Research And Implementation Of PM2.5 Concentration Prediction Based On Neural Network

People’s awareness of PM2.5 prevention and control is constantly strengthened.Accurate and timely PM2.5 concentration prediction is important for air pollution prevention and control,as well as people’s travel health.Based on the air quality in Xi ’an history data,this thesis uses the concentration of PM2.5 impact factors and the historical data based on BP neural network to establish a methods of concentrations of PM2.5 hours multistep prediction.It optimizes the forecasting model based on the season.Then this thesis use two kinds of prediction model,with the help of crawler technique to design and carry out a PM2.5concentrations real-time prediction system.This thesis proposes a multistep prediction of PM2.5 hourly concentration.It combines the influencing factors of PM2.5 concentration with its K-order historical concentration value as the input characteristics of the prediction model.First,the air quality data of Xi ’an in recent one year were obtained and preprocessed.Then,the correlation between PM2.5 and PM10,SO2,NO2,CO and O3 was analyzed by drawing correlation scatter diagram and correlation coefficient analysis.Results show there is a certain correlation between PM2.5 and these influencing factors.The correlation degree from high to low is CO,PM10,SO2,NO2 and O3.Aamong which O3 is negatively correlated with PM2.5.The influencing factors of PM2.5concentration were determined by correlation analysis.The lag order K of PM2.5concentration value and the parameters of BP neural network were determined by repeated trial and error experiments.And the PM2.5 concentration prediction model of BP neural network was established.Because of the lack of seasonal characteristics in the current PM2.5 prediction model,this thesis proposes to build a seasonal PM2.5 concentration prediction model.First,through statistical analysis of the daily mean value,maximum value and daily variation rule line chart of PM2.5 concentration in different seasons in Xi ’an,it is concluded PM2.5 concentration in this region does have significant seasonal characteristics.For example,PM2.5 concentration in winter is much higher than that in other seasons.And the daily variation curve chart has obvious peak and valley values compared with other seasons.Then,based on the seasonal characteristics of PM2.5 concentration,the earned data sets are classified according to the seasons.The matching training sets of different seasons are used to train different BP neural network prediction models for the season.Besides,the two PM2.5 concentration prediction models are compared and analyzed.Compared with the overall prediction model,the mean absolute error of the seasonal prediction model is reduced by 34%.And the mean square error is reduced by 58% and the R-squared error is increased by 5.2%.In this thesis,a real-time PM2.5 prediction system is built based on the above two prediction models combined with crawler technology.It statistics and analyzes historical data of the air quality in Xi ’an.It displays Xi ’an real-time air quality data from air quality monitoring sites.And it concentrates PM2.5 hour forecasts of monitoring sites.According to the air quality data and PM2.5 predicted,it gives corresponding health and travel safety tips to help people easier to master air quality situation.That reduces the air pollution damage to human body health.The development of the system is based on Python development language,Djangoweb framework and Mysql database.