Lightning arrester operation state on-line monitoring and evaluation method and system

Abstract

The invention discloses a lightning arrester operation state online monitoring and evaluation method and system, and relates to the technical field of lightning arrester operation state monitoring, and the method comprises three steps of cooperative correction, health degree evaluation and local defect discrimination. According to the cooperative correction, data is acquired through a multi-sensor network, an environment compensation model is established to optimize the leakage current measurement precision, and a filter is dynamically adjusted to suppress interference signals. Health degree evaluation is fused with multi-dimensional features, a dynamic entropy weight method and a self-adaptive membership function are adopted to realize accurate state grading, and the parameter coupling analysis capability is improved through covariance correction. Local defect identification utilizes a distributed optical fiber sensing network and a digital twinning technology to realize real-time positioning and growth prediction of microdefects. The monitoring precision and reliability of the operation state of the lightning arrester can be effectively improved, potential faults can be found in time, the maintenance cost is reduced, and safe and stable operation of a power system is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of arrester operation status monitoring, and specifically to an on-line monitoring and evaluation method and system for arrester operation status. Background Technique

[0002] As a key infrastructure for the operation of modern society, the safe and stable operation of the power system is of crucial importance. And the arrester, as a core protection component in the power system, undertakes the important task of limiting overvoltage and protecting electrical equipment from lightning strikes and switching overvoltage damages. With the continuous expansion of the power grid scale and the increasingly complex operating environment, the requirements for the reliability and stability of arresters are also constantly increasing. Traditional methods such as manual inspection and regular preventive tests have gradually revealed many deficiencies in monitoring the operation status of arresters, such as untimely detection, inability to monitor in real time, and difficulty in detecting internal defects, which are difficult to meet the requirements of modern power systems for efficient and accurate monitoring of arresters.

[0003] Currently, the existing on-line monitoring systems for arresters have a series of problems in terms of functions and performance. On the one hand, the data acquisition accuracy is not ideal enough and is easily interfered by environmental factors. For key electrical characteristic parameters such as leakage current, there are often large errors in the measurement results, which cannot accurately reflect the true operation status of the arrester. On the other hand, the multi-source data fusion analysis ability is lacking. Most systems can only simply process single-type monitoring data and are difficult to organically integrate multi-dimensional data such as electrical, mechanical, thermodynamic, and environmental data, so they cannot comprehensively and deeply evaluate the health status of arresters, and it is easy to have misjudgment or missed judgment situations. In addition, in terms of fault warning, the warning mechanism of existing systems is relatively simple and crude, and it cannot issue hierarchical warning signals in a timely and accurate manner according to different fault degrees of arresters, and cannot provide sufficient and effective decision-making support for operation and maintenance personnel.

[0004] In order to solve the above defects, a technical solution is provided now. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of existing arrester monitoring technologies in data acquisition accuracy, multi-dimensional feature fusion, fault warning mechanism, and microscopic defect monitoring, and to propose an on-line monitoring and evaluation method and system for arrester operation status.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An on-line monitoring and evaluation method for arrester operation status includes the following steps:

[0008] S1. Collaborative calibration, collecting data through a multi-sensor network, establishing an environmental compensation model, optimizing the measurement accuracy of leakage current, and dynamically adjusting the filter to suppress interference signals;

[0009] S2. Health assessment, integrating multi-dimensional features of electricity, mechanics, thermodynamics, and environment, using the dynamic entropy weight method and adaptive membership function to achieve accurate state classification, and improving the parameter coupling analysis ability through covariance correction;

[0010] S3. Local defect identification, using a distributed optical fiber sensing network and digital twin technology to achieve real-time positioning and growth prediction of microscopic defects, and optimizing the defect classification efficiency by combining edge-cloud collaborative computing.

[0011] Furthermore, the specific process of S1 is as follows:

[0012] S101. Construction of a distributed sensing network: Four groups of broadband current sensors are arranged annularly within 0.5 m above the top of the arrester body to form an anti-interference differential measurement array;

[0013] Five groups of thin-film temperature sensors are embedded equidistantly along the axial direction of the arrester skirt to form an axial temperature gradient monitoring chain;

[0014] A three-axis acceleration sensor is installed at the base to collect the mechanical vibration spectrum in the range of 0.1 - 2000 Hz in real time;

[0015] A micro weather station is integrated at the top to synchronously monitor environmental temperature and humidity, wind speed, and air pressure;

[0016] S102. Establish a leakage current correction model: , where is the corrected leakage current, is the original leakage current, is the temperature compensation coefficient, is the humidity correction factor, T is the real-time temperature, is the rated operating temperature of the equipment, are the real-time air pressure and standard air pressure respectively, is the air pressure normalization factor, and RH is the real-time relative humidity;

[0017] S103. On the basis of the leakage current correction model, add a surface contamination compensation factor to achieve non-intrusive contamination degree estimation through multi-sensor data fusion;

[0018] S104. Intelligent filtering processing: Use a finite impulse response filter bank with adaptive parameters, and the cut-off frequency is dynamically adjusted according to the real-time environmental temperature. Specifically:

[0019] When the temperature rises by 1 °C, the cut-off frequency moves up by 0.5 Hz;

[0020] When the humidity increases by 10%RH, the stopband attenuation increases by 3 dB;

[0021] Used to suppress interference signals in the 20 - 50 kHz frequency band caused by rain and fog weather.

[0022] Furthermore, the specific operation steps of S103 are as follows:

[0023] Establish a pollution rate prediction equation based on environmental parameters: , where is the surface pollution density, is the temperature decay factor, is the humidity influence coefficient, is the wind speed suppression coefficient, with an initial value of 0.08, is the relative humidity at time τ, is the wind speed at time τ, is the surface temperature at time τ, e is the base of the natural logarithm, is the integration variable, representing each moment from the starting time to the current time t, represents the integration with respect to the time variable from time 0 to time t, cumulatively calculating the pollution rate;

[0024] Introduce a pollution compensation term in the leakage current correction model: , where is the corrected leakage current, is the pollution sensitivity coefficient, is the current surface pollution density, is the critical pollution density, is the hyperbolic tangent function;

[0025] Deploy a sliding window optimizer at the edge computing node, and calculate the pollution compensation effect every 30 minutes. The calculation logic is: ; where is the pollution compensation effect, N is the number of data points in the sliding window, is the partial derivative, representing the rate of change of the corrected leakage current with respect to the pollution density p, is the time interval of the i - th data point;

[0026] When is triggered for parameter self - tuning: ; where is the new pollution sensitivity coefficient; is the original pollution sensitivity coefficient; is the sign function, used to judge the positive or negative of, if is positive, then the value is 1, if is 0, then the value is 0, if is negative, then the value is - 1;

[0027] Obtain meteorological station and temperature sensor data, and use the output of the pollution model as the feature input for health assessment. The compensation result is fed back to the adaptive filter, and when it automatically enhances the filtering in the 50 - 100 Hz frequency band.

[0028] Furthermore, the specific operation steps of S2 are as follows:

[0029] S201. Construct multi - dimensional feature engineering and extract state indicators, including:

[0030] Electrical features: the third - harmonic distortion rate of resistive current, the discharge pulse repetition rate;

[0031] Mechanical features: the energy ratio of the pedestal vibration signal in the 100 - 400 Hz frequency band;

[0032] Thermodynamic features: the product of the maximum value and the standard deviation of the axial temperature gradient;

[0033] Environmental features: the weighted comprehensive index of the temperature - humidity change rate and the wind speed;

[0034] S202. Dynamic weight assignment mechanism. Calculate the index weights in real - time by using the improved entropy weight method, and calculate the dispersion degree of each feature parameter: ;

[0035] Generate dynamic weights: ; where is the dispersion degree of the j - th feature parameter, is the dynamic weight of the j - th feature; is the standardized value of the j - th feature of the i - th sample, and n and m are the total number of samples and the total number of features respectively; make the weight assignment automatically adjust with the equipment aging process;

[0036] S203. Fuzzy decision - making mechanism. Construct an S - type membership function: , where is the membership degree of the j - th feature, is the sensitivity adjustment factor, with an initial value of 2.5, is the historical baseline value of each feature quantity; and introduce a dynamic membership function parameter adaptive adjustment mechanism;

[0037] By calculating the comprehensive health degree , establish multi - level early warning, where S is the comprehensive health degree, is the weight of the j - th feature, is the membership degree of the j - th feature;

[0038] When , trigger the emergency isolation of the equipment. When , notify the operator for manual review and decision - making. When Keep the regular monitoring mode when...

[0039] Furthermore, the specific operation steps for introducing the dynamic membership function parameter self - adaptive adjustment mechanism in S203 are as follows:

[0040] Deploy a sliding window statistics module at the edge computing node for each feature parameter to calculate the data distribution characteristics within the window: , , and dynamically update the membership function threshold: ; where are respectively the sliding window mean and the sliding window standard deviation of the feature ; N is the number of data points within the sliding window, is the value of the feature x of the k - th data point, is the updated membership function threshold, is the original membership function threshold;

[0041] Dual - mode adjustment strategy: Steady - state operating condition, i.e., the health change rate < 0.01 / h: ; where is the new feature weight coefficient, used to adjust the weight of the feature under steady - state operating conditions, is the original feature weight coefficient, is the sign function, used to judge the positive or negative of , is the partial derivative of the health S with respect to the feature ;

[0042] Transient operating condition, i.e., the health change rate ≥ 0.01 / h: , is the increment of the parameter in the last 1 hour;

[0043] Introduce a covariance matrix correction mechanism, and calculate the feature parameter covariance matrix every 6 hours : ; where N is the number of samples, is the i - th eigenvalue of the κ - th sample, is the mean of the i - th feature;

[0044] Modify the membership function: ; where is the membership of the i - th feature after modification, is the sensitivity adjustment factor, is the covariance between feature i and feature j, is the product operation for all features, are respectively the standard deviations of the i - th and j - th features;

[0045] Obtain the entropy weight value calculated by the dynamic weight allocation module. When the weight of a certain parameter is such that the adjustment step size is expanded to 0.15;

[0046] Abnormal condition automatically triggers the parameter freezing mechanism: When the mutation of a certain parameter exceeds 3σ, its membership parameter is locked to the value of the previous time period;

[0047] The optimization result is fed back to the manual warning. When the cumulative adjustment exceeds 15%, a parameter calibration work order is generated.

[0048] Furthermore, the specific operation steps of S3 are as follows:

[0049] S301. Deployment of the fiber optic sensing network. Embed a distributed fiber optic array between the valve disc laminations inside the lightning arrester:

[0050] Use Φ0.2mm OFDR fiber, arrange it in a helical winding along the axial direction to cover all valve disc contact surfaces; Set FBG grating sensors every 10 cm to form a temperature-strain dual-parameter sensing chain;

[0051] Integrate a miniaturized edge computing node at the top fitting, with a built-in fiber optic demodulation module and a sampling rate of 2 kHz;

[0052] S302. Resolution of multi-physical field coupling signals. Construct a temperature-strain joint decoupling model: , where is the temperature sensitivity coefficient, is the strain-temperature cross-influence factor, is the strain sensitivity coefficient, is the current wavelength, is the reference wavelength; is the temperature change amount, is the wavelength change amount, and L is the fiber optic length; is the temperature coefficient;

[0053] S303. Real-time diagnosis on the edge side. Use a lightweight defect recognition model for diagnosis. Specifically, through improved one-dimensional depthwise separable convolution, with a convolution kernel width of 5 and 64 channels, extract the axial abnormal signal, introduce the Squeeze-and-Excitation module in the spatial dimension to weight and focus on the defect-sensitive area, and use the random forest algorithm to achieve defect classification;

[0054] S304. Collaborative optimization mechanism, dynamic task allocation strategy: The edge node processes more than 80% of the regular data, with a latency requirement <50 ms. The cloud platform centrally analyzes the complex working condition data, including the damage assessment after lightning strikes;

[0055] Synchronously update the edge node model parameters weekly, with the incremental learning data volume ≥ 10,000 groups. When a new defect mode is detected, automatically trigger the federated learning process, and the number of participating nodes ≥ 50;

[0056] S305. Three-dimensional defect visualization, constructing a digital twin of the lightning arrester: Map the optical fiber monitoring data to a three-dimensional model, and use the HSV color space to encode the temperature-strain composite index:

[0057] Red channel: Temperature anomaly degree, mapped from 0 - 120 °C to 0 - 255;

[0058] Blue channel: Strain over-standard rate, mapped from 0 - 2000 με to 0 - 255;

[0059] Defect growth simulation: Predict the crack propagation path based on the phase field method. When the strain exceeds 80% of the material yield strength, automatically generate repair coordinates.

[0060] An on-line monitoring and evaluation system for the operating state of a lightning arrester, comprising:

[0061] A multi-source data acquisition module, used to integrate a broadband current sensor, a temperature gradient monitoring chain, a three-axis acceleration sensor, and a micro weather station to collect multi-dimensional data of the lightning arrester's electricity, mechanics, thermodynamics, and environment in real time;

[0062] An environment compensation and filtering module, used to dynamically suppress interference through a leakage current correction model and a pollution compensation algorithm, combined with an adaptive filter, to improve data reliability;

[0063] A dynamic feature evaluation module, used to extract five types of features, calculate the dynamic weight using an improved entropy weight method, and generate a comprehensive health score through covariance correction and parameter adaptive adjustment;

[0064] A collaborative diagnosis module, which processes conventional data through edge nodes and analyzes complex working conditions on the cloud platform; uses a distributed optical fiber array to capture microscopic defects, and combines a lightweight deep learning model and federated learning to achieve high-precision defect classification and positioning;

[0065] A digital twin decision module, constructs a three-dimensional digital twin of the lightning arrester, encodes temperature-strain data through the HSV color space, predicts the crack propagation path based on the phase field method, automatically generates repair coordinates, and triggers a multi-level warning strategy.

[0066] Compared with the prior art, the beneficial effects of the present invention are:

[0067] (1) In the present invention, through multi-sensor collaborative calibration and dynamic filtering technologies, the measurement accuracy and anti-interference ability of the monitoring data of the operating state of the lightning arrester are effectively improved. A multi-sensor network is used to collect multi-dimensional data, and an environmental compensation model is established to optimize the measurement accuracy of the leakage current; at the same time, a filter bank with adaptive parameters adjusts dynamically according to the real-time environment to suppress interference signals. The problems of low data acquisition accuracy and susceptibility to environmental interference in the existing monitoring system are solved, providing a high-quality data basis for subsequent health state assessment;

[0068] (2) In the present invention, the dynamic entropy weight method and the adaptive membership function are adopted to realize the accurate hierarchical assessment of the health state of the lightning arrester by fusing multi-dimensional features, and the parameter coupling analysis ability is improved through covariance correction; the dynamic weight is calculated by extracting state indicators, so that the weight distribution is automatically adjusted with the aging of the equipment; the constructed membership function is dynamically optimized according to real-time data, and a multi-level early warning strategy is established; enabling the operation and maintenance personnel to grasp the health state of the lightning arrester in real time, discover potential faults in advance, and effectively improve the operation reliability of the lightning arrester;

[0069] (3) In the present invention, the distributed optical fiber sensing network and digital twin technology are used to realize the real-time positioning and growth prediction of the micro-defects of the lightning arrester, and the defect classification efficiency is optimized by combining edge-cloud collaborative computing; a distributed optical fiber array is embedded inside the lightning arrester to form a temperature-strain dual-parameter sensing chain; through a lightweight defect recognition model and a federated learning algorithm, the edge node processes conventional data, and the cloud platform analyzes complex working conditions to achieve high-precision defect classification and positioning; at the same time, a three-dimensional digital twin body is constructed, and the crack propagation path is predicted based on the phase field method to automatically generate maintenance coordinates. These technologies significantly improve the monitoring and prediction capabilities of the micro-defects inside the lightning arrester, providing a scientific basis for equipment maintenance and repair. Brief Description of the Drawings

[0070] For the convenience of those skilled in the art to understand, the present invention will be further described below in conjunction with the accompanying drawings;

[0071] Figure 1 It is the method flow chart of the present invention. Detailed Embodiments

[0072] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0073] It should be understood that the terms "comprising" and "including" as used in the specification and claims of this disclosure indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0074] It should also be understood that the terms used in this disclosure specification are for the purpose of describing particular embodiments only and are not intended to limit this disclosure. As used in this disclosure specification and claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms. It should further be understood that the term "and / or" as used in this disclosure specification and claims refers to any combination and all possible combinations of one or more of the associated listed items and includes such combinations.

[0075] As Figure 1 shown, an on-line monitoring and evaluation method for the operating state of a lightning arrester includes the following steps:

[0076] Step 1. Cooperative calibration: Collect data through a multi-sensor network, establish an environmental compensation model (temperature and humidity, air pressure, pollution compensation), optimize the measurement accuracy of the leakage current, and dynamically adjust the filter to suppress interference signals;

[0077] Distributed sensing network construction: Arrange 4 groups of broadband current sensors (frequency response of 100 kHz - 10 MHz) in a ring within 0.5 m above the top of the lightning arrester body to form an anti-interference differential measurement array; Embed 5 groups of thin-film temperature sensors (resolution of 0.1 °C) equidistantly along the axial direction of the lightning arrester skirt to form an axial temperature gradient monitoring chain; Install a three-axis acceleration sensor (range of ±20 g) at the base to collect the mechanical vibration spectrum of 0.1 - 2000 Hz in real time; Integrate a micro weather station at the top to synchronously monitor the environmental temperature and humidity (accuracy of ±1%RH), wind speed (range of 0 - 60 m / s), and air pressure (300 - 1100 hPa);

[0078] Establish a leakage current correction model: , where is the corrected leakage current, is the original leakage current, is the temperature compensation coefficient (0.0042 / °C for silicone rubber material), is the humidity correction factor (take 0.018), T is the real-time temperature, is the rated operating temperature of the device and is set to 40 °C, is the air pressure normalization factor, are the real-time air pressure and the standard air pressure respectively; RH is the real-time relative humidity;

[0079] On the basis of the leakage current correction model, the surface pollution compensation factor is added, and the non-sensing pollution degree estimation is realized through multi-sensor data fusion. The specific steps are as follows:

[0080] Establish the pollution rate prediction equation based on environmental parameters: ,in is the surface contamination density, is the temperature attenuation factor, with an initial value of 0.03 / ℃; is the humidity influence coefficient, the initial value is 0.15, is the wind speed suppression coefficient, with an initial value of 0.08. is the relative humidity at time τ, is the wind speed at time τ, is the surface temperature at time τ; e is the base of the natural logarithm, is the integral variable, representing each moment from the start time to the current time t, and is used to cumulatively calculate the pollution density; Represents the time variable Integrate from time 0 to time t and calculate the pollution rate cumulatively;

[0081] Introducing pollution compensation term into the leakage current correction model: ,in is the corrected leakage current, is the pollution sensitivity coefficient, with a value of 0.12. is the current surface pollution density, is the critical pollution density, , It is a hyperbolic tangent function, which realizes nonlinear compensation and avoids overcorrection;

[0082] A sliding window optimizer is deployed on the edge computing node to calculate the pollution compensation effect every 30 minutes. The calculation logic is: ;in is the pollution compensation effect, which is used to measure the influence of pollution compensation on leakage current correction. N is the number of data points in the sliding window. is the partial derivative, which represents the leakage current after correction The rate of change with the pollution density p reflects the degree of influence of the change of pollution density on the leakage current; is the time interval of the i-th data point;

[0083] when When triggering parameter auto-tuning: ;in is the new pollution sensitivity coefficient, which is based on the pollution compensation effect The pollution sensitivity coefficient after self-tuning is used to update the sensitivity parameters in the pollution compensation model; is the original pollution sensitivity coefficient; is the sign function used to judge the positive or negative of, if is positive, the value is 1, if is 0, the value is 0, if is negative, the value is -1;

[0084] Obtain the data of the weather station (temperature, humidity, wind speed) and the temperature sensor. The output of the pollution model is used as the feature input for health assessment. The compensation result is fed back to the adaptive filter. When is satisfied, the filtering of the 50 - 100 Hz frequency band is automatically enhanced.

[0085] Intelligent filtering processing: Adopt a finite impulse response filter bank with adaptive parameters, and its cut-off frequency is dynamically adjusted according to the real-time ambient temperature: for every 1 °C increase in temperature, the cut-off frequency moves up by 0.5 Hz; for every 10%RH increase in humidity, the stopband attenuation is increased by 3 dB; suppress the interference signal in the 20 - 50 kHz frequency band caused by rainy and foggy weather.

[0086] Step 2: Health assessment. Integrate multi-dimensional features (electrical, mechanical, thermodynamics, environment), adopt the dynamic entropy weight method and the adaptive membership function to achieve accurate state classification, and improve the parameter coupling analysis ability through covariance correction;

[0087] Extract state indicators, including:

[0088] Electrical characteristics: the third harmonic distortion rate of resistive current, the partial discharge pulse repetition rate; Mechanical characteristics: the energy ratio of the pedestal vibration signal in the 100 - 400 Hz frequency band; Thermodynamic characteristics: the product of the maximum value and the standard deviation of the axial temperature gradient; Environmental characteristics: the weighted comprehensive index of the temperature and humidity change rate and the wind speed;

[0089] The dynamic weight allocation mechanism calculates the index weights in real time by adopting the improved entropy weight method, and calculates the dispersion degree of each characteristic parameter: , generate the dynamic weight: , where is the dispersion degree of the j-th characteristic parameter, is the dynamic weight of the j-th characteristic; is the normalized value of the j-th characteristic of the i-th sample, and n and m are the total number of samples and the total number of characteristics respectively; make the weight allocation automatically adjust with the equipment aging process. When the temperature gradient increases abnormally, its weight coefficient is automatically increased by 35% - 60%;

[0090] Fuzzy decision-making mechanism, construct an S-shaped membership function: , where is the membership degree of the j-th characteristic, is the sensitivity adjustment factor, and the initial value is 2.5, is the historical baseline value of each characteristic quantity;

[0091] Introduce a dynamic membership function parameter adaptive adjustment mechanism, and the process is as follows:

[0092] Deploy a sliding window statistics module (window size = 24 hours, step size = 1 hour) at the edge computing node, for each characteristic parameter , calculate the data distribution characteristics within the window: , , dynamically update the membership function threshold: ; where are respectively the sliding window mean and sliding window standard deviation of the characteristic; N is the number of data points within the sliding window, is the value of the characteristic x of the k-th data point, is the updated membership function threshold, is the original membership function threshold;

[0093] Dual-mode adjustment strategy: Steady-state operating condition (health change rate < 0.01 / h): ; where is the new characteristic weight coefficient, used to adjust the weight of the characteristic under steady-state operating conditions, is the original characteristic weight coefficient, is the sign function, used to judge positive or negative, is the partial derivative of the health S with respect to the characteristic ; Transient operating condition (health change rate ≥ 0.01 / h): , is the increment of the parameter in the most recent 1 hour;

[0094] Introduce a covariance matrix correction mechanism, calculate the characteristic parameter covariance matrix every 6 hours : ; where N is the number of samples, is the i-th eigenvalue of the κ-th sample, is the mean of the i-th characteristic;

[0095] Modify the membership function: ; where is the membership of the i-th characteristic after modification, is the sensitivity adjustment factor, is the covariance between characteristic i and characteristic j, is for performing a product operation on all characteristics, are respectively the standard deviations of the i-th and j-th characteristics;

[0096] Obtain the entropy weight value calculated by the dynamic weight allocation module. When the weight of a certain parameter is less than, its corresponding adjustment step is expanded to 0.15; the abnormal condition automatically triggers the parameter freezing mechanism: when the mutation of a certain parameter exceeds 3σ, its membership parameter is locked to the value of the previous time period; the optimization result is fed back to the manual warning. When the cumulative adjustment exceeds 15%, a parameter calibration work order is generated.

[0097] By calculating the comprehensive health degree , a multi-level warning is established, where S is the comprehensive health degree, is the weight of the jth feature, is the membership degree of the jth feature; when , the equipment emergency isolation is triggered. When , the manual is notified to make a manual review decision. When , the conventional monitoring mode is maintained.

[0098] Step 3: Local defect identification. Utilize the distributed optical fiber sensing network and digital twin technology to achieve real-time positioning and growth prediction of microscopic defects, and combine edge-cloud collaborative computing to optimize the defect classification efficiency;

[0099] Embed a distributed optical fiber array between the valve disc laminations inside the lightning arrester:

[0100] Adopt Φ0.2mm high-temperature-resistant OFDR optical fiber (spatial resolution 1cm), arrange it in a helical winding along the axis, covering all valve disc contact surfaces; set FBG grating sensors (wavelength accuracy ±5pm) every 10cm to form a temperature-strain dual-parameter sensing chain; integrate a miniaturized edge computing node (computing power ≥4TOPS) at the top fitting, and build an optical fiber demodulation module (sampling rate 2kHz) inside;

[0101] Construct a temperature-strain joint decoupling model: , where is the temperature sensitivity coefficient, with an initial value of 10.8pm / ℃, is the strain-temperature cross-influence factor, with an initial value of , is the strain sensitivity coefficient, with an initial value of ; is the current wavelength, representing the wavelength of the fiber grating sensor in the current environment, is the reference wavelength, representing the wavelength of the fiber grating sensor under standard conditions; is the temperature change, is the wavelength change, and L is the fiber length; is the temperature coefficient, representing the thermal expansion coefficient of the material;

[0102] Diagnosis is carried out using a lightweight defect recognition model. Specifically, through an improved one-dimensional depthwise separable convolution (kernel width = 5, number of channels = 64), axial abnormal signals are extracted. The Squeeze-and-Excitation module is introduced in the spatial dimension to weight and focus on the defect-sensitive areas. The random forest algorithm (tree depth = 8, proportion of feature subset = √d) is used to achieve defect classification;

[0103] Dynamic task allocation strategy: Edge nodes process more than 80% of the conventional data (delay requirement < 50 ms), and the cloud platform centrally analyzes complex working condition data (such as damage assessment after lightning strike). The model parameters of the edge nodes are updated synchronously every week (incremental learning data volume ≥ 10,000 groups). When a new defect mode is detected, the federated learning process is automatically triggered (number of participating nodes ≥ 50);

[0104] Constructing a digital twin of the lightning arrester: Mapping the fiber optic monitoring data to a three-dimensional model (mesh accuracy ≤ 1 mm³), and using the HSV color space to encode the temperature-strain composite index:

[0105] Red channel: Temperature anomaly degree (0 - 120 °C is mapped to 0 - 255); Blue channel: Strain over-standard rate (0 - 2000 με is mapped to 0 - 255); Defect growth simulation: Based on the phase field method to predict the crack propagation path (time step = 1 h). When the local strain exceeds 80% of the material yield strength, the repair coordinates are automatically generated (latitude and longitude error < 5 cm).

[0106] An on-line monitoring and evaluation system for the operating status of lightning arresters, including a multi-source data acquisition module, an environmental compensation and filtering module, a dynamic feature evaluation module, a collaborative diagnosis module, and a digital twin decision module:

[0107] The multi-source data acquisition module integrates a broadband current sensor, a temperature gradient monitoring chain, a three-axis acceleration sensor, and a micro meteorological station to collect multi-dimensional data of the electrical, mechanical, thermodynamic, and environmental aspects of the lightning arrester in real time, providing the original signal for subsequent analysis;

[0108] The environmental compensation and filtering module uses a leakage current correction model (temperature / humidity / pressure compensation) and a pollution compensation algorithm (hyperbolic tangent non-linear correction), combined with an adaptive filter to dynamically suppress interference (such as 20 - 50 kHz band noise in rainy and foggy weather), improving data reliability;

[0109] The dynamic feature evaluation module extracts five types of features such as the harmonic distortion rate of the resistive current and the proportion of vibration energy, calculates the dynamic weight using the improved entropy weight method (such as the weight increases by 35% - 60% when the temperature is abnormal), and generates a comprehensive health score through covariance correction and parameter adaptive adjustment;

[0110] The edge nodes of the collaborative diagnosis module process conventional data (time delay < 50 ms), and the cloud platform analyzes complex working conditions (such as lightning damage); the distributed fiber optic array (OFDR + FBG) is used to capture microscopic defects, and combined with lightweight deep learning models (such as depthwise separable convolution) and federated learning to achieve high-precision defect classification and localization;

[0111] The digital twin decision module constructs a three-dimensional digital twin of the lightning arrester, encodes temperature-strain data through the HSV color space, predicts the crack propagation path based on the phase field method, automatically generates maintenance coordinates (error < 5 cm), and triggers a multi-level early warning strategy (emergency isolation / artificial review / conventional monitoring).

[0112] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only the specific embodiments. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. An on-line monitoring and evaluation method for the operating state of a lightning arrester, characterized in that, It includes the following steps: S1. Collaborative calibration: Collect data through a multi-sensor network, establish an environmental compensation model, optimize the measurement accuracy of leakage current, and dynamically adjust the filter to suppress interference signals; S2. Health assessment: Integrate multi-dimensional features of electricity, mechanics, thermodynamics, and environment, use the dynamic entropy weight method and adaptive membership function to achieve accurate state classification, and improve the parameter coupling analysis ability through covariance correction; S3. Local defect identification: Utilize a distributed optical fiber sensing network and digital twin technology to achieve real-time positioning and growth prediction of microscopic defects, and optimize the defect classification efficiency by combining edge-cloud collaborative computing.

2. The on-line monitoring and evaluation method for the operation state of a lightning arrester according to claim 1, characterized in that, The specific process of S1 is as follows: S101. Construction of a distributed sensing network: Ring-arrange 4 groups of broadband current sensors within 0.5 m above the top of the arrester body to form an anti-interference differential measurement array; Embed 5 groups of thin-film temperature sensors at equal intervals along the axial direction of the arrester skirt to form an axial temperature gradient monitoring chain; Install a three-axis acceleration sensor at the base to collect the mechanical vibration spectrum in the range of 0.1 - 2000 Hz in real time; Integrate a micro-meteorological station at the top to synchronously monitor environmental temperature, humidity, wind speed, and air pressure; S102. Establish a leakage current correction model: , where is the corrected leakage current, is the original leakage current, is the temperature compensation coefficient, is the humidity correction factor, T is the real-time temperature, is the rated operating temperature of the device, are the real-time air pressure and the standard air pressure respectively, is the air pressure normalization factor, and RH is the real-time relative humidity; S103. On the basis of the leakage current correction model, add a surface contamination compensation factor to achieve non-intrusive contamination degree estimation through multi-sensor data fusion; S104. Intelligent filtering processing: Adopt a finite impulse response filter bank with adaptive parameters, and the cut-off frequency is dynamically adjusted according to the real-time environmental temperature. Specifically: When the temperature rises by 1 °C, the cut-off frequency shifts up by 0.5 Hz; When the humidity increases by 10%RH, the stopband attenuation increases by 3 dB; It is used to suppress interference signals in the 20 - 50 kHz frequency band caused by rain and fog weather.

3. An on-line monitoring and evaluation method for the operating state of a lightning arrester according to claim 2, characterized in that, The specific operation steps of S103 are as follows: Establish a pollution rate prediction equation based on environmental parameters: , where is the surface pollution density, is the temperature decay factor, is the humidity influence coefficient, is the wind speed suppression coefficient, with an initial value of 0.08, is the relative humidity at time τ, is the wind speed at time τ, is the surface temperature at time τ, and e is the base of the natural logarithm, is the integration variable, representing each moment from the starting time to the current time t, represents the time variable is integrated, from time 0 to time t, to cumulatively calculate the pollution rate; Introduce a pollution compensation term into the leakage current correction model: , where is the corrected leakage current, is the pollution sensitivity coefficient, is the current surface pollution density, is the critical pollution density, is the hyperbolic tangent function; Deploy a sliding window optimizer at the edge computing node, and calculate the pollution compensation effect every 30 minutes. The calculation logic is as follows: ; where is the pollution compensation effect, N is the number of data points within the sliding window, is the partial derivative, representing the corrected leakage current The rate of change with respect to the pollution density p, is the time interval of the i-th data point; When parameter self-tuning is triggered: ; where is the new pollution sensitivity coefficient; is the original pollution sensitivity coefficient; is the sign function, used to judge positive or negative. If is positive, the value is 1. If is 0, the value is 0. If is negative, the value is -1; Obtain the data of the weather station and temperature sensors, use the output of the pollution model as the feature input for health assessment, and feedback the compensation result to the adaptive filter. When it automatically enhances the filtering of the 50 - 100 Hz frequency band.

4. The on-line monitoring and evaluation method for the operation state of a lightning arrester according to claim 1, characterized in that, The specific operation steps of S2 are as follows: S201. Construction of multi-dimensional feature engineering, extract state indicators, including: Electrical features: Third-harmonic distortion rate of resistive current, discharge pulse repetition rate; Mechanical features: Energy ratio of the base vibration signal in the frequency band of 100 - 400 Hz; Thermodynamic features: Product of the maximum value and standard deviation of the axial temperature gradient; Environmental features: Weighted comprehensive index of temperature and humidity change rate and wind speed; S202. The dynamic weight allocation mechanism calculates the index weights in real time by using the improved entropy weight method and calculates the dispersion degree of each characteristic parameter: ; Generate dynamic weights: ; where is the dispersion of the j-th characteristic parameter, is the dynamic weight of the j-th characteristic; is the normalized value of the j-th feature of the i-th sample, and n and m are the total number of samples and the total number of features respectively; enabling the weight allocation to be automatically adjusted with the aging process of the device; S203. Fuzzy decision-making mechanism, constructing an S-shaped membership function: , where is the membership degree of the j-th feature, is the sensitivity adjustment factor, with an initial value of 2.5, is the historical baseline value of each feature quantity; and introduce a dynamic membership function parameter adaptive adjustment mechanism; By calculating the comprehensive health degree , a multi-level early warning is established, where S is the comprehensive health degree, is the weight of the jth feature, is the membership degree of the jth feature; When occurs, trigger the emergency isolation of the device. When occurs, notify the operator for manual review and decision-making. When occurs, maintain the normal monitoring mode.

5. The on-line monitoring and evaluation method for the operation state of a lightning arrester according to claim 4, characterized in that The specific operation steps of introducing the dynamic membership function parameter adaptive adjustment mechanism in S203 are as follows: Deploy a sliding window statistics module on the edge computing node, and for each feature parameter , calculate the data distribution characteristics within the window: , , dynamically update the membership function threshold: ; where are respectively the sliding window mean and the sliding window standard deviation of the feature ; N is the number of data points within the sliding window, is the value of the feature x of the k-th data point, is the updated membership function threshold, is the original membership function threshold; Dual - mode adjustment strategy: Under steady - state conditions, i.e., the health - degree change rate < 0.01 / h: ; where is the new feature weight coefficient, used to adjust the weights of features under steady - state conditions, is the original feature weight coefficient, is the sign function, used to judge the positive or negative of is the partial derivative of the health - degree S with respect to the feature ; Transient operating conditions, i.e., health rate of change ≥ 0.01 / h: , is the increment of the parameter in the most recent 1 hour; Introduce a covariance matrix correction mechanism to calculate the covariance matrix of feature parameters every 6 hours : ; where N is the number of samples, is the i-th eigenvalue of the κ-th sample, is the mean of the i-th feature; Revised membership function: ; where is the membership of the i-th feature after revision, is the sensitivity adjustment factor, is the covariance between feature i and feature j, is for performing product operations on all features, are the standard deviations of the i-th and j-th features respectively; Obtain the entropy weight value calculated by the dynamic weight allocation module. When the weight of a certain parameter is, its corresponding adjustment step is expanded to 0.15; Automatic trigger of the parameter freezing mechanism under abnormal conditions: When a certain parameter mutation exceeds 3σ, its membership parameter is locked to the value of the previous time period; The optimization results are fed back to the manual warning, and when a cumulative adjustment exceeds 15%, a parameter calibration work order is generated.

6. The on-line monitoring and evaluation method for the operating state of a lightning arrester according to claim 1, characterized in that The specific operation steps of S3 are as follows: S301. Deployment of the optical fiber sensing network, embed a distributed optical fiber array between the valve disc laminations inside the arrester: Adopt Φ0.2mm OFDR optical fiber, helically wind and arrange it along the axial direction to cover all valve disc contact surfaces; Set FBG grating sensors every 10 cm to form a temperature-strain dual-parameter sensing chain; Integrate a miniaturized edge computing node at the top fitting, with an optical fiber demodulation module built-in and a sampling rate of 2 kHz; S302. Analyze the multi-physical field coupling signals and construct a temperature-strain joint decoupling model: , where is the temperature sensitivity coefficient, is the strain-temperature cross-influence factor, is the strain sensitivity coefficient, is the current wavelength, is the reference wavelength; is the temperature change, is the wavelength change, and L is the optical fiber length; is the temperature coefficient; S303. Edge-side real-time diagnosis is performed using a lightweight defect recognition model. Specifically, through improved one-dimensional depthwise separable convolution with a convolution kernel width of 5 and 64 channels, axial abnormal signals are extracted. The Squeeze-and-Excitation module is introduced in the spatial dimension to weight and focus on the defect-sensitive areas, and the random forest algorithm is used to achieve defect classification; S304. Cooperative optimization mechanism, dynamic task allocation strategy: Edge nodes process more than 80% of the regular data with a latency requirement of <50 ms, and the cloud platform centrally analyzes complex working condition data, including damage assessment after lightning strikes; The edge node model parameters are synchronously updated weekly, and the incremental learning data volume is ≥10,000 groups. When a new defect mode is detected, the federated learning process is automatically triggered, and the number of participating nodes is ≥50; S305. 3D defect visualization, constructing a digital twin of the lightning arrester: Mapping the fiber optic monitoring data to a 3D model and using the HSV color space to encode the temperature-strain composite index: Red channel: Temperature anomaly, mapped from 0 - 120 °C to 0 - 255; Blue channel: Strain exceedance rate, mapped from 0 - 2000 με to 0 - 255; Defect growth simulation: Predicting the crack propagation path based on the phase field method, and automatically generating repair coordinates when the strain exceeds 80% of the material yield strength.

7. A system for the online monitoring and evaluation method of the operating state of a lightning arrester according to any one of claims 1-6, characterized in that Including: Multi-source data acquisition module, used to integrate broadband current sensors, temperature gradient monitoring chains, triaxial acceleration sensors, and micro-meteorological stations to collect multi-dimensional data of the electrical, mechanical, thermodynamic, and environmental aspects of the lightning arrester in real time; Environmental compensation and filtering module, used to dynamically suppress interference through a leakage current correction model and a pollution compensation algorithm, combined with an adaptive filter, to improve data reliability; Dynamic feature evaluation module, used to extract five types of features, calculate dynamic weights using an improved entropy weight method, and generate a comprehensive health score through covariance correction and parameter adaptive adjustment; Cooperative diagnosis module, processing regular data through edge nodes and analyzing complex working conditions on the cloud platform; Capturing microscopic defects using a distributed fiber optic array, and achieving high-precision defect classification and positioning through a lightweight deep learning model and federated learning; Digital twin decision module, constructing a 3D digital twin of the lightning arrester, encoding temperature-strain data through the HSV color space, predicting the crack propagation path based on the phase field method, automatically generating repair coordinates, and triggering a multi-level warning strategy.

Images