Intelligent lightning arrester for line and state diagnosis method thereof

Abstract

The invention provides an intelligent lightning arrester for a line. The intelligent lightning arrester comprises a lightning arrester body and an intelligent control box, the lightning arrester body is used for providing a low-impedance grounding channel for overcurrent when overvoltage caused by lightning or switching operation occurs on a line, so that the overvoltage is limited below a safety level, subsequent electrical equipment is protected from being damaged, the lightning arrester body recovers to an insulation state after the overvoltage disappears, and normal operation of a power system is guaranteed. The intelligent control box comprises an embedded intelligent sensing module and a diagnosis module, and the embedded intelligent sensing module is used for collecting operation state parameters of the lightning arrester body in real time; and the diagnosis module is used for evaluating the insulation state of the lightning arrester body by adopting a partial discharge mode recognition model according to the operation state parameters of the lightning arrester body, and predicting the aging trend of a valve plate unit of the lightning arrester body by adopting a valve plate aging trend prediction model. The lightning arrester disclosed by the invention has the dual advantages of compact structure and high-performance protection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lightning arresters, in particular to a line intelligent lightning arrester and a state diagnosis method thereof. BACKGROUND

[0002] Metal oxide arresters are key devices for protecting power system equipment from lightning overvoltage and operating overvoltage. With the development of ultra-high voltage and compact transmission lines, higher performance requirements are placed on lightning arresters: on the one hand, they are required to have higher protection characteristics (i.e. higher nonlinearity coefficient and gradient) to reduce the volume and save line corridor space; on the other hand, they are required to have state self-awareness capability to realize the transition from "periodic maintenance" to "condition-based maintenance".

[0003] Currently, traditional 500kV and above line lightning arresters have the following outstanding problems: (1) the electric field gradient is generally low (about 250-280V / mm), resulting in lightning arrester height exceeding 4 meters, large diameter, installation inconvenience and landscape impact; (2) there is a lack of effective built-in state monitoring means, and existing monitoring mostly uses external leakage current meters or periodic pre-testing, which cannot realize real-time sensing of internal valve piece aging, dampness and uneven current distribution of multiple column parallel connection; (3) the lightning arrester body will generate extremely high ground potential and strong electromagnetic interference, and the operating environment is harsh (large temperature difference, high electromagnetic intensity), so conventional electronic sensors are difficult to work stably for a long time in this environment.

[0004] Chinese patent CN116487137A discloses an "intelligent anti-aging device for lightning arrester", which combines the current mature online detection technology for lightning arrester aging, uses the automatic resistance changing method to reduce the charge rate of the protected lightning arrester, and delays the aging process of the lightning arrester that has already aged, but does not improve the structure and performance of the lightning arrester, and cannot realize real-time sensing of the state of the lightning arrester. SUMMARY

[0005] The purpose of the present application is to overcome the defects of the prior art, and provide a line intelligent lightning arrester, which comprises: a lightning arrester body and an intelligent control box; The lightning arrester body is used to provide a low-impedance ground path for overcurrent when overvoltage caused by lightning or switch operation occurs on the line, so as to limit the overvoltage below a safe level and protect the subsequent electrical equipment from being damaged, and restore to an insulating state after the overvoltage disappears, thereby ensuring the normal operation of the power system; The intelligent control box comprises an embedded intelligent sensing module and a diagnosis module, the embedded intelligent sensing module is used for collecting the operating state parameters of the lightning arrester body in real time, and the diagnosis module is used for evaluating the insulation state of the lightning arrester body by using a partial discharge mode recognition model and predicting the aging trend of the valve unit of the lightning arrester body by using a valve aging trend prediction model according to the operating state parameters of the lightning arrester body.

[0006] Further, the lightning arrester body comprises a valve unit and an insulation shell sleeved outside the valve unit, the valve unit is composed of a plurality of zinc oxide resistance pieces doped with Bi2O3 in series, the Bi2O3 doping concentration of each zinc oxide resistance piece gradually decreases from the edge to the center, and the insulation shell is made of high-temperature vulcanized silicone rubber material.

[0007] Further, the embedded intelligent sensing module comprises a temperature sensor, a leakage current sensor and a UHF partial discharge sensor, the leakage current sensor collects the leakage current signal of the lightning arrester body mixed with electromagnetic interference in real time, the UHF partial discharge sensor collects the UHF partial discharge signal of the lightning arrester body mixed with electromagnetic interference in real time, and the temperature sensor collects the ambient temperature in real time.

[0008] Further, the diagnosis module is provided with a noise suppression algorithm, a partial discharge mode recognition model and a valve aging trend prediction model, the noise suppression algorithm removes electromagnetic interference from the leakage current signal mixed with electromagnetic interference and the UHF partial discharge signal mixed with electromagnetic interference, to obtain a denoised leakage current signal and a denoised UHF partial discharge signal; The denoised UHF partial discharge signal is input into the partial discharge mode recognition model to obtain an insulation state recognition result of the lightning arrester body, and the insulation state recognition result is normal insulation function or a preset insulation defect type; The denoised leakage current signal is input into the valve aging trend prediction model to obtain a current aging degree quantitative value of the valve unit, and the valve aging trend prediction model outputs the change trend of the aging degree quantitative value of the valve unit in a future preset time window in combination with the ambient temperature and the ambient humidity.

[0009] Further, the specific method for removing electromagnetic interference from the leakage current signal mixed with electromagnetic interference and the UHF partial discharge signal mixed with electromagnetic interference by the noise suppression algorithm is as follows: The noise suppression algorithm comprises a reference preset noise sample library and an adaptive filter. The adaptive filter is configured to perform adaptive noise suppression based on a least mean square algorithm, the adaptive filter adjusts its weight coefficients in real time to minimize the output mean square error, and then suppresses the corresponding signal noise component related to the noise sample in the preset noise sample library from the leakage current signal mixed with electromagnetic interference and the ultra-high frequency partial discharge signal mixed with electromagnetic interference according to the real-time adjusted weight coefficients, and extracts the signal with a signal-to-noise ratio greater than a signal-to-noise ratio threshold from the leakage current signal mixed with electromagnetic interference as a denoising leakage current signal, and extracts the signal with a signal-to-noise ratio greater than a signal-to-noise ratio threshold from the ultra-high frequency partial discharge signal mixed with electromagnetic interference as a denoising ultra-high frequency partial discharge signal.

[0010] Further, the partial discharge pattern recognition model is obtained by the following method: The historical ultra-high frequency partial discharge signals in the historical ultra-high frequency partial discharge signal set are labeled, and the label is the insulation state recognition result. The labeled historical ultra-high frequency partial discharge signal set is divided into a training set and a validation set. The convolutional neural network is trained according to the training set and the validation set, and the partial discharge pattern recognition model is obtained.

[0011] Further, the specific method of inputting the denoising ultra-high frequency partial discharge signal into the partial discharge pattern recognition model to obtain the insulation state recognition result of the lightning arrester body is: The partial discharge pattern recognition model includes an input unit, a feature extraction unit, and a classification decision unit. The input unit receives the denoising ultra-high frequency partial discharge signal. The feature extraction unit includes a convolution layer and a pooling layer. The convolution layer is used to extract high-frequency features and phase distribution features from the partial discharge signal through a ReLU activation function. The pooling layer is used to reduce the dimensions of the high-frequency features and the phase distribution features through a maximum pooling strategy. The reduced high-frequency features and phase distribution features are concatenated into partial discharge features in the channel dimension. The classification decision unit includes a fully connected layer and a Softmax classifier. The fully connected layer linearly combines the partial discharge features through a weight matrix and a bias to generate a score vector. The Softmax classifier maps the score vector to the insulation state recognition result.

[0012] Further, the valve plate aging trend prediction model includes: A signal decomposition unit, a resistive current calculation unit, a health index calculation unit, and a trend prediction unit. The signal decomposition unit is used to preprocess the denoising ultra-high frequency partial discharge signal by wavelet transform, and decompose the denoising ultra-high frequency partial discharge signal after preprocessing into a fundamental component mixed with third harmonic noise by using an adaptive noise complete ensemble empirical mode decomposition algorithm. The resistive current calculation unit is used to extract a resistive current signal from a fundamental wave component doped with third harmonic noise by using a capacitive current compensation algorithm, and the resistive current signal is temperature corrected in amplitude by using the ambient temperature collected by the temperature sensor, the change rate of the peak value of the temperature corrected resistive current signal with time is calculated to obtain a resistive current change rate, and the ratio of the third harmonic noise amplitude to the fundamental wave amplitude in the resistive current signal is calculated to obtain a resistive current third harmonic content; The health index calculation unit is used to calculate the current health index of the valve piece unit by using a fuzzy comprehensive evaluation model according to the resistive current change rate, the resistive current third harmonic content and the partial discharge characteristic value, and the health index represents the aging degree quantitative value of the valve piece unit. The trend prediction unit is used to adopt a long short-term memory neural network to take a historical health index set as input and output the change trend of the health index in a future preset time window.

[0013] A state diagnosis method of an intelligent lightning arrester for a line comprises: When the lightning arrester body is subjected to overvoltage caused by lightning or switch operation on the line, the lightning arrester body provides a low-impedance ground channel for overcurrent, thereby limiting the overvoltage below a safe level and protecting subsequent electrical equipment from being damaged, and the lightning arrester body returns to an insulation state after the overvoltage disappears, thereby ensuring normal operation of the power system. The intelligent control box comprises an embedded intelligent sensing module and a diagnosis module, the embedded intelligent sensing module collects lightning arrester body operation state parameters in real time, and the diagnosis module evaluates the insulation state of the lightning arrester body by using a partial discharge mode recognition model and predicts the aging trend of the valve piece unit of the lightning arrester body by using a valve piece aging trend prediction model according to the lightning arrester body operation state parameters.

[0014] A computer program product comprises computer programs / instructions, which, when executed by a processor, implement the above-mentioned state diagnosis method of the intelligent lightning arrester for the line.

[0015] The present application has the following advantages: 1. The present application optimizes the electric field distribution inside the resistor disc by using the zinc oxide resistor disc with a Bi2O3 doping concentration that decreases from the edge to the axis, significantly improves the average electric field gradient (which can exceed 330 V / mm). This enables the radial size of the lightning arrester body to be reduced by 20%-30% under the premise of meeting the same voltage level and protection level, effectively saving material costs, reducing tower load, and facilitating installation in lines with limited space. At the same time, the gradient doping structure suppresses local electric field distortion, enhances the nonlinearity and energy absorption capacity of the valve piece, and enables the lightning arrester to have the dual advantages of compact structure and high-performance protection.

[0016] 2. Through the integration of temperature, leakage current, and ultra-high frequency partial discharge multi-dimensional sensors, the embedded intelligent sensing module can synchronously collect key multi-physical field parameters reflecting the operating state of the lightning arrester. In particular, the ultra-high frequency partial discharge sensor can effectively capture the weak discharge signals generated by early insulation defects, overcoming the shortcomings of traditional monitoring methods, such as being susceptible to electromagnetic interference and insufficient sensitivity, and laying a data foundation for early fault diagnosis.

[0017] 3. The noise suppression algorithm built in the diagnosis module innovatively combines a preset noise sample library and an adaptive filter based on the least mean square criterion. This algorithm can dynamically track and suppress complex electromagnetic interference mixed in the leakage current and partial discharge signals, adaptively extract pure fault characteristic signals with a required signal-to-noise ratio from a strong noise background, and greatly improve the accuracy and reliability of subsequent diagnostic analysis.

[0018] 4. The partial discharge pattern recognition model constructed by the convolutional neural network can automatically learn and extract deep features from the de-noised ultra-high frequency partial discharge signals, and complete end-to-end intelligent recognition of the insulation state. This method avoids the problem of relying on expert experience and strong subjectivity in traditional diagnosis, and can quickly and accurately classify the insulation state as normal or a specific defect type, greatly improving the automation level and recognition accuracy of diagnosis.

[0019] 5. The valve piece aging trend prediction model constructed by the present application accurately calculates the resistive current change rate and third harmonic content representing key aging characteristics through wavelet transform and adaptive noise complete ensemble empirical mode decomposition algorithm, and combines capacitance current compensation and temperature correction technology. Further, a fuzzy comprehensive evaluation model is used to fuse multiple feature parameters to calculate an intuitive health index, realizing accurate quantification of the aging degree. Finally, the long short-term memory neural network is used to learn the time series of the health index, realizing accurate prediction of the aging trend in the future preset time window, thereby changing the maintenance strategy of the lightning arrester from regular maintenance or post-maintenance to forward-looking predictive maintenance, effectively avoiding sudden failures, prolonging the service life of the equipment, and optimizing operation and maintenance resources.

[0020] 6. The present application deeply integrates advanced sensing technology, signal processing algorithm, and artificial intelligence model, and constructs an intelligent system integrating real-time monitoring, accurate diagnosis, and trend prediction. It is not only a lightning protection device, but also an intelligent line health management node, providing key technical support and data foundation for building smart grids and realizing condition-based maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The structure diagram of the intelligent lightning arrester for lines of the present application. DETAILED DESCRIPTION

[0022] In order to make the technical problems, technical solutions and beneficial effects to be solved in the present application more clear and explicit, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0023] Noun explanation Leakage current sensor: a precision measuring device for high-voltage electrical equipment state monitoring, its core function is to collect the micro leakage current signal of the metal oxide arrester (MOA) valve in real time and high precision, and convert it into a standard electrical signal for subsequent intelligent analysis unit processing.

[0024] UHF partial discharge sensor: a special sensor for online monitoring of the insulation state of high-voltage electrical equipment, its core function is to detect and collect the UHF electromagnetic wave signal in the frequency band of 300MHz to 3GHz excited by the internal partial discharge activity of insulation.

[0025] High temperature vulcanized rubber: a material based on high molecular weight (usually 400-800 million) polyorganosiloxane, adding reinforcing filler (such as fumed white carbon black), vulcanizing agent (such as organic peroxide) and other additives, and forming an elastomer by high temperature (100-300℃) heating vulcanization.

[0026] Convolutional neural network (CNN): a deep learning model designed specifically for processing grid topology data (such as images, time series signal spectrum). Its core idea is to efficiently extract features from low to high level by local connection, weight sharing and spatial down-sampling, and finally complete the classification or recognition task. It is usually composed of feature extractor (convolutional layer, pooling layer) and classifier (fully connected layer, Softmax) in series, and its working process is an end-to-end feature learning and mapping process.

[0027] Adaptive noise suppression based on LMS: An advanced signal processing technology, the core of which is to use the least mean square (LMS) adaptive algorithm to dynamically adjust the filter parameters to estimate and cancel the noise component in the signal in real time, so as to extract the effective signal with high signal-to-noise ratio from the strong interference environment. The algorithm system usually contains three basic components: reference input, derived from the preset power frequency interference and high frequency noise sample library, which contains typical noise patterns extracted from historical data or prior knowledge, providing noise samples that need to be suppressed for adaptive filter; Primary input, derived from the real original signal mixed with strong electromagnetic interference collected by leakage current sensor or ultra-high frequency partial discharge sensor; adaptive filter, the core of which is a set of weight coefficient vector, which is the "brain" and adjustable parameter of the filter.d(n)=s(n)+n0(n), where s(n) is the effective state signal we expect (such as real leakage current, partial discharge pulse), n0(n) is the mixed strong noise, d(n) is the Primary input, the adaptive filter processes the reference input x(n) to produce a noise estimate y(n), subtracts the filter output y(n) from the Primary input d(n) to get the error signal e(n)=d(n)-y(n), which is the purified output signal we expect, LMS algorithm uses error signal e(n) and reference input x(n) to update weight vector in real time according to the following rules: W(n+1)=W(n)+μ×e(n)×X(n), where μ is the step factor, which controls the convergence speed and stability, the purpose of this formula is to minimize the mean square value of error signal e(n). Through countless iterations, the weight vector W(n) will converge to an optimal value, at which time the filter output y(n) can best approximate the real noise n0(n) in the Primary input, so that the error signal e(n) is infinitely close to the pure signal s(n).

[0028] Capacitive current compensation algorithm: A method specially used to separate the resistive current representing the aging state of the valve piece from the total leakage current with high precision. Its essence is to offset the capacitive component in the leakage current through mathematical modeling and electrical characteristic analysis, so as to extract the pure resistive component and provide reliable data basis for state diagnosis. In the equivalent circuit of the arrester, the valve piece under operating voltage can be regarded as a parallel model of capacitance and non-linear resistance. The total leakage current contains: capacitive current, generated by the geometric capacitance of the arrester, leading the voltage by about 90°, and irrelevant to aging; resistive current, in phase with the voltage, directly reflecting the degradation of the non-linear characteristics of the valve piece, which is the core of diagnosis. The algorithm process is generally as follows: collect the phase of bus voltage and total current, construct a compensation vector with the same amplitude and opposite direction as the capacitive current, and get the resistive current by subtracting the compensation vector from the total current.

[0029] Wavelet transform: A powerful time-frequency analysis tool, its core idea is to use a short oscillating waveform that can be scaled and shifted, called wavelet, to scan the signal, thus revealing the local details of the signal in time and frequency. It is very suitable for analyzing non-stationary signals, that is, signals whose statistical properties change over time, such as partial discharge, vibration, speech and biomedical signals. The ultra-high frequency signal generated by partial discharge is a transient pulse, which is submerged in strong background electromagnetic noise. Wavelet transform can accurately locate the time when the discharge pulse occurs, and by setting a threshold, it can filter out the noise components with weak energy and retain the discharge components with strong energy, thus effectively improving the signal-to-noise ratio.

[0030] Complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN): An advanced and adaptive signal processing technique used to decompose complex non-stationary and nonlinear signals into a series of intrinsic mode functions with physical meaning. It is an improved algorithm of empirical mode decomposition (EMD), which effectively solves the mode mixing and reconstruction error problems of traditional EMD method by introducing adaptive white noise and complete reconstruction mechanism. In this invention, CEEMDAN works with wavelet transform to form a signal preprocessing pipeline: the leakage current full signal or ultra-high frequency partial discharge signal after denoising processing; the signal is adaptively decomposed into a series of intrinsic mode functions (IMF), including: fundamental component (power frequency component, about 50Hz), third harmonic component (150Hz component), other high-frequency harmonics and residual noise; the pure fundamental component and third harmonic component obtained by decomposition are delivered to the resistive current calculation unit for extracting resistive current peak value, active loss value and third harmonic content.

[0031] Fuzzy comprehensive evaluation model: A decision analysis method based on fuzzy mathematics (Fuzzy Logic), specifically designed to handle complex system evaluation problems involving uncertainty, fuzziness and multiple factor influence. Its core idea is to convert fuzzy, non-quantitative subjective evaluation into clear, quantifiable evaluation results through mathematical methods.

[0032] Long short-term memory neural network: Long Short-Term Memory, abbreviated as LSTM, is a special type of recurrent neural network (RNN) designed specifically to process and predict time series data.

[0033] Example 1 In high-voltage transmission lines, surge arresters are key devices for protecting electrical equipment from lightning overvoltage and operating overvoltage. However, traditional line surge arresters and their monitoring methods have many defects, making it difficult to meet the requirements of smart grid for device miniaturization, intelligentization and high reliability.

[0034] First, in terms of structural performance, the conventional lightning arrester is usually large in size and weight to achieve sufficient insulation strength and current-carrying capacity, which not only increases the mechanical load and installation difficulty of the tower, but also occupies valuable line space. The valve pieces of the conventional lightning arrester are usually made of uniformly doped zinc oxide resistance pieces, and the electric field distribution is uneven, which limits the further improvement of the electric field gradient, resulting in the sacrifice of protection performance when pursuing miniaturization.

[0035] Secondly, in terms of state monitoring and diagnosis, the conventional method mainly relies on regular inspection and preventive test, and cannot realize real-time perception of the running state. Even if some lightning arresters are equipped with simple leakage current monitoring function, due to the lack of effective anti-interference means, the signal-to-noise ratio of the collected signals is low, and it is difficult to extract the characteristic quantity truly reflecting the aging or insulation deterioration of the valve piece from the strong electromagnetic noise environment. In addition, the existing diagnosis methods mostly rely on threshold alarm or artificial experience judgment, lack of deep identification of potential faults and prediction ability of aging trend, and cannot realize the transition from regular maintenance to predictive maintenance, so the risk of sudden failure of equipment is high and the operation and maintenance cost is huge.

[0036] Based on the above reasons, the present application provides an intelligent lightning arrester for lines, as shown in the accompanying drawings, which comprises: Figure 1 The lightning arrester body 1, the connecting piece 2 and the intelligent control box 3 are electrically connected between the lightning arrester body 1 and the intelligent control box 3, and the connecting piece 2 physically fixes the lightning arrester body 1 and the intelligent control box 3; The lightning arrester body 1 comprises an insulating shell 5 and a valve piece unit 4 in the insulating shell 5, the valve piece unit 4 is composed of a plurality of zinc oxide resistance pieces doped with Bi2O3 in series, and the Bi2O3 doping concentration of the zinc oxide resistance piece gradually decreases from the edge to the center; The intelligent control box 3 comprises an embedded intelligent sensing module 6 and a diagnosis module 7, and the embedded intelligent sensing module 6 and the diagnosis module 7 are electrically connected, the embedded intelligent sensing module 6 is used for real-time acquisition of the running state parameters of the lightning arrester body 1, and the diagnosis module 7 is used for evaluating the insulation state of the lightning arrester body 1 and predicting the aging trend of the valve piece unit 4 according to the running state parameters of the lightning arrester body 1.

[0037] As a preferred embodiment, the insulating shell 5 is made of high-temperature vulcanized silicone rubber material, and the radial size is reduced by 20%-30% compared with the conventional lightning arrester of the same voltage level. The high-temperature vulcanized silicone rubber material has high electrical strength, high temperature resistance, aging resistance and ultraviolet resistance, which ensures the long-term stable operation of the insulating shell in harsh environments such as high and low temperature and humidity.

[0038] ​As a preferred embodiment, the Bi2O3 doping concentration of the edge of the zinc oxide resistor disc is 2.0-5.5wt% higher than that of the axis, ensuring that the average electric field gradient of the resistor disc is >330V / mm. By designing the inclined voltage-sharing ring, the axial electric field distribution is optimized, so that the insulation height is reduced to ≤2.8m (≥4m for the traditional structure) under the same current-carrying capacity, and the overall size is reduced by 20%-30% compared with the traditional product in combination with the compact silicon rubber composite insulator designed by optimizing the electric field.

[0039] By the doping difference in the specific concentration range (2.0-5.5wt%), the electric field gradient of the edge and the axis of the resistor disc is smoothly transitioned, avoiding electric field concentration and reducing the risk of partial discharge and breakdown. This design improves the non-linear characteristics of the resistor disc, so that the lightning arrester responds more quickly and has a lower residual voltage when protecting against overvoltage. The concentration gradient design reduces the thermal stress of the resistor disc, delays aging, and improves the long-term stability of the valve unit.

[0040] As a preferred embodiment, the embedded intelligent sensing module 6 includes a temperature sensor, a leakage current sensor, and a UHF partial discharge sensor. These sensors are protected by a multi-physical field protective package, ensuring that the sensors can work reliably for a long time in harsh environments of high voltage, strong electromagnetic interference, and high and low temperature (-40℃ to +85℃).

[0041] The application will be further described in detail below in combination with the embodiment of the 500kV AC line high-gradient small-size intelligent lightning arrester.

[0042] S1 Preparation of high-gradient valve unit: zinc oxide resistor discs with a diameter of about 76-80mm are selected, and the Bi2O3 doping concentration of the edge region is made to be 10.9wt% and the center region is 8.9wt% by a special process. The gradient of a single valve disc is tested to be 380V / mm. According to the target residual voltage level, the number of series-connected valve discs is determined.

[0043] S2 Compact insulating shell forming: a high-temperature vulcanized silicone rubber material is used to manufacture a compact insulating shell by a precision mold injection molding process; the shell is tested for electrical strength, weather resistance, and mechanical strength to ensure that it meets the requirements of high voltage, strong electromagnetic interference, and high and low temperature environments. The overall electric field of the lightning arrester is simulated by using a finite element software (such as ANSYS), the size of the voltage-sharing ring and the shape of the umbrella skirt are optimized, and finally the designed lightning arrester has a height of about 2.82 meters and a diameter of about 180mm, which is reduced by about 28% compared with the size of the traditional product.

[0044] S3 Embedded intelligent sensing module integration: The temperature sensor adopts a miniature fiber Bragg grating (FBG) temperature sensor; the leakage current sensor adopts a miniature broadband current sensor based on the Rogowski coil principle. Integration location: The FBG sensor is pasted on the side of the valve piece at the key position; the current sensor is embedded near the grounding lead of the valve piece string; the leakage current sensor is placed at the grounding side of the bottom of the surge arrester.

[0045] S4 Diagnosis module development: The hardware adopts a low-power ARM Cortex-M7 core processor, equipped with a high-precision ADC unit (analog-to-digital converter) and supports multi-channel synchronous data acquisition. The communication unit simultaneously integrates a LoRa wireless unit and a single-mode optical fiber interface. The adaptive filtering algorithm based on LMS is used to denoise the collected leakage current signal, and the above algorithm is integrated into the embedded system.

[0046] S5 Overall assembly and testing: The assembled 500kV intelligent surge arrester prototype needs to pass the following tests: (1) Electrical performance test: passed the standard specified steep wave, lightning wave, operating wave impulse test and power frequency withstand test in the high voltage test room, the measured protection characteristic meets the design requirement, the gradient is 352V / mm. (2) Sensor accuracy test: inject standard square wave current to test the current sensor, the accuracy is better than ±1%. (3) Communication and function test: in the simulated running state, the diagnosis module successfully receives the sensor data, executes the diagnosis algorithm, and uploads the "normal state" report to the background system through the optical fiber, and the LoRa communication as a backup link is also tested successfully.

[0047] By adopting gradient-doped valve piece group and inclined grading ring design, the electric field gradient of the surge arrester is significantly improved to >330V / mm, and the insulation height is reduced to ≤2.8m, the overall size is reduced by 20%-30% compared with traditional products, effectively saving the line corridor space and reducing the installation difficulty. The embedded intelligent sensing module integrates miniature temperature, leakage current and partial discharge sensors, and cooperates with the multi-physical field protection package to realize stable operation in high voltage, strong electromagnetic interference and high and low temperature harsh environment, solving the problem of lack of built-in state monitoring means of traditional surge arresters. The intelligent diagnosis system can process data on site and realize state evaluation and fault warning through intelligent algorithm, combined with LoRa and optical fiber dual-mode communication interface, supports IEC61850 standard protocol, uploads the result to the monitoring center in real time, promotes the transformation from "periodic maintenance" to "condition-based maintenance", and greatly improves the reliability of power grid operation.

[0048] Example 2 A state diagnosis method of an intelligent surge arrester for a line, comprising: The leakage current sensor collects the leakage current signal of the arrester body 1 mixed with electromagnetic interference in real time, the ultra-high frequency partial discharge sensor collects the ultra-high frequency partial discharge signal of the arrester body 1 mixed with electromagnetic interference in real time, and the temperature sensor collects the ambient temperature in real time; The diagnostic module 7 is built-in with a noise suppression algorithm, a partial discharge pattern recognition model, and a valve piece aging trend prediction model. The noise suppression algorithm removes electromagnetic interference from the leakage current signal mixed with electromagnetic interference and the ultra-high frequency partial discharge signal mixed with electromagnetic interference, obtaining a denoised leakage current signal and a denoised ultra-high frequency partial discharge signal. The denoised ultra-high frequency partial discharge signal is input into the partial discharge pattern recognition model to obtain an insulation state recognition result of the arrester body 1. The insulation state recognition result is normal insulation function or a preset insulation defect type (such as internal partial discharge, surface partial discharge, and surface flashover). The denoised leakage current signal is input into the valve piece aging trend prediction model to obtain a current aging degree quantization value of the valve piece unit 4, and the change trend of the aging degree quantization value of the valve piece unit 4 in a future preset time window is output in combination with the ambient temperature and the ambient humidity.

[0049] As a preferred embodiment, the specific method of the noise suppression algorithm for removing electromagnetic interference from the leakage current signal mixed with electromagnetic interference and the ultra-high frequency partial discharge signal mixed with electromagnetic interference is as follows: The noise suppression algorithm includes a reference preset noise sample library (including power frequency interference and high frequency interference) and an adaptive filter. The adaptive filter is configured to perform adaptive noise suppression based on the least mean square algorithm. The adaptive filter adjusts its weight coefficients in real time to minimize the output mean square error, and then suppresses the corresponding signal noise components related to the noise samples in the preset noise sample library from the leakage current signal mixed with electromagnetic interference and the ultra-high frequency partial discharge signal mixed with electromagnetic interference according to the real-time adjusted weight coefficients, and extracts signals with a signal-to-noise ratio greater than a signal-to-noise ratio threshold from the leakage current signal mixed with electromagnetic interference as a denoised leakage current signal, and extracts signals with a signal-to-noise ratio greater than a signal-to-noise ratio threshold from the ultra-high frequency partial discharge signal mixed with electromagnetic interference as a denoised ultra-high frequency partial discharge signal.

[0050] The adaptive filter adjusts the weight coefficients in real time through the least mean square algorithm to suppress power frequency and high frequency noise from the mixed signal. This improves the signal-to-noise ratio of the signal and reduces the influence of environmental electromagnetic interference. The algorithm extracts denoised signals with a signal-to-noise ratio greater than a threshold from the leakage current and ultra-high frequency partial discharge signals, providing a reliable data basis for subsequent diagnosis. The reference noise sample library allows the algorithm to adapt to different noise environments, improves robustness, and ensures effective operation under strong interference conditions.

[0051] As a preferred embodiment, the partial discharge pattern recognition model is obtained by the following method: The historical UHF partial discharge signals in the historical UHF partial discharge signal set are labeled with labels, which are insulation state recognition results. The labeled historical UHF partial discharge signal set is divided into a training set and a validation set. The convolutional neural network is trained according to the training set and the validation set to obtain the partial discharge pattern recognition model. The training process is a typical supervised training, which will not be described in detail in the present application.

[0052] The CNN can automatically extract features from the historical UHF partial discharge signals without manual feature engineering, improving the efficiency and accuracy of pattern recognition. Through supervised training, the model can accurately classify insulation states (such as normal insulation function or specific preset insulation defect types), reducing human intervention and reducing false positives.

[0053] As a preferred embodiment, the specific method of inputting the denoised UHF partial discharge signal into the partial discharge pattern recognition model to obtain the insulation state recognition result of the lightning arrester body 1 is as follows: The partial discharge pattern recognition model includes an input unit, a feature extraction unit, and a classification decision unit. The input unit receives the denoised UHF partial discharge signal. The feature extraction unit includes a convolutional layer and a pooling layer. The convolutional layer is used to extract high-frequency features and phase distribution features from the partial discharge signal through a ReLU activation function. The pooling layer is used to reduce the dimensionality of the high-frequency features and the phase distribution features through a max-pooling strategy. The reduced high-frequency features and phase distribution features are concatenated in the channel dimension to form partial discharge features. The classification decision unit includes a fully connected layer and a Softmax classifier. The fully connected layer performs linear combination on the partial discharge features through a weight matrix and a bias to generate a score vector. The Softmax classifier maps the score vector to the insulation state recognition result.

[0054] The convolutional layer extracts high-frequency features and phase distribution features of the partial discharge signal through a ReLU activation function, and the pooling layer reduces the feature dimension while retaining key information. This improves the representativeness and computational efficiency of the features. The fully connected layer and the Softmax classifier map the features to the insulation state recognition result, enabling end-to-end classification and outputting clear state labels (such as normal insulation function or specific preset insulation defect types) to support rapid decision-making. The model structure is optimized and suitable for embedded system deployment, enabling real-time monitoring and diagnosis to meet online application requirements.

[0055] As a preferred embodiment, the valve sheet aging trend prediction model includes: a signal decomposition unit, a resistive current calculation unit, a health index calculation unit, and a trend prediction unit. The signal decomposition unit is configured to preprocess the denoised ultrahigh frequency partial discharge signal by using a wavelet transform, and decompose the preprocessed denoised ultrahigh frequency partial discharge signal into a fundamental wave component with doped third harmonic noise by using an adaptive noise complete set empirical mode decomposition algorithm; The resistive current calculation unit is configured to extract a resistive current signal from the fundamental wave component with doped third harmonic noise by using a capacitance current compensation algorithm, correct the amplitude of the resistive current signal by using the ambient temperature collected by the temperature sensor, calculate the rate of change of the peak value of the temperature-corrected resistive current signal with time to obtain a resistive current change rate, and calculate the ratio of the amplitude of the third harmonic noise to the amplitude of the fundamental wave in the resistive current signal to obtain a resistive current third harmonic content. The health index calculation unit is configured to calculate the current health index of the valve unit 4 according to the resistive current change rate, the resistive current third harmonic content, and the partial discharge characteristic value by using a fuzzy comprehensive evaluation model, wherein the health index represents the aging degree quantization value of the valve unit 4. In the present application, each factor (resistive current third harmonic content, resistive current change rate, and partial discharge characteristic) is evaluated separately to determine the degree of belonging to each evaluation grade, forming a membership degree vector. The membership degree vectors of all factors are combined to form a fuzzy relationship matrix R. The weight vector A is combined with the fuzzy relationship matrix R to obtain a comprehensive evaluation vector B, for example, by using a weighted average operator. Assuming that the calculation result is B=(0.1, 0.2, 0.6, 0.1), the final judgment is made according to the comprehensive evaluation vector B. Generally, the maximum membership degree principle is adopted, that is, the evaluation grade of the value with the maximum value (i.e., the health index) in the vector is selected as the final result. In the comprehensive evaluation vector B=(0.1, 0.2, 0.6, 0.1), the maximum value is 0.6, i.e., the health index is 0.6, and the corresponding evaluation grade is abnormal. Therefore, it is determined that the health state of the lightning arrester valve is abnormal.

[0056] The trend prediction unit is configured to use a long short-term memory neural network to input a set of historical health indexes and output the change trend of the health index in a future preset time window. After learning the historical pattern from the set of historical health indexes, the long short-term memory neural network outputs a prediction of the change trend of the health index in the future preset time window (e.g., the next 6 months). The prediction result is a trend line that intuitively shows how the health index may change. When the prediction shows that the health index will deteriorate rapidly and approach the warning threshold, the system will generate a maintenance suggestion in advance.

[0057] Embodiment 3 A computer program product includes computer programs / instructions that, when executed by a processor, implement the method for using the intelligent lightning arrester for a line in embodiment 2.

[0058] Those aspects of the specification that are not otherwise explicitly described will be understood as comprising part of the known art to which the application pertains. Those skilled in the art will appreciate that the embodiments of the application can be practiced with embodiments other than those described, which will vary with respect to the specific implementation details. That is, it will be appreciated that the described embodiments are intended to be illustrative only and in no way limiting of the scope of the application as it is described and claimed herein. It is therefore contemplated to be within the scope of the application to comprise, consist of, or consist essentially of, particular excipients or combinations thereof in the amounts described and / or claimed.

[0059] The present application is described in reference to the drawings using a flowchart and / or a block diagram of the method, apparatus (system) and computer program product according to the embodiments of the application. It will be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for carrying out each of the one or more functions specified in the flowchart and / or block diagram block or blocks.

[0060] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for carrying out each of the one or more functions specified in the flowchart and / or block diagram block or blocks.

[0061] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for carrying out each of the one or more functions specified in the flowchart and / or block diagram block or blocks.

[0062] It should be pointed out finally that the above examples are only used for illustrating the technical solutions of the present application but not for limiting the protection scope thereof, and although the present application has been described in detail with reference to the above examples, it should be understood by those skilled in the art that the specific embodiments of the present application can be changed, modified or replaced equivalently by those skilled in the art after reading the present application, but these changes, modifications or equivalent replacements are all within the protection scope of the claims to be filed of the present application.

Claims

1. A line intelligent surge arrester, characterized by, Comprise: Lightning arrester body (1) and intelligent control box (3); The lightning arrester body (1) is used for providing a low impedance ground channel for overcurrent when overvoltage caused by lightning or switch operation occurs on the line, thereby limiting the overvoltage below a safe level, protecting the subsequent electrical equipment from being damaged, and restoring to an insulation state after the overvoltage disappears, ensuring the normal operation of the power system; The intelligent control box (3) comprises an embedded intelligent sensing module (6) and a diagnosis module (7), the embedded intelligent sensing module (6) is used for collecting the running state parameters of the lightning arrester body (1) in real time, and the diagnosis module (7) is used for evaluating the insulation state of the lightning arrester body (1) by using a partial discharge pattern recognition model and predicting the aging trend of the valve unit (4) of the lightning arrester body (1) by using a valve aging trend prediction model according to the running state parameters of the lightning arrester body (1).

2. The intelligent lightning arrester for lines according to claim 1, characterized in that: The lightning arrester body (1) comprises a valve unit (4) and an insulation shell (5) sleeved outside the valve unit (4), the valve unit (4) is composed of a plurality of zinc oxide resistance pieces doped with Bi2O3 in series, the Bi2O3 doping concentration of each zinc oxide resistance piece gradually decreases from the edge to the center, and the insulation shell (5) is made of high-temperature vulcanized silicone rubber material.

3. The intelligent lightning arrester for lines according to claim 1, characterized in that: The embedded intelligent sensing module (6) comprises a temperature sensor, a leakage current sensor and a UHF partial discharge sensor, the leakage current sensor collects the leakage current signal mixed with electromagnetic interference of the lightning arrester body (1) in real time, the UHF partial discharge sensor collects the UHF partial discharge signal mixed with electromagnetic interference of the lightning arrester body (1) in real time, and the temperature sensor collects the ambient temperature in real time.

4. The intelligent lightning arrester for lines according to claim 3, characterized in that: The diagnosis module (7) is built-in noise suppression algorithm, partial discharge pattern recognition model and valve aging trend prediction model, the noise suppression algorithm removes electromagnetic interference from the leakage current signal mixed with electromagnetic interference and the UHF partial discharge signal mixed with electromagnetic interference, to obtain a denoising leakage current signal and a denoising UHF partial discharge signal; The denoising UHF partial discharge signal is input into the partial discharge pattern recognition model to obtain an insulation state recognition result of the lightning arrester body (1), and the insulation state recognition result is normal insulation function or a preset insulation defect type; The denoising leakage current signal is input into the valve aging trend prediction model to obtain a current aging degree quantitative value of the valve unit (4), and the change trend of the aging degree quantitative value of the valve unit (4) in a future preset time window is output in combination with the ambient temperature and the ambient humidity.

5. The line intelligent surge arrester according to claim 4, characterized in that, The specific method for removing electromagnetic interference from the leakage current signal mixed with electromagnetic interference and the UHF partial discharge signal mixed with electromagnetic interference by the noise suppression algorithm is: The noise suppression algorithm comprises a reference preset noise sample library and an adaptive filter; The adaptive filter is configured to perform adaptive noise suppression based on a least mean square algorithm, and the adaptive filter adjusts its weight coefficients in real time to minimize the output mean square error, and then suppresses the corresponding signal noise component related to the noise sample in the preset noise sample library from the leakage current signal mixed with electromagnetic interference and the ultra-high frequency partial discharge signal mixed with electromagnetic interference according to the real-time adjusted weight coefficients, and extracts the signal with a signal-to-noise ratio greater than a signal-to-noise ratio threshold from the leakage current signal mixed with electromagnetic interference as a denoised leakage current signal, and extracts the signal with a signal-to-noise ratio greater than a signal-to-noise ratio threshold from the ultra-high frequency partial discharge signal mixed with electromagnetic interference as a denoised ultra-high frequency partial discharge signal.

6. The line intelligent surge protector according to claim 4, wherein The partial discharge pattern recognition model is obtained by the following method: The historical ultra-high frequency partial discharge signals in the historical ultra-high frequency partial discharge signal set are labeled, and the label is an insulation state recognition result. The labeled historical ultra-high frequency partial discharge signal set is divided into a training set and a validation set. The convolutional neural network is trained according to the training set and the validation set to obtain the partial discharge pattern recognition model.

7. The line intelligent surge protector according to claim 6, wherein The specific method of inputting the denoised ultra-high frequency partial discharge signal into the partial discharge pattern recognition model to obtain the insulation state recognition result of the lightning arrester body (1) is: The partial discharge pattern recognition model includes an input unit, a feature extraction unit, and a classification decision unit. The input unit receives the denoised ultra-high frequency partial discharge signal. The feature extraction unit includes a convolution layer and a pooling layer. The convolution layer is used to extract high-frequency features and phase distribution features from the partial discharge signal through a ReLU activation function. The pooling layer is used to reduce the dimensions of the high-frequency features and the phase distribution features through a max-pooling strategy. The high-frequency features and the phase distribution features after dimension reduction are concatenated into partial discharge features in the channel dimension. The classification decision unit includes a fully connected layer and a Softmax classifier. The fully connected layer performs linear combination on the partial discharge features through a weight matrix and a bias to generate a score vector. The Softmax classifier maps the score vector to an insulation state recognition result.

8. The line intelligent surge protector according to claim 7, wherein The valve sheet aging trend prediction model includes: a signal decomposition unit, a resistive current calculation unit, a health index calculation unit, and a trend prediction unit. The signal decomposition unit is used to preprocess the denoised ultra-high frequency partial discharge signal by wavelet transform, and decompose the denoised ultra-high frequency partial discharge signal after preprocessing into a fundamental component mixed with third harmonic noise by using an adaptive noise complete set empirical mode decomposition algorithm. The resistive current calculation unit is used to extract a resistive current signal from the fundamental component mixed with third harmonic noise by using a capacitance current compensation algorithm, and to correct the amplitude of the resistive current signal by temperature using the ambient temperature collected by a temperature sensor, to calculate the change rate of the peak value of the temperature-corrected resistive current signal over time to obtain a resistive current change rate, and to calculate the ratio of the third harmonic noise amplitude to the fundamental amplitude in the resistive current signal to obtain a resistive current third harmonic content. The health index calculation unit is configured to calculate a current health index of the valve piece unit (4) according to the resistive current change rate, the resistive current third harmonic content and the partial discharge characteristic value by using a fuzzy comprehensive evaluation model, the health index representing a quantitative value of the aging degree of the valve piece unit (4); The trend prediction unit is configured to use a long short-term memory neural network to take a set of historical health indexes as input and output a change trend of the health index in a future preset time window.

9. A method of diagnosing a state of a line intelligent surge protector, characterized by, Comprise: The arrester body (1) provides a low-impedance ground channel for overcurrent when overvoltage caused by lightning or switch operation occurs on the line, thereby limiting the overvoltage below a safe level and protecting subsequent electrical equipment from being damaged, and restoring to an insulation state after the overvoltage disappears to ensure normal operation of the power system. The intelligent control box (3) comprises an embedded intelligent sensing module (6) and a diagnosis module (7), the embedded intelligent sensing module (6) is configured to collect running state parameters of the arrester body (1) in real time, and the diagnosis module (7) is configured to evaluate an insulation state of the arrester body (1) by using a partial discharge pattern recognition model and predict an aging trend of the valve piece unit (4) of the arrester body (1) by using a valve piece aging trend prediction model according to the running state parameters of the arrester body (1).

10. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instruction is executed by the processor to realize the state diagnosis method of the intelligent arrester for the line in claim 9.

Images