Distribution box electric leakage monitoring and early warning system

Abstract

The invention specifically relates to a power distribution box electric leakage monitoring and early warning system, and relates to the technical field of power equipment monitoring. A heterogeneous signal processing and fusion module; and an intelligent diagnosis and three-dimensional positioning engine module. According to the invention, a multi-mode fusion sensing technology is adopted, early insulation degradation characteristics such as electric field distortion and radio frequency pulses are captured, a prediction model with an attention mechanism is combined, an insulation degradation risk value in the next 24 hours is output, and temperature and humidity compensation and anti-interference filtering design are matched, so that false alarm and missing alarm caused by power grid harmonic waves and electromagnetic radiation are effectively avoided; the crossing from post-event alarm to pre-event prediction is realized, and sufficient disposal time is reserved for operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, and in particular to a leakage current monitoring and early warning system for distribution boxes. Background Technology

[0002] As the core equipment for power distribution and control in a power system, the insulation performance of the distribution box directly determines the safety and stability of the power supply. When insulation deteriorates inside the distribution box (such as due to moisture, carbonization, or aging), it can easily lead to leakage faults. If timely monitoring and early warning are not provided, it may result in serious accidents such as equipment burnout, fire, or even electric shock to personnel.

[0003] Existing leakage current monitoring technology for distribution boxes mainly relies on a single current transformer to collect leakage current signals, which has the following significant drawbacks: The early warning is delayed, and the alarm can only be triggered when the insulation deteriorates and forms a stable leakage path and the leakage current reaches the threshold. It cannot capture the subtle characteristics of the early stage of insulation deterioration. Poor positioning accuracy can only determine that there is leakage in the circuit, but cannot accurately locate the fault point inside the box, which causes great inconvenience to operation and maintenance. The diagnostic depth is insufficient, making it impossible to distinguish the physical type of leakage fault (such as corona discharge, air gap discharge, etc.), which makes it difficult to support targeted operation and maintenance; fourth, the anti-interference capability is weak, and it is easily affected by power grid harmonics and environmental electromagnetic radiation, leading to false alarms or missed alarms.

[0004] Therefore, developing a distribution box monitoring system capable of early warning, accurate location, and in-depth diagnosis of leakage faults has become an urgent technical problem to be solved in the field of power operation and maintenance. Summary of the Invention

[0005] The purpose of this invention is to provide a leakage current monitoring and early warning system for distribution boxes in order to solve the above-mentioned problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A leakage current monitoring and early warning system for distribution boxes, comprising: The multimodal sensing array module is configured to use a hybrid deployment of distributed arrays and integrated probes to simultaneously acquire electric field distortion, radio frequency / ultra-high frequency pulses, and current / temperature / vibration signals; The heterogeneous signal processing and fusion module is configured to achieve deep fusion of heterogeneous signals and generate joint feature vectors through noise reduction by dedicated conditioning circuits, multi-dimensional feature extraction and spatiotemporal alignment technology. The intelligent diagnosis and 3D localization engine module is configured to use a Bi-LSTM prediction model, CNN and SVM classifiers, and electric field-RF fusion localization algorithm to achieve early prediction of insulation degradation, fault type classification, and 3D localization.

[0007] Preferably, the multimodal sensing array module specifically includes: Space electric field distortion sensing submodule: A preset number of MEMS electric field sensors are deployed in a grid-like layout on the inner wall of the distribution box. After the system is powered on for the first time, it continuously collects electric field data under healthy conditions, stores the amplitude and phase information of each sensor, and generates a reference three-dimensional electric field distribution map. During normal operation, the sensor array collects data in real time, and each frame of data is compared with the corresponding position of the reference spectrum to calculate the electric field distortion coefficient.

[0008] Preferably, the method further includes a radio frequency / ultra-high frequency detection and injection submodule: Five microstrip patch antennas are deployed in a layout of four corners and the center, located at the four corners and the center of the top of the distribution box; Passive mode: The antenna array receives high-frequency electromagnetic pulses in space in real time, amplifies them with a low-noise amplifier, filters out clutter, and then converts them into digital signals. Spectral analysis of digital signals is performed to extract the energy proportion, pulse repetition rate, and pulse amplitude distribution of characteristic frequency bands of partial discharge; the discharge area is preliminarily determined by the first acquisition antenna and the signal arrival time difference. Active mode: The signal generator produces a sinusoidal scanning signal, which is amplified by the power amplifier and then injected into the bus through the coupler; The reflected and transmitted signals are captured by the receiving end of the coupler, the reflection coefficient and transmission attenuation are calculated, and an impedance-frequency response curve is generated. When the insulation resistance decreases, the impedance to ground decreases, the reflection coefficient increases and the transmission attenuation decreases. By comparing the response curve under healthy conditions, the change in insulation resistance is quantified. Passive mode output: high-frequency pulse time-domain waveform, spectrum, PRPD map, energy percentage of characteristic frequency bands, and preliminary location area; Active mode output: incident / reflected / transmitted signal waveforms, impedance-frequency response curves, reflection coefficient-frequency curves, and estimated insulation resistance values.

[0009] Preferably, the method further includes a multi-parameter fusion probe submodule: Core sensor integration: miniature Rogowski coil, high-precision temperature sensor, micro-vibration sensor, UHF antenna contact; It is nested onto busbars and cables using a snap-fit ​​structure; The system controls all sensors to collect data synchronously, and the sampling trigger signal is provided by a unified system clock to ensure that current, temperature and vibration data at the same timestamp are aligned. The raw data collected is preliminarily processed, including harmonic analysis of current signals, moving average filtering of temperature signals, and peak detection of vibration signals. Output: Synchronization data frames, exception trigger data.

[0010] Preferably, the heterogeneous signal processing and fusion module specifically includes: Dedicated circuit for signal conditioning and noise reduction: Modular conditioning circuits are designed to address the signal characteristics of different types of sensors, with each module independently packaged. Electric field sensor signal conditioning circuit: Input stage: Employs a high-impedance operational amplifier; Amplification stage: Employs an instrumentation amplifier; Filtering stage: A second-order active low-pass filter is used; Output stage: Employs a voltage follower; RF signal conditioning circuit: Passive signal channels: broadband low-noise amplifier, programmable bandpass filter, variable gain amplifier; Active signal injection channel: signal generator, power amplifier, directional coupler, reflected signal receiving amplifier; Fusion probe signal conditioning circuit: Current signal path: The Rogowski coil output signal is converted into a voltage signal by an integrator, and then conditioned by a low-pass filter and an amplifier; Temperature signal channel: driven by a constant current source, the voltage signal is amplified by an instrumentation amplifier; Vibration signal channel: preamplifier, bandpass filter, peak hold circuit, to extract vibration peak signal.

[0011] Preferably, the method further includes feature extraction: Electric field data feature extraction: Spatial characteristics: Calculate the spatial gradient of the electric field distribution, the distortion of equipotential surfaces, and the area of ​​the distorted region; Temporal characteristics: Extract the mean, standard deviation, maximum value, rate of change, and peak factor of the electric field distortion coefficient; Radio frequency data feature extraction: Spectral characteristics: Calculate the energy proportion of characteristic frequency bands, spectral centroid, and spectral flatness; Pulse characteristics: Extract pulse repetition rate, pulse amplitude distribution entropy, pulse rise time, and pulse width; Feature extraction from fused probe data: Current characteristics: Calculate the effective value, peak value, waveform factor, harmonic distortion rate, amplitude and phase of each harmonic; Temperature characteristics: Extract average temperature, maximum temperature, rate of temperature change, and temperature fluctuation amplitude; Vibration characteristics: The characteristic frequencies, harmonic content, and vibration energy of the vibration signal are extracted by FFT transformation; Correlation characteristics: Calculate the correlation coefficient between current and temperature, and the time difference between the peak vibration and the peak current.

[0012] Preferably, the intelligent diagnosis and three-dimensional positioning engine module specifically includes: Early prediction model for insulation degradation: A bidirectional LSTM network is used, which contains 3 hidden layers, each with 128 neurons. The input dimension is 10-dimensional and the output dimension is 1-dimensional. An attention mechanism is added after the LSTM layer to automatically focus on key features that have a significant impact on the prediction results, and the risk value is mapped to the 0~1 range. Dataset construction: Measured data at different insulation degradation stages were collected. Each sample group contained 24 hours of time-series data. The sample size was expanded and divided into training set, validation set, and test set in a ratio of 7:2:1. The training parameters were trained using the Adam optimizer. The model outputs an insulation degradation risk curve for the next 24 hours by inputting the latest 2-hour time series features every 5 minutes. The slope of the risk value curve is analyzed using a linear fitting algorithm; Fault type precise classifier: Feature input layer: Input a 15-dimensional fused feature vector; Feature enhancement layer: A 1D convolutional layer is used to extract local features, and the activation function is ReLU; Feature aggregation layer: Max pooling is used to reduce feature dimensionality, and then fully connected layers are used for feature aggregation; Classification output layer: An SVM classifier is used to output the probability distribution of 6 types of faults; Dataset construction: Collect measured data of 6 types of faults, collect a preset number of samples for each type of fault, and each sample contains 15-dimensional fusion features; Training process: The model was trained using the 5-fold cross-validation method.

[0013] Preferably, the method further includes: Location based on solving the inverse problem of electric field array: Electric field propagation model: Establish a three-dimensional finite element model of electric field propagation inside the distribution box, taking into account conductor shape, dielectric constant of insulating material, and air medium; Inverse problem solution: Using the electric field distortion vector matrix as input, the inverse problem is solved by regularized least squares algorithm to obtain the equivalent charge distribution of insulation defects; The centroid coordinates of the equivalent charge distribution are the preliminary location coordinates of the fault point; Time-difference positioning based on radio frequency antenna arrays: The time difference of high-frequency pulses arriving at the five antennas was calculated using a cross-correlation algorithm. Based on the known coordinates of the antenna array, a set of time difference of arrival (TDOA) positioning equations is established; the positioning coordinates of the fault point are obtained by solving the set of equations; and the final positioning coordinates are calculated by weighted average fusion.

[0014] Preferably, the method further includes: The graded early warning and self-calibration output module is configured to use a four-level dynamic early warning strategy to release fault information through multiple channels, combined with regular self-calibration, sensor health monitoring and degradation operation mechanism.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention uses multimodal fusion sensing technology to capture early characteristics of insulation degradation such as electric field distortion and radio frequency pulses. Combined with a prediction model with attention mechanism, it outputs the insulation degradation risk value for the next 24 hours. With the addition of temperature and humidity compensation and anti-interference filtering design, it effectively avoids false alarms and missed alarms caused by power grid harmonics and electromagnetic radiation. It realizes the leap from post-event alarm to pre-event prediction, and reserves sufficient time for operation and maintenance to deal with the situation.

[0016] 2. This invention achieves fault location through finite element model inversion and Newton-Raphson algorithm, and intuitively marks the fault in a three-dimensional digital model. At the same time, based on a multimodal feature classifier, it can accurately distinguish faults such as insulation flashover and air gap discharge, and output the physical nature and typical characteristics of the fault. This solves the problem of blind troubleshooting in traditional operation and maintenance, allowing operation and maintenance personnel to quickly locate the fault, clarify the direction of handling, and shorten the troubleshooting time. Attached Figure Description

[0017] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0018] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0019] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0020] Example 1

[0021] Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0022] Appendix Figure 1 This invention provides a structural block diagram of a power distribution box leakage monitoring and early warning system, which shows the connection relationship between the multimodal sensor array module and the hierarchical early warning and self-calibration output module, and marks the main functional interaction flow of each module.

[0023] In this embodiment, it includes: The multimodal sensing array module is configured to use a hybrid deployment of distributed arrays and integrated probes to simultaneously acquire multi-dimensional signals such as electric field distortion, radio frequency / ultra-high frequency pulses, current / temperature / vibration, etc., providing accurate raw data for subsequent processing; Specifically, it includes: Space electric field distortion sensing submodule: It adopts a customized MEMS electric field sensor with a chip size of ≤3mm×3mm×1mm, a measurement range of 0~100kV / m, an amplitude measurement accuracy of ±1%FS, a phase measurement accuracy of ±0.5°, a response time of ≤10μs, a power supply voltage of 3.3V, and a power consumption of ≤5mW. Deployment scheme: Deploy a preset number of sensors in a grid layout on the inner wall of the distribution box (adjustable according to the box size). Deploy one sensor every 15cm above the busbar, and in areas with dense cable bundles, deploy one sensor every 10cm. The sensors are fixed with insulating brackets, with a distance of ≥5cm from the conductor to avoid signal interference caused by mechanical contact. Auxiliary components: Each sensor integrates a miniature temperature and humidity compensation unit (measurement range 0~100%RH, accuracy ±2%RH) to counteract the effects of ambient temperature and humidity on electric field measurement.

[0024] After the system is powered on for the first time, it enters calibration mode and continuously collects electric field data under healthy conditions for 30 minutes. It stores the amplitude and phase information of each sensor at a frequency of 10ms / frame and generates a reference three-dimensional electric field distribution map through a three-dimensional interpolation algorithm. The map resolution is 1cm×1cm×1cm. At the same time, the current ambient temperature and humidity are recorded as a reference benchmark. During normal operation, the sensor array collects data in real time at a frequency of 5ms / frame. Each frame of data is compared with the corresponding position in the reference spectrum to calculate the electric field distortion coefficient K (K=|measured value-reference value| / reference value). When K is greater than a preset threshold (≥3%), it is marked as a potential anomaly point. The electric field distortion vector (including magnitude, direction, and range of action) is calculated using a vector synthesis algorithm. The specific formula is as follows: ;in, Let i be the electric field measured by the i-th sensor. As the baseline value, The sensor position unit vector; Digital filtering algorithms (Kalman filtering, moving average filtering) are used to eliminate false signals caused by power grid harmonics and external electromagnetic radiation. The size of the filtering window can be adaptively adjusted according to the actual interference intensity.

[0025] It also includes a radio frequency / ultra-high frequency detection and injection submodule: Microstrip patch antennas are used, each with dimensions ≤5cm×5cm×0.5cm, operating frequency band 30MHz~1.5GHz, gain ≥2dBi, VSWR ≤1.5, and noise figure ≤3dB. Five antennas are deployed in a four-corner and center layout, located at the four corners and center of the top of the distribution box, with a distance of ≥10cm between the antennas and the conductor, and connected to the signal processing unit via coaxial cables (characteristic impedance 50Ω). Active injection coupler: It adopts a through-hole coupling structure, is compatible with Φ10~50mm busbars, has a coupling degree of -20dB~-10dB, an operating frequency band of 100kHz~10MHz, a withstand voltage rating of ≥10kV, an insulation resistance of ≥100MΩ, and an injection signal power of ≤10mW, so as to avoid affecting the normal operation of the power grid. Signal switching switch: adopts radio frequency relay, with switching time ≤100μs and isolation ≥60dB, to realize rapid switching between passive detection and active injection modes (switching cycle can be set to 1~10 minutes). Working principle: Passive mode: Signal acquisition: The antenna array receives high-frequency electromagnetic pulses in space in real time. After being amplified by a low-noise amplifier (adjustable gain, 20~40dB), the signal is filtered for noise by a bandpass filter (30MHz~1.5GHz) and then converted into a digital signal by a high-speed ADC (sampling rate ≥1GSps, resolution ≥12bit). Feature analysis: Perform spectrum analysis on the digital signal (using FFT algorithm, spectrum resolution ≤10kHz) to extract the energy proportion, pulse repetition rate (PRPD), and pulse amplitude distribution of the characteristic frequency bands of partial discharge (e.g., 30~300MHz for corona discharge, 300MHz~1.5GHz for air gap discharge); identify the discharge area by the first acquisition antenna and the signal arrival time difference. Active mode: Signal injection: The signal generator generates a sine wave scanning signal (frequency range 1MHz~10MHz, step size 10kHz, dwell time of 10ms at each frequency point), which is amplified by the power amplifier and then injected into the bus through the coupler; Response monitoring: The reflected and transmitted signals are captured by the receiver of the coupler, and the reflection coefficient Γ (Γ = reflected signal amplitude / incident signal amplitude) and transmission attenuation A (A = 20lg(transmitted signal amplitude / incident signal amplitude)) are calculated to generate an impedance-frequency response curve. When the insulation resistance decreases, the impedance to ground decreases, the reflection coefficient increases, and the transmission attenuation decreases. By comparing the response curve under healthy conditions, the change in insulation resistance is quantified. Passive mode output: high-frequency pulse time-domain waveform, spectrum (amplitude-frequency curve), PRPD spectrum, energy proportion of characteristic frequency bands, and preliminary location area; Active mode output: incident / reflected / transmitted signal waveforms, impedance-frequency response curves, reflection coefficient-frequency curves, and estimated insulation resistance values; Data interface: Employs Ethernet interface (TCP / IP protocol), with a data transmission rate ≥10Mbps, supporting real-time upload and local caching.

[0026] It also includes a multi-parameter fusion probe submodule: Core sensor integration: Miniature Rogowski coil: inner diameter Φ5~10mm, measurement range 0~500A (RMS), bandwidth 10Hz~1MHz, accuracy ±0.5%FS, phase error ≤1°, coil turns 1000, encapsulated with polyimide film; High-precision temperature sensor: PT1000 platinum resistance thermometer, measuring range -40℃~200℃, accuracy ±0.1℃, response time ≤100ms, with thermally conductive silicone tightly attached to the measuring point; Micro-vibration sensor: Employs a piezoelectric accelerometer with a measurement range of 0~50g, sensitivity of 100mV / g, frequency response of 10Hz~1kHz, and resolution ≤0.01g; UHF antenna contacts: Uses a small loop antenna with a diameter ≤3mm, operating frequency band 300MHz~1.5GHz, and gain ≥0dB; Packaging and Interface: The overall package size of the probe is ≤15mm×20mm×8mm, using ceramic-epoxy resin composite insulation material (dielectric strength ≥20kV / mm, thermal conductivity ≥1.5W / (m・K)); it integrates a high-speed MCU (main frequency ≥100MHz) to achieve synchronous data sampling and preprocessing; the interface uses a waterproof aviation plug, supports hot-swapping, and the cable length is ≤2m (can be customized to extend to 5m). Working principle: By using a snap-fit ​​structure to nest on conductors such as busbars and cables, or by using thermally conductive adhesive to attach to key parts such as joints and switches, the temperature sensor is ensured to be in close contact with the measurement point, and the vibration sensor can capture the conductor vibration signal. Synchronous sampling: The MCU controls each sensor to synchronously collect data at a sampling frequency of 1kHz~10kHz (configurable). The sampling trigger signal is provided by the system's unified clock to ensure that the current, temperature, and vibration data at the same timestamp are strictly aligned (synchronization error ≤1μs). Data preprocessing: The MCU performs preliminary processing on the acquired raw data, including harmonic analysis of the current signal (extracting the 1st to 13th harmonics), moving average filtering of the temperature signal (window size 10 frames), and peak detection of the vibration signal. The preprocessed data is stored in frame format (each frame contains 200 sampling points, frame interval 100ms). Output: Synchronization data frame: includes timestamp, current RMS / peak / harmonic content, instantaneous / average temperature, vibration acceleration peak / characteristic frequency, and UHF signal amplitude (optional); Abnormal trigger data: When a parameter exceeds a preset threshold, the sampling frequency is automatically increased to 10kHz, data is continuously collected for 5 seconds and marked as abnormal event data, and is uploaded to the processing module first.

[0027] The heterogeneous signal processing and fusion module is configured to achieve deep fusion of heterogeneous signals through noise reduction by dedicated conditioning circuits, multi-dimensional feature extraction and spatiotemporal alignment technology, and generate high-value joint feature vectors. Specifically, it includes: Dedicated circuit for signal conditioning and noise reduction: Modular conditioning circuits are designed to address the signal characteristics of different types of sensors. Each module is independently packaged for easy maintenance and expansion. Electric field sensor signal conditioning circuit: Input stage: High-impedance operational amplifiers (input impedance ≥ 10¹²Ω, bias current ≤ 1pA) are used to buffer the signal and avoid load effects; Amplification stage: Employs an instrumentation amplifier (adjustable gain, 10~100 times), with a common-mode rejection ratio ≥120dB@50Hz, suppressing power frequency interference; Filtering stage: A second-order active low-pass filter is used, with a cutoff frequency of 1kHz and an attenuation slope of -40dB / decade to eliminate high-frequency noise. Output stage: Employs a voltage follower with an output impedance ≤10Ω to ensure stable signal transmission to the ADC; RF signal conditioning circuit: Passive signal channels: wideband low-noise amplifier (30MHz~1.5GHz, gain 30dB, noise figure ≤2.5dB), programmable bandpass filter (center frequency adjustable, bandwidth 10MHz~100MHz), variable gain amplifier (gain adjustable from 0 to 20dB); Active signal injection channel: signal generator (frequency accuracy ±1ppm, amplitude accuracy ±0.5dB), power amplifier (output power 10mW, efficiency ≥30%), directional coupler (coupling degree -20dB, isolation degree ≥60dB), reflected signal receiving amplifier (gain 20dB). Fusion probe signal conditioning circuit: Current signal channel: The Rogowski coil output signal is converted into a voltage signal by an integrator (integration time constant 1ms), and then conditioned by a low-pass filter (cutoff frequency 1kHz) and an amplifier (gain 20~50 times); Temperature signal channel: driven by a constant current source (drive current 1mA, accuracy ±0.1%), the voltage signal is amplified by an instrumentation amplifier (gain 100~1000 times) to eliminate the influence of lead resistance; Vibration signal channel: preamplifier (gain 40dB, noise figure ≤5dB), bandpass filter (10Hz~1kHz), peak hold circuit, to extract vibration peak signal; Common design features: all circuits adopt a multi-layer PCB design, with power and ground layers separated, and critical signal lines using differential routing and shielding; equipped with an electromagnetic shielding shell (shielding effectiveness ≥80dB@30MHz~1GHz) to avoid electromagnetic interference inside and outside the module.

[0028] It also includes feature extraction: Electric field data feature extraction: Spatial characteristics: Calculate the spatial gradient of the electric field distribution, the equipotential surface distortion (calculated based on the phase difference between adjacent sensors, distortion = standard deviation of phase difference / reference phase difference), and the area of ​​the distorted region (extract the region K≥3% through a threshold segmentation algorithm and calculate the volume in three-dimensional space). Among them, spatial gradient ; in, This is the gradient operator, which represents the calculation of the spatial rate of change of a physical quantity and outputs a vector. , , Electric field strength Components along the three axes of a rectangular coordinate system; , , These are the three-dimensional rectangular space coordinates inside the distribution box; electric field Component along Partial derivatives in direction; electric field Component along Partial derivatives in direction; electric field Component along (partial derivatives in direction); Temporal characteristics: Extract the mean, standard deviation, maximum value, rate of change, and peak factor (peak value / effective value) of the electric field distortion coefficient to reflect the stability and development trend of the distortion; Radio frequency data feature extraction: Spectral characteristics: Calculate the energy proportion of characteristic frequency bands (e.g., energy in the 30~300MHz band / total energy), spectral centroid ( ;in, For frequency points, Power at that frequency point), spectral flatness ( ); Pulse characteristics: Extract pulse repetition rate (number of pulses per unit time), pulse amplitude distribution entropy (reflecting the uniformity of amplitude distribution), pulse rise time (time from 10% amplitude to 90% amplitude), and pulse width (duration of amplitude ≥ 50% peak value). Feature extraction from fused probe data: Current characteristics: Calculate the effective value, peak value, waveform factor (effective value / average value), harmonic distortion rate (THD), amplitude and phase of each harmonic; Temperature characteristics: Extract average temperature, maximum temperature, rate of temperature change, and temperature fluctuation range (maximum value - minimum value). Vibration characteristics: The characteristic frequencies (the three frequencies with the largest amplitude), harmonic content, and vibration energy of the vibration signal are extracted through FFT transformation. ,in (for acceleration signals) Correlation characteristics: Calculate the correlation coefficient between current and temperature ( ;in, Let be the covariance of current and temperature; For current Standard deviation; For temperature The standard deviation of the vibration peak and the time difference between the peak current and the peak current (reflecting the causal relationship); Spatiotemporal alignment and data fusion: The system adopts the IEEE 1588PTP precision clock synchronization protocol, with a master clock accuracy of ≤10ns. Each sensor and processing module achieves clock synchronization via Ethernet, with a synchronization error of ≤1μs. For sensors not connected to Ethernet (such as fusion probes), synchronization pulse signals are transmitted via CAN bus to ensure consistent timestamps. The three-dimensional coordinates of each sensor are pre-stored (based on the CAD model of the power distribution box, with an accuracy of ±1mm), and a sensor ID-coordinate mapping table is established; during data processing, the signal of each sensor is bound to the corresponding coordinate to achieve precise matching of spatial positions; Data fusion algorithm: Data layer fusion: Time-domain superposition: The synchronized fused probe current waveform is superimposed with the vibration waveform to generate a current-vibration joint waveform, which intuitively shows the timing relationship between the two. Spatial superposition: The electric field distortion region is spatially superimposed with the antenna position captured by the radio frequency pulse, and the overlapping region is extracted by Boolean operation to narrow down the potential fault range; Data completion: For missing data from some sensors (such as temporary communication interruptions), the K-nearest neighbor interpolation algorithm (K=3) is used to complete the data to ensure data integrity; Feature layer fusion: Feature dimensionality reduction: Principal component analysis (PCA) algorithm is used to reduce the dimensionality of high-dimensional feature vectors (initial dimension ≥ 50 dimensions), retaining principal components with a cumulative contribution rate ≥ 95%, reducing the dimension to 10~15 dimensions, thereby reducing the computational complexity of subsequent diagnostic algorithms; Feature weighted fusion: Weights of each feature are calculated based on the information gain ratio (IGR) (the sum of the weights is 1). Features with higher IGRs have larger weights. A joint feature vector is generated by weighted summation. ,in, The weights of the i-th type of features are... Let i be the feature vector of the i-th class; Fusion Validation: The DS evidence theory is used to fuse and validate the diagnostic results of different features. The basic probability assignment function (BPA) is calculated, and the final fusion confidence is obtained through the synthesis rule, thereby improving the reliability of diagnosis.

[0029] The intelligent diagnosis and 3D localization engine module is configured to use a Bi-LSTM prediction model, CNN and SVM classifiers and electric field-RF fusion localization algorithm to achieve early prediction of insulation degradation, accurate classification of fault types and 3D localization. Specifically, it includes: Early prediction model for insulation degradation: A bidirectional LSTM (Bi-LSTM) network is used, which contains 3 hidden layers, each with 128 neurons. The input dimension is 10-dimensional (core features of electric field distortion and auxiliary features of temperature and humidity), and the output dimension is 1-dimensional (insulation degradation risk value in the next 24 hours, ranging from 0 to 1). Optimize the structure: Add an attention mechanism after the LSTM layer to automatically focus on key features that have a significant impact on the prediction results (such as the rate of change of electric field distortion) and improve prediction accuracy; the output layer uses the Sigmoid activation function to map the risk value to the 0~1 range; Dataset Construction: Measured data were collected at different insulation degradation stages (healthy, slightly damp, moderately carbonized, and severely degraded). 1000 samples were collected for each stage, and each sample contained 24 hours of time-series data (sampling frequency 5ms / frame). The sample size was expanded to 5000 sets using data augmentation techniques (time stretching, noise addition, and feature perturbation), and divided into training, validation, and test sets in a 7:2:1 ratio. Training parameters: Adam optimizer is used, the initial learning rate is 0.001, decays exponentially (decay rate 0.95 / epoch), batch size = 32, number of training epochs = 100, and the loss function is mean squared error (MSE). Real-time prediction: Input the latest 2-hour time series features every 5 minutes, and the model outputs the insulation degradation risk value curve for the next 24 hours; when the risk value is ≥0.5, an L1 level attention alert is triggered; when the risk value is ≥0.7, an L2 level warning is triggered. Trend Analysis: The slope of the risk value curve is analyzed using a linear fitting algorithm. When the slope is ≥0.02 / h, it is judged as rapid deterioration, and the warning level is upgraded; when the slope is ≤0.005 / h, it is judged as slow deterioration, and the current warning level is maintained. Fault type precise classifier: Feature input layer: Input a 15-dimensional fused feature vector (features after PCA dimensionality reduction); Feature enhancement layer: A 1D convolutional layer (kernel size 3, number of layers 64, stride 1) is used to extract local features, and the activation function is ReLU; Feature aggregation layer: Max pooling layer (pooling kernel size 2, stride 2) is used to reduce feature dimensionality, and then feature aggregation is performed through a fully connected layer (128 neurons); Classification output layer: An SVM classifier (with RBF kernel, penalty coefficient C=10, gamma=0.1) is used to achieve multi-class classification, outputting the probability distribution of 6 types of faults (summing up to 1). Dataset construction: Measured data of 6 types of faults (insulation surface flashover, internal air gap discharge, metal tip corona, mechanical loosening arc, cable insulation aging, and joint overheating) were collected. A preset number of samples were collected for each type of fault, and each sample contained 15-dimensional fusion features. Training process: The model is trained using 5-fold cross-validation to avoid overfitting; the kernel parameters (C and gamma) of the SVM are optimized through grid search to maximize the cross-validation accuracy. Real-time classification: The model outputs the probability of each type of fault when the latest fused feature vector is input every 100ms; when the probability of a certain type of fault is ≥0.8, it is judged as that type of fault; when the highest probability is between 0.5 and 0.8, it is judged as a suspected fault, and further confirmed by combining radio frequency pulse characteristics. Fault tracing: Based on the classification results, output the typical characteristics of this type of fault (such as the energy proportion of metal tip corona in the 30~300MHz frequency band ≥0.6) to help maintenance personnel understand the nature of the fault.

[0030] Location based on solving the inverse problem of electric field array: Electric field propagation model: A three-dimensional finite element model (FEM) of electric field propagation inside the distribution box is established, taking into account factors such as conductor shape, dielectric constant of insulating material, and air medium. The model mesh resolution is ≤5mm×5mm×5mm. Solving the inverse problem: Using the electric field distortion vector matrix as input, the inverse problem is solved using a regularized least squares algorithm (regularization parameter λ=0.01) to obtain the equivalent charge distribution of the insulation defect. ; Location calculation: The centroid coordinates of the equivalent charge distribution are the preliminary location coordinates of the fault point. ; Time-difference positioning based on radio frequency antenna arrays: Time difference measurement: The time difference of high-frequency pulses arriving at the five antennas is calculated using a cross-correlation algorithm; Positioning equation establishment: Based on the known coordinates of the antenna array, a set of time difference of arrival (TDOA) positioning equations is established: ;in, Let i be the coordinates of the i-th antenna. For reference antenna coordinates, At the speed of light, Let i be the time difference between the i-th antenna and the reference antenna; Equation Solution: The Newton-Raphson iterative algorithm is used to solve the system of equations to obtain the location coordinates of the fault point. ; Integration of verification and visualization: The final positioning coordinates are calculated using a weighted average fusion method (with preset electric field positioning weights and radio frequency positioning weights). ; Based on the CAD model of the distribution box, a 1:1 scale 3D digital model is constructed, and the final positioning coordinates are mapped onto the model. Suspected fault points are marked with red highlighted spheres (radius = positioning error). The model supports rotation, scaling, and sectioning operations, and maintenance personnel can intuitively view the fault location through the host computer software.

[0031] The tiered early warning and self-calibration output module is configured to use a four-level dynamic early warning strategy to release fault information through multiple channels. Combined with regular self-calibration, sensor health monitoring, and degradation operation mechanisms, it ensures the long-term stable and reliable operation of the system. Specifically, it includes: Dynamic hierarchical early warning strategy: The basic threshold is set based on the national standards GB / T14048.1-2022 and GB / T7251.1-2022 (e.g., the leakage current threshold is 10mA). By statistically analyzing 1,000 sets of measured fault data, the characteristic thresholds of each warning level were determined (e.g., the electric field distortion coefficient threshold for L2 warning is 5%, and the radio frequency pulse repetition rate threshold is 3 times / hour). It supports user-defined thresholds (adjustable within ±20%) to adapt to the needs of different application scenarios.

[0032] Level 4 warning includes: Level 1 (Attention): Triggering conditions: The 24-hour risk value output by the insulation degradation prediction model is ≥0.5, and there are no abnormal signals such as electric field distortion or radio frequency pulses; Output method: pop-up prompts in the host computer software, background system log recording, no audible or visual alarms; Warning message: Equipment number [XXX], insulation performance is showing a slow downward trend (current risk value [XX]), it is recommended to re-inspect every 72 hours, with special attention to [XX area].

[0033] Level 2 (Warning): Triggering conditions: electric field distortion coefficient ≥5%, or radio frequency pulse repetition rate ≥3 times / hour, or prediction model risk value ≥0.7, with no overheating or vibration abnormalities; Output methods: PC software pop-up, sound prompts, mobile APP push notifications, and background system alarm records; Notification: Equipment number [XXX], early insulation degradation detected! Fault type: [Specific type], Suspected area: [X,Y,Z]±5cm, It is recommended to arrange inspection within 48 hours, Inspection focus: [Fault-related components, such as A-phase busbar connectors].

[0034] Level 3 (Alarm): Triggering conditions: RF pulse repetition rate ≥ 5 times / hour, or temperature detected by fusion probe ≥ 85℃ (or exceeding ambient temperature by 50℃), or current harmonic distortion rate ≥ 10% accompanied by abnormal vibration; Output methods: local audible and visual alarm (red LED flashing, frequency 2Hz; buzzer volume ≥85dB), forced pop-up window in host computer software, mobile APP push, SMS notification, and linkage with industrial IoT platform; Notification: Equipment No. [XXX], insulation fault has occurred! Fault type: [Specific type], Precise location: [X,Y,Z]±4cm, Fault severity: Moderate. It is recommended to stop the machine immediately for inspection and repair to prevent the fault from escalating.

[0035] Level 4 (Critical): Triggering conditions: The fusion probe detects a leakage current ≥10mA (national standard threshold), or the multimodal characteristics (electric field distortion coefficient ≥10% + radio frequency pulse repetition rate ≥10 times / hour + temperature ≥100℃ + vibration acceleration ≥10g) are met simultaneously; Output methods: local strong audible and visual alarm (red LED light stays on + buzzer sounds continuously), full-screen alarm on host computer software, mobile APP push + SMS + telephone notification, output remote trip signal (passive contact, capacity AC220V / 5A). Warning message: Equipment number [XXX], Critical fault! Short circuit risk imminent! Fault type: [Specific type], Precise location: [X,Y,Z]±4cm. Immediately disconnect power, do not approach the equipment, and contact professional personnel for emergency handling.

[0036] System self-calibration and health management: Calibration cycle: Default is 24 hours / cycle, users can set it to 12 hours / cycle or 48 hours / cycle; Calibration steps: The system automatically switches to calibration mode and pauses the warning function (data acquisition is retained). The radio frequency active injection module injects standard high-frequency signals (1MHz, 5MHz, and 10MHz, with an amplitude of 1Vpp) into the bus. The signal processing module acquires reflected and transmitted signals and generates calibration response curves. The amplitude deviation ΔA and phase deviation Δφ are calculated by comparing the curve with the preset standard response curve (calibrated at the factory and stored in non-volatile memory). If ΔA≤±5% and Δφ≤±3°, the calibration is deemed qualified, and the calibration timestamp is updated. If ΔA>±5% or Δφ>±3°, the gain and filtering parameters of the signal conditioning circuit are automatically adjusted, and the data is re-acquired and compared. If the adjustment still fails to meet the standard, it is marked as a calibration anomaly, triggering an L1 level attention prompt.

[0037] Sensor health monitoring: Monitoring indicators: background noise (e.g., background noise of electric field sensor ≥0.5kV / m is considered abnormal), signal output stability (signal amplitude fluctuation ≤±2% for 10 consecutive frames is considered normal), communication status (three consecutive communication failures are considered offline). Monitoring frequency: Real-time monitoring (communication status), and monitoring of background noise and output stability every hour; Anomaly Handling: Mark abnormal sensors (e.g., sensor ID [XX], abnormal background noise) and prompt maintenance in the background system; for offline sensors, attempt automatic reconnection (3 times / minute, lasting 5 minutes), and trigger L1 level attention prompt if reconnection fails.

[0038] Degradation operation mechanism: Sensor partial failure: When ≤30% of the sensors fail, the system automatically adjusts the feature extraction and fusion algorithm (e.g., removing features from failed sensors and increasing the weight of features from valid sensors) to maintain the warning and positioning functions (accuracy will decrease slightly, e.g., positioning error ≤6cm). Partial module failure: When a channel of the signal processing module fails, it automatically switches to a backup channel (each channel is equipped with one backup channel); when a part of the intelligent diagnostic engine algorithm fails, a simplified algorithm is enabled (such as replacing CNN and SVM classifiers with decision trees) to ensure that the core functions are available; Degradation Notice: Add a message to the warning information indicating that the system has been degraded and that maintenance is recommended as soon as possible, reminding users to handle the fault in a timely manner.

[0039] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0040] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0041] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0042] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0043] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0044] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0045] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0046] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0047] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0048] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A leakage current monitoring and early warning system for a distribution box, characterized in that, include: The multimodal sensing array module is configured to use a hybrid deployment of distributed arrays and integrated probes to simultaneously acquire electric field distortion, radio frequency / ultra-high frequency pulses, and current / temperature / vibration signals; The heterogeneous signal processing and fusion module is configured to achieve deep fusion of heterogeneous signals and generate joint feature vectors through noise reduction by dedicated conditioning circuits, multi-dimensional feature extraction and spatiotemporal alignment technology. The intelligent diagnosis and 3D localization engine module is configured to use a Bi-LSTM prediction model, CNN and SVM classifiers, and electric field-RF fusion localization algorithm to achieve early prediction of insulation degradation, fault type classification, and 3D localization.

2. The leakage current monitoring and early warning system for a distribution box according to claim 1, characterized in that, The multimodal sensor array module specifically includes: Space electric field distortion sensing submodule: A predetermined number of MEMS electric field sensors are deployed in a grid pattern on the inner wall of the distribution box. After the system is powered on for the first time, it continuously collects electric field data under healthy conditions, stores the amplitude and phase information of each sensor, and generates a reference three-dimensional electric field distribution map. During normal operation, the sensor array collects data in real time, and each frame of data is compared with the corresponding position of the reference spectrum to calculate the electric field distortion coefficient.

3. The leakage current monitoring and early warning system for a distribution box according to claim 2, characterized in that, It also includes a radio frequency / ultra-high frequency detection and injection submodule: Five microstrip patch antennas are deployed in a layout of four corners and the center, located at the four corners and the center of the top of the distribution box; Passive mode: The antenna array receives high-frequency electromagnetic pulses in space in real time, amplifies them with a low-noise amplifier, filters out clutter, and then converts them into digital signals. Spectral analysis of digital signals is performed to extract the energy proportion, pulse repetition rate, and pulse amplitude distribution of characteristic frequency bands of partial discharge; the discharge area is preliminarily determined by the first acquisition antenna and the signal arrival time difference. Active mode: The signal generator produces a sinusoidal scanning signal, which is amplified by the power amplifier and then injected into the bus through the coupler; The reflected and transmitted signals are captured by the receiving end of the coupler, the reflection coefficient and transmission attenuation are calculated, and an impedance-frequency response curve is generated. When the insulation resistance decreases, the impedance to ground decreases, the reflection coefficient increases and the transmission attenuation decreases. By comparing the response curve under healthy conditions, the change in insulation resistance is quantified. Passive mode output: high-frequency pulse time-domain waveform, spectrum, PRPD map, energy percentage of characteristic frequency bands, and preliminary location area; Active mode output: incident / reflected / transmitted signal waveforms, impedance-frequency response curves, reflection coefficient-frequency curves, and estimated insulation resistance values.

4. The leakage current monitoring and early warning system for a distribution box according to claim 1, characterized in that, It also includes a multi-parameter fusion probe submodule: The core sensor integration includes a miniature Rogowski coil, a high-precision temperature sensor, a micro-vibration sensor, and UHF antenna contacts; it is nested onto busbars and cables via a snap-fit ​​structure. The system controls all sensors to collect data synchronously, and the sampling trigger signal is provided by a unified system clock. The raw data collected is preliminarily processed, including harmonic analysis of current signals, moving average filtering of temperature signals, and peak detection of vibration signals. Output synchronous data frames and exception trigger data.

5. The leakage current monitoring and early warning system for a distribution box according to claim 1, characterized in that, The heterogeneous signal processing and fusion module specifically includes: Dedicated circuit for signal conditioning and noise reduction: Modular conditioning circuits are designed to address the signal characteristics of different types of sensors, with each module independently packaged. Electric field sensor signal conditioning circuit: Input stage: Employs a high-impedance operational amplifier; Amplification stage: Employs an instrumentation amplifier; Filtering stage: A second-order active low-pass filter is used; Output stage: Employs a voltage follower; RF signal conditioning circuit: Passive signal channels: broadband low-noise amplifier, programmable bandpass filter, variable gain amplifier; Active signal injection channel: signal generator, power amplifier, directional coupler, reflected signal receiving amplifier; Fusion probe signal conditioning circuit: Current signal path: The Rogowski coil output signal is converted into a voltage signal by an integrator, and then conditioned by a low-pass filter and an amplifier; Temperature signal channel: driven by a constant current source, the voltage signal is amplified by an instrumentation amplifier; Vibration signal channel: preamplifier, bandpass filter, peak hold circuit, to extract vibration peak signal.

6. The leakage current monitoring and early warning system for a distribution box according to claim 5, characterized in that, It also includes feature extraction: Electric field data feature extraction: Spatial characteristics: Calculate the spatial gradient of the electric field distribution, the distortion of equipotential surfaces, and the area of ​​the distorted region; Temporal characteristics: Extract the mean, standard deviation, maximum value, rate of change, and peak factor of the electric field distortion coefficient; Radio frequency data feature extraction: Spectral characteristics: Calculate the energy proportion of characteristic frequency bands, spectral centroid, and spectral flatness; Pulse characteristics: Extract pulse repetition rate, pulse amplitude distribution entropy, pulse rise time, and pulse width; Feature extraction from fused probe data: Current characteristics: Calculate the effective value, peak value, waveform factor, harmonic distortion rate, amplitude and phase of each harmonic; Temperature characteristics: Extract average temperature, maximum temperature, rate of temperature change, and temperature fluctuation amplitude; Vibration characteristics: The characteristic frequencies, harmonic content, and vibration energy of the vibration signal are extracted by FFT transformation; Correlation characteristics: Calculate the correlation coefficient between current and temperature, and the time difference between the peak vibration and the peak current.

7. The leakage current monitoring and early warning system for a distribution box according to claim 1, characterized in that, The intelligent diagnostic and 3D localization engine module specifically includes: Early prediction model for insulation degradation: A bidirectional LSTM network is used, which contains 3 hidden layers, each with 128 neurons. The input dimension is 10-dimensional and the output dimension is 1-dimensional. An attention mechanism is added after the LSTM layer to automatically focus on key features that have a significant impact on the prediction results, and the risk value is mapped to the 0~1 range. Dataset construction: Measured data at different insulation degradation stages were collected. Each sample group contained 24 hours of time-series data. The sample size was expanded and divided into training set, validation set, and test set in a ratio of 7:2:

1. The training parameters were trained using the Adam optimizer. The model outputs an insulation degradation risk curve for the next 24 hours by inputting the latest 2-hour time series features every 5 minutes. The slope of the risk value curve is analyzed using a linear fitting algorithm; Fault type precise classifier: Feature input layer: Input a 15-dimensional fused feature vector; Feature enhancement layer: A 1D convolutional layer is used to extract local features, and the activation function is ReLU; Feature aggregation layer: Max pooling is used to reduce feature dimensionality, and then fully connected layers are used for feature aggregation; Classification output layer: An SVM classifier is used to output the probability distribution of 6 types of faults; Dataset construction: Collect measured data of 6 types of faults, collect a preset number of samples for each type of fault, and each sample contains 15-dimensional fusion features.

8. The leakage current monitoring and early warning system for a distribution box according to claim 1, characterized in that, Also includes: Location based on solving the inverse problem of electric field array: Electric field propagation model: Establish a three-dimensional finite element model of electric field propagation inside the distribution box, taking into account conductor shape, dielectric constant of insulating material, and air medium; Inverse problem solution: Using the electric field distortion vector matrix as input, the inverse problem is solved by regularized least squares algorithm to obtain the equivalent charge distribution of insulation defects; The centroid coordinates of the equivalent charge distribution are the preliminary location coordinates of the fault point; Time difference location based on RF antenna array: the time difference of high-frequency pulses arriving at the five antennas is calculated by cross-correlation algorithm; Based on the known coordinates of the antenna array, a set of time difference of arrival (TDOA) positioning equations is established; the positioning coordinates of the fault point are obtained by solving the set of equations; and the final positioning coordinates are calculated by weighted average fusion.

9. A leakage current monitoring and early warning system for a distribution box according to claim 1, characterized in that, Also includes: The graded early warning and self-calibration output module is configured to use a four-level dynamic early warning strategy to release fault information through multiple channels, combined with regular self-calibration, sensor health monitoring and degradation operation mechanism.

Images