Prediction Of PM_2.5 In Zhengzhou Based On Weighted Support Vector Machine Model

In recent years,air pollution in Henan is serious.It is of great significance to understand the air pollution situation in Henan province and predict the change of the concentration of pollution factors in the future.This paper chooses Zhengzhou,the capital city of Henan,as the research object.First of all,the air pollution situation of Zhengzhou is analyzed,and the change and law of pollutant PM_2.5 is mastered,and the pollution situation is found to be serious.Through the analysis of monthly variation of PM_2.5 concentration,it is found that its seasonal distribution characteristics are obvious,so it is necessary to strengthen the management of PM_2.5 in winter.At the same time,the correlation between PM_2.5 and its influencing factors is analyzed.According to the positive and negative relationship and correlation intensity between these factors and PM_2.5,targeted measures can be taken in pollution control.Secondly,the correlation coefficient and the importance value of features are used to screen the feature variables,and the Grid Search（GS）,Genetic Algorithm（GA）and Particle Swarm Optimization（PSO）are used to optimize the three important parameters of the Support Vector Machine（SVM）model,and the PM_2.5 concentration is predicted.The R² of GS-SVM model is 0.9117,The R² of GA-SVM model is 0.9181,and the R²of PSO-SVM model is 0.9201.After comparison,it is found that the parameters optimized by Particle Swarm Optimization can achieve higher accuracy of the model,so PSO-SVM model is selected as the prediction model of PM_2.5 concentration.Finally,two weighted methods are proposed based on PSO-SVM model.One is the E-PSO-SVM model which uses the added value of error after removing variables as the weight,and the other is R-PSO-SVM model which uses the mean value of two correlation coefficients as the weight.The results show that the prediction accuracy of PSO-SVM reaches 92.01%,the accuracy of E-PSO-SVM used error increment as weight reaches 92.78%,and the prediction accuracy of R-PSO-SVM based on the mean of correlation coefficient reaches 93.50%.Therefore,the R-PSO-SVM model is finally selected as the prediction model of PM_2.5 concentration to provide help for air pollution control.