Research On PM2.5 Concentration Prediction Of Improved GS-SVM Based On Wavelet Decomposition And Principal Component Analysis

In recent years,the haze of many regions in our country is frequent,which has a serious impact on people's health and social development.As the most harmful component of the haze,the PM2.5 can be suspended in the air for a long time,and is sucked into the human body to easily cause a series of diseases such as the cardiovascular system and the respiratory tract.When the haze occurs,the increase in the concentration of the particulate matter can cause the change of the atmospheric water vapor,which can cause the change of the tropospheric delay.The GPS-sounding atmospheric water vapor has the advantages of real-time,not affected by the weather,high precision and the like.Therefore,it is of great significance to use the GPS technology to establish the high-efficiency PM2.5 prediction model for the prevention and treatment of the haze.In this paper,taking Beijing area as an example,the zenith tropospheric delay ZTD and atmospheric precipitable water vapor PWV,are inversed by using the method of precise single point positioning to retrieve atmospheric water vapor.Combined with air pollutant data and meteorological data,an improved GS-SVM model combining wavelet decomposition and principal component analysis is established to predict PM2.5.The main research contents of this paper are as follows:1.Several factors affecting the solution of zenith troposphosphere by precise single point positioning technology are analyzed,and the effects of satellite altitude cut-off angle,satellite ephemeris and clock difference,and mapping function on solving ZTD are discussed respectively.It is verified that the accuracy of real-time ephemeris clock error data can meet the needs of real-time atmospheric water vapor inversion applications.2.The relationship between the four seasons PM2.5 and ZTD/ PWV,the atmospheric pollutants and the meteorological conditions is analyzed.The results show that the correlation of PWV and PM2.5 in summer is lower,and the other three seasons show strong positive correlation;ZTD has negative correlation in spring and PM2.5,and the remaining three seasons show positive correlation;The positive correlation of PM2.5 and SO2,NO2 and CO in the four seasons,and negative correlation with O3,PM2.5 and wind speed,air pressure and precipitation show a negative correlation,and the temperature,relative humidity and dew point temperature show positive correlation.On the other hand,when the haze is frequent,the wind direction is mainly the southwest wind and the south wind.3.The grid search algorithm was improved by combining rough search and fine search.The improved GSSVM was established by combining PWV and PWV in combination with meteorological factors,combining PWV,meteorological factor modeling and air pollutants.4.The improved GS-SVM combination model based on wavelet decomposition and principal component analysis is established.Compared with the improved GS-SVM model,the MAE and RMSE of the combined model are lower than those of the improved GS-SVM model in spring,summer,autumn and winter.The prediction accuracy of PM2.5 is improved effectively.