Air Pollution Forecasting In Zhengzhou Based On Data-driven Models

The atmosphere is one of the indispensable environmental elements for human existence.At the same time,air pollution is an important risk factor for global diseases and has a significant impact on human health.However,with the rapid development of modern industry and the rapid concentration of urban population,traffic pressure has increased rapidly and a lot of fossil fuels have been burned.The rapid increase of PM₁₀,PM_2.5,NO₂,SO₂,CO and O₃ has greatly exceeded the capacity of the atmospheric environment.Therefore,it is urgent to control air pollution.In 2018,China also launched the"blue sky defense".Therefore,the study on the prediction of pollutant concentration of air pollution can give a timely warning of the occurrence of serious pollution events,and provide an effective reference for the decision-making and management of the government and other departments,as well as the travel arrangements.This paper presents a hybrid forecasting model which combines random forest missing data interpolation with empirical mode decomposition and NARX neural network model.In order to evaluate the accuracy of model prediction,five cities were selected:East?47th Middle School?,West?Water Supply Company?,South?City Monitoring Station?,North?Gangli Reservoir?and City Center?Tobacco Factory?in Zhengzhou City,Henan Province,China The monitoring station analyzes and predicts the data of the concentration of six pollutants per hour from 2015 to 2017.In order to establish a prediction model with good performance,the missing data is first compared with the interpolation effect of random forest,MICE,kNN,MA,Kalman interpolation,and Stineman interpolation method,and finally the random forest algorithm with the best interpolation effect is selected for interpolation Build complete data.Secondly,empirical mode decomposition is used to decompose the complete data into a finite number of eigenmode functions and residual terms with similar frequencies.The eigenmode functions and residual terms are used as inputs to the NARX neural network,and the output values are accumulated to obtain the final The output prediction result.The prediction effect of the EMD+NARX hybrid model built is compared with the prediction results of the four single model ARIMA model,ets model,Holt-Winters model and SVM model.The results show that the hybrid model built in this paper has higher prediction accuracy and faster running speed,and can basically fit the fluctuation and change trend of the time series of atmospheric pollutant concentration.Therefore,it can provide some support for the prediction and research of atmospheric pollutant concentration in Zhengzhou City.