Research On Chengdu Air Quality Index Forecast Based On Ensemble Forecast

In recent years,many cities in our country have become more and more serious in the air quality,which affects people's health travel and quality of life.Therefore,scientific analysis and prediction of air quality index(AQI)has become an essential research topic.The Air Quality Index is based on the level of pollutants in the air to determine the level.The change of air quality index is related to the emission of pollutants and the meteorological conditions that affect the ability of pollutants to sink and disperse.Therefore,the factors influencing the change of air quality index are pollutant factors and meteorological factors.In order to improve the accuracy of prediction,the predicting factors need to be screened,based on the researches both at home and abroad,this paper proposes a new screening method of pollutant factors and meteorological factors.When atmospheric pollutant emission is relatively stable,meteorological factors are important factors that affect air quality index.In the current study of air quality prediction,most meteorological factors take the historical data or the numerical forecast data as input for the prediction model.It takes the historical data as the input,ignores the meteorological forecast information and reduces the prediction accuracy.With a single numerical forecast data as the input,the singularity and chaos will result in the inaccuracy of the final forecast result.Therefore,this paper proposes to use meteorological factors based on ensemble forecast as the input of the air quality index prediction model to improve the accuracy of the forecast.At present,the research on air quality prediction model mainly includes the numerical model prediction model and the statistical prediction model,in which the prediction model widely used at home and abroad is the statistical prediction model.The neural network predictive model is widely used in statistical prediction model because of its strong nonlinear relationship fitting ability.Therefore,the qualitative and quantitative nested BP neural network prediction model is used as the prediction model of air quality index of the municipal district of Chengdu.The research contents of this paper mainly include:(1)Collecting experimental data and preprocessing the data.At the same time,the numerical forecast data of meteorological factors interpolated into the meteorological numerical forecast data of the municipal district of Chengdu.(2)Screening predictive factors.The factors affecting the change of air quality index mainly include pollutant factors and meteorological factors.The screening of pollutant factors,according to the new ambient air quality standard(GB3095-2012).The screening of meteorological factors,this paper proposes that the meteorological factor screening method adopts the transfer entropy method based on the increase of characteristic variables.At the same time,the related properties and theoretical knowledge of the transfer entropy method are sorted and analyzed,and the results show that the method is feasible.(3)Prediction factor forecast.The prediction factor forecast is mainly aimed at the meteorological factors,this paper proposes to use ensemble forecast method for meteorological factors.The meteorological ensemble forecast method based on Bayes model average(BMA),which is adapted to process the ensemble forecast values of the meteorological factors that affect the change of air quality index,to improve the accuracy of meteorological forecast.(4)The construction of a predictive model.In order to enhance the reliability of the final prediction conclusion and the stability of the predictive modeling method,on the basis of pre-work,this paper proposes a qualitative and quantitative nested BP neural network prediction model to predict the Air quality index of the municipal district of Chengdu,and the experimental results show that the predicted results of this model are more accurate than the prediction results of the single quantitative BP neural network prediction model.