Research On Prediction Method Of Insulator Contamination Based On Grey Relational Analysis Of Meteorological Factors

In recent years,the continuous upgrading of power grid voltage has greatly improved the efficiency of trans-regional power transmission,promoted the realization of the strategic goal of smart grids,and laid a solid foundation for China’s modernization.At the same time,the probability of pollution flashover accident also increases,which seriously threatens the safe and stable operation of the power system and affects people’s normal life.The occurrence of pollution flashover is closely related to the pollution degree of insulator surface.In recent years,the evaluation of equivalent salt density of insulator surface pollution degree has become a research hotspot.Based on this,this thesis uses Grey Relational Analysis,BP Neural Network and Genetic Algorithm based on multi-non-unifmutation to study a method for predicting insulator pollution degree based on Grey Relational Analysis of meteorological factors and MNUM-GA-BPNN,and an evaluation method of insulator surface pollution degree is proposed.Firstly,this thesis introduces the characteristics of meteorological conditions and the basic principle of Grey Relational Analysis method,and expounds the close relationship between meteorological factors and surface equivalent salt deposit density of insulators.Secondly,the natural fouling test platform of insulator was built and the test sampling equipment was introduced.The surface salt density of XP-70 type insulator was measured regularly,and various relevant meteorological parameters were recorded.The total cycle of natural fouling test was 8 months,and a total of 60 groups of samples were obtained,thus providing a large amount of test data for the training of the prediction model.Thirdly,the Grey Relational Analysis method is used to calculate the correlation degree between various meteorological factors and equivalent salt density,and then the meteorological factor with a greater correlation degree with equivalent salt density is selected as the input of the prediction model of insulator surface pollution degree.Finally,respectively introduces the BP neural network,genetic algorithm,and based on the principle of multi-point non-uniform mutation genetic algorithm,and their respective characteristics,and the correlation between major meteorological factors(temperature,humidity,wind speed,precipitation,PM2.5,SO2 and NO2,a total of 7 classes)as input characteristic.Insulator surface equivalent salt as the only output model.Established the BP neural network(BPNN)and genetic algorithm to optimize the BP neural network(GA-BPNN)and based on the multi-point non-uniform mutation of genetic algorithm to optimize the BP neural network(MNUM-GA-BPNN)three kinds of prediction model.In addition,the network training is carried out with the relevant test data collected,and the influence of the selection of various parameters on the network error is verified.Seven groups of samples were randomly selected for prediction simulation,and the relative errors and determination coefficients of the three models were compared.The results show that MNUM-GA-BPNN has smaller error,higher determination coefficient and better accuracy than GA-BPNN and BPNN in predicting XP-70 type insulator surface equivalent salt density by using meteorological factors,which indicates that the BP neural network optimized by multi-non-uniform variation genetic algorithm has higher effectiveness in predicting insulator surface pollution degree by using meteorological factors.