PM2.5 Combination Prediction Based On ARIMA-BP Neural Network

With the improvement of the quality of life,air purification and environmental protection issues have attracted more and more attention and thinking.Why the environment has deteriorated and how to effectively manage it have become issues that we urgently need to explore.In the prevention and control of environmental pollution,the problem of smog is extremely prominent,and PM2.5 is the main component of smog.Therefore,the analysis and prediction of PM2.5 mass concentration is very important to control the haze.This paper selects the air quality data of four cities in Beijing,Suzhou,Chongqing and Lanzhou from January 1,2020 to December 31,2020,a total of 366 days.Through feature analysis,model selection and empirical analysis,the optimal prediction model is selected to predict the PM2.5 mass concentration,and based on the prediction results,it provides a real and effective numerical reference for relevant departments to formulate policies and improve systems.Firstly,a feature analysis of the PM2.5 mass concentration data of four cities in2020 was carried out.The results showed that the annual PM2.5 mass concentration data of the four cities showed certain seasonal characteristics.Secondly,the scatter plot and correlation coefficient table of PM2.5 and PM10,SO₂,CO,NO₂,O₃ are used to explore the correlation between these six air pollutants.The results showed that there were different levels of positive correlations between PM2.5 and PM10,SO₂,CO,NO₂ in the four cities,respectively.The correlation between PM2.5 and O₃ is very weak in Beijing and Suzhou.There were different levels of negative correlation between PM2.5and O₃ in Chongqing and Lanzhou,respectively.Finally,this paper chooses to establish ARIMA and BP neural network models for the PM2.5 mass concentration data of four cities in 2020.On the basis of these two models,three combined prediction models are established by using equal weight fusion method,weighted average fusion method and reciprocal root mean square error fusion method respectively.By comparing and analyzing the above five models,it can be seen that for Beijing,Chongqing and Lanzhou,the inverse root mean square error fusion method has the highest prediction accuracy and the best prediction effect for PM2.5.It can predict PM2.5 more accurately.As far as Suzhou City is concerned,the BP neural network model has the highest prediction accuracy and the best prediction effect on PM2.5.That is,for the four research cities selected in this paper,there is no universal optimal prediction model.Therefore,each city should adapt measures to local conditions and establish an optimal prediction model for the PM2.5 mass concentration data according to the city’s air quality.