviolet sensor, with a duration of 50 μs - 200 μs; The low - frequency time - slice is used to activate the ultrasonic AE sensor, the transient earth voltage (TEV) sensor, and the infrared sensor, with a duration of 200 μs - 2 ms; Among them, the decoupling module avoids the time - domain overlap of signals from different sensors through a time - division activation mechanism.

4. The accurate monitoring system for hidden dangers of abnormal discharge of the adaptive intelligent distribution box according to claim 2, wherein, The frequency - dimension decoupling of the decoupling module includes: A wavelet packet decomposition sub - unit for performing 8 - layer wavelet packet decomposition on the VHF UHF sensor signal and extracting a 128 - dimensional time - frequency feature vector in the 200 MHz - 300 MHz frequency band; A compressive sensing reconstruction sub - unit for sparsely representing the ultrasonic AE sensor signal and reconstructing the effective signal in the 20 kHz - 100 kHz frequency band to suppress the power - frequency interference below 20 kHz; The decoupling module improves the signal - to - noise ratio to more than 25 dB through a frequency - domain feature separation algorithm.

5. The adaptive intelligent power distribution box abnormal discharge hidden danger precise monitoring system according to claim 2, characterized in that, The space - dimension decoupling unit of the decoupling module adopts an ultra - wideband time - difference - of - arrival (TDoA) positioning method. A three - dimensional array is formed by at least 3 VHF UHF sensors and 2 ultrasonic AE sensors, and the discharge point coordinates are calculated based on the time - difference of arrival (TDoA) of the signals. The mathematical model of the positioning algorithm is: Among them, (x i , y i , z i ) are the three-dimensional coordinates of the i-th sensor, Δt i-j is the time difference for the signal to reach the i-th and j-th sensors, and c is the propagation speed of electromagnetic waves in air.

6. The accurate monitoring system for hidden dangers of abnormal discharge of the adaptive intelligent distribution box according to claim 1, characterized in that, The edge computing node adopts a heterogeneous computing platform of an FPGA unit + a DSP unit; The FPGA unit realizes signal pre - processing, including baseline subtraction, digital filtering, and analog - to - digital conversion; The DSP unit runs a fusion model of a lightweight convolutional neural network (MobileNetV3) and a long short - term memory network (LSTM) to perform pattern recognition on the discharge signal.

7. The accurate monitoring system for hidden dangers of abnormal discharge of the adaptive intelligent distribution box according to claim 6, characterized in that, The FPGA unit pre - processes the signal through a dynamic baseline calibration algorithm, specifically including: Constructing a Gaussian mixture model based on the historical data of the sensor as the signal baseline and calculating the Mahalanobis distance between the current signal and the baseline in real - time; When the Mahalanobis distance exceeds 3 times the standard deviation, an abnormal warning is triggered, and at the same time, the warning threshold is dynamically adjusted through a particle swarm optimization algorithm (PSO).

8. The precise monitoring system for hidden dangers of abnormal discharge of the adaptive intelligent distribution box according to claim 1, characterized in that The cloud platform is also used to construct a graph neural network (GNN) device health model. The graph neural network (GNN) device health model includes: The input layer fuses the discharge signal features, temperature field data, load current curve, and equipment ledger information; The graph convolution layer captures the non-linear correlations between multi-source data through the attention mechanism and generates a three-dimensional fault tree graph; The output layer predicts the remaining life of the insulating material.

9. The accurate monitoring system for hidden dangers of abnormal discharge of the adaptive intelligent distribution box according to claim 2, wherein, The sensor array is built-in with a self-calibration module; The self-calibration module includes a temperature compensation circuit and a quantum dot fluorescence intensity monitoring unit; The quantum dot fluorescence spectrum monitoring unit collects the emission wavelength shift of the quantum dot-sensitized thin film in real time and establishes a temperature drift error correction model: S(T) = S0·(1 + k·ΔT) Among them, S(T) is the sensor sensitivity at temperature T, S0 is the reference sensitivity at 25°C, k is the temperature drift coefficient of quantum dots (k ≤ 5×10 -4 / °C), and ΔT is the temperature deviation; The temperature compensation circuit stabilizes the operating temperature of the sensor's sensitive element within the range of 25°C ± 0.1°C through a feedback control loop composed of a thermoelectric cooler and a temperature sensor to compensate for the temperature drift effect of the quantum dot material.

Images