Research On The Temporal And Spatial Characteristics Of Pm_2.5 In Xi’an And Narx Neural Network Prediction

The haze weather has become a serious threat and challenge in our country.PM2.5 is one of the main pollutants that attract people’s attention in air pollutants.It is harmful and widely distributed.The research and prediction of PM2.5 is of great significance.This article aims at the current deficiencies of artificial neural networks in PM2.5 prediction.First,this paper analyzes the temporal and spatial distribution characteristics of PM2.5 and air oxidation in Xi’an based on the meteorological parameters of atmospheric pollutant concentration at 11 monitoring stations in Xi’an.Provide a basis for PM2.5 forecasts.Secondly,11 monitoring sites in Xi’an are taken as research objects.A nonlinear feedback autoregressive neural network NARX model with external input is established to predict PM2.5.Finally,based on principal component analysis and wavelet decomposition,four NARX network models are established to predict and evaluate each site.Determine the optimal network model.In order to achieve high-precision PM2.5 prediction of 11 stations in Xi’an.Provide theoretical reference for practical applications.Contribute to the development of environmental protection.The results of the study are as follows:1.There are obvious seasonal differences in PM2.5 in Xi’an.And the air quality during the heating period is worse than that during the non-heating period.From 2015 to 2017,the air pollution in Xingqing station in Xi’an was relatively low,and the Jingkai station and Gaoya station were poor.2.PM2.5 is significantly related to atmospheric oxides,and atmospheric oxidizability has a certain impact on PM2.5 levels.3.NARX neural network can predict PM2.5 more accurately,and the correlation is above 0.8.Both principal component analysis and wavelet decomposition can improve the prediction accuracy of NARX neural network.The prediction accuracy of NARX neural network based on principal component analysis is better than the model based on wavelet decomposition.4.The prediction accuracy of NARX network model based on the combination of principal component analysis and wavelet decomposition is better than the other three models.Both principal component analysis and wavelet denoising can improve the prediction accuracy of NARX network to a certain extent.The NARX network based on the combination of principal component analysis and wavelet noise reduction has the highest accuracy.5.The prediction effect of this model in each month of the year can meet the requirements of normal use.The monthly average results of each evaluation index show good results,and the correlation between the predicted value and the measured value is strong.