| The requirements for safety and reliability of the power supply system are becoming higher and higher with the continuous development of the power system.As a widely used overvoltage protection device in power systems,the Metal Oxide Arrester（MOA）is susceptible to aging,moisture and other insulation degradation in actual operation due to frequency voltages and environmental factors,which can lead to serious loss of protection and even explosion.Therefore,online monitoring and diagnosis of MOA are of great significance.However,the factors affecting the online monitoring and diagnosis of MOA are complex,and the existing system suffers from low accuracy of online monitoring,incomplete collection of characteristic quantities,low accuracy of online diagnosis and imperfect online diagnosis functions in actual complex working conditions.In response to the above problems,this topic carries out research on key technologies for online monitoring and online diagnosis of MOA,and designs a set of online monitoring system for MOA.（1）MOA online monitoring technology research.To improve the accuracy of online monitoring,the advantages and disadvantages of nine online monitoring methods,such as the capacitive current compensation method,were first analyzed,and the harmonic analysis method was determined as the online monitoring method in this paper.An improved harmonic analysis method is proposed to address the problem that the accuracy of this method can be degraded due to spectral leakage and fence effects caused by asynchronous sampling when the grid frequency fluctuates.The hardware adaptively adjusts the sampling frequency according to the grid frequency to fundamentally reduce the non-synchronous sampling;the software compares various window functions and interpolation algorithms,adopts Nuttall window to suppress the spectrum leakage,and the three-spectrum interpolation algorithm to reduce the fence effect;meanwhile,the maximum spectrum at the peak frequency of each harmonic of the interpolation algorithm is directly obtained through the hardware acquisition of the fundamental frequency,which reduces the number of calculations for the maximum spectral line solution from 30 in the original algorithm to 1.This reduces the computational effort and helps to improve the real-time performance of the system.Simulation results show that,compared to the harmonic analysis method,the algorithm reduces the maximum relative error magnitude from 10-2 to 10-7 over the range of grid frequency fluctuations,effectively improving the accuracy of the algorithm.Then,the five types of errors affecting the online monitoring of MOA,such as inter-phase capacitive interference,are summarized and analyzed,and the corresponding error compensation scheme is designed.Finally,for the characteristics of small amplitude and large span（μA～m A）of MOA leakage current,a customized active zero flux current transformer is used to achieve high accuracy conversion of small current signals;a seven-stage adaptive amplification circuit is designed to achieve adaptive amplification of leakage currents of different amplitudes in order to improve AD acquisition accuracy.To comprehensively collect the characteristic quantities to meet the subsequent online diagnosis requirements,the acquisition scheme of seven characteristic quantities,such as resistive current fundamental wave,was determined.Based on the above scheme,the software and hardware design of the signal acquisition device and the software design of the station control host were completed to realize the construction of the MOA online monitoring system.Tests have shown that the system is more accurate than existing devices in the market and industry standards.（2）MOA online diagnostic technology research.To improve the online diagnostic accuracy and perfect the online diagnostic function,firstly a 10 k V MOA test environment was built to verify the specific effects of busbar voltage,temperature and humidity on the leakage current,thus determining the appropriate online diagnostic characteristic quantities.Secondly,the data set is optimized by means of a designed high-precision,multi-characteristic quantity signal acquisition device for data acquisition.Finally,an MOA online diagnosis strategy using resistive current fundamental and third harmonic as key eigenvolumes for fault warning and then seven eigenvolumes such as resistive current fundamental,temperature and humidity for fault classification is proposed.An MOA fault warning model based on PSO-GRU is proposed for fault warning,using PSO to optimize the GRU hyperparameters to improve the model prediction accuracy,and introducing dropout to reduce the overfitting problem of deep learning and improve the network performance.The experimental results show that the MAE and RMSE of the model are 0.0176 and 0.0259,respectively,which have lower prediction errors relative to the models based on BP,RNNN,LSTM,and GRU,indicating that its prediction accuracy is high and can accurately realize MOA fault warning.An MOA fault classification model based on PSO-XGBOOST was proposed for fault classification,and the same PSO was used to optimize the XGBOOST hyperparameters to improve the classification accuracy of the model.The experimental results show that the model has 99.62%precision,98.14%recall,98.87%F1-score,and 0.968 AUC,which is better than the models based on NB,MLP,SVM,RF,and XGBOOST,and can accurately achieve MOA fault classification.The MOA online monitoring system researched and designed in this paper achieves high precision and comprehensive signal acquisition,improves online diagnostic accuracy,perfects online diagnostic functions and has wide practical application value. |
PDF Full Text Request