Implementation And Comparison Of Typical Machine Learning Algorithms In PM2.5 Concentration Prediction

With the continuous development of China’s economy,people’s material and cultural living standards are constantly improving,but air quality pollution problems occur frequently,especially haze pollution.Haze and other air pollution not only bring a lot of inconvenience to human’s normal production,life,work and study,but also bring serious harm to people’s physical and mental health and hinder the sustainable development of society.PM2.5 as the main reason of haze formation,scientific and effective prediction of PM2.5 can make people do a good job of protection in advance and minimize the harm to human body.Industrialization is the foundation of modernization.Our country has been trying to realize industrialization.With the air pollution,the prediction of PM2.5 concentration has become a very realistic topic.In the contemporary research of PM2.5,few typical machine learning algorithms are used for comparative study.This paper mainly uses several theoretical models under the traditional machine learning algorithm,deep learning algorithm and integrated algorithm,selects the panel data of PM2.5 from September 1 to 30,2019 in Beijing,a representative city of China,takes the actual concentration value of PM2.5 as the dependent variable,and takes the data values of dewp,temp,Pres,cbwd,LWS,LS,LR in Beijing as the independent variable,using three kinds of algorithms,namely The efficiency and accuracy of PM2.5 concentration prediction in Beijing are compared by traditional regression algorithm,deep learning algorithm and multiple models under integrated algorithm.The results show that the recognition effect of the integrated algorithm in PM2.5 concentration prediction is significantly better than the other two kinds of algorithms,because the integrated model can be better than the single model and is more robust.Among the several models under the integrated algorithm,the random forest has the best effect.This is because the expression ability of the regression tree is stronger than that of the support vector regression,but if the depth is not limited,the regression tree is likely to over fit,but if the over fit is well controlled,the regression tree used for the base model to do the integrated model will be better than that of the support vector Better return.Compared with linear regression,ridge regression and lasso regression,support vector regression(SVR)can better predict PM2.5 concentration in traditional machine learning algorithms.The prediction ability of several models under the deep learning algorithm is the lowest and has many limitations.To sum up,the random forest model method has a high prediction accuracy in the data of selected cities,has a good performance in the medium and short term prediction application,reflects its feasibility and effectiveness,and can provide a certain technical reference for pollution early warning,air quality assessment and environmental governance.In this study,the method of combining machine learning with traditional weather measurement is used to solve people’s demand for air quality supervision and prediction,and improve the accuracy and real-time of prediction.For example,for machine learning modeling,the amount of data is small,so the test error is relatively large;moreover,only three machine learning algorithms are used in this paper,resulting in the test results are not comprehensive.Although this study has the above shortcomings,it is undeniable that AI algorithm has great reference value and Enlightenment for the future social and technological development.It can produce a lot of applications in various fields,and has a very important practical significance for solving the increasing problems and needs of the contemporary society.