Optimal-combined Adaptive Model Based On Discrimination And Its Application In AQI And CO Prediction

After the end of the Second World War,the world’s industry ushered in rapid development.Although this has brought many positive changes to people’s lives,it has also caused more serious air pollution problems,affecting people’s daily life and travel.Therefore,the country and the government have issued many policies to strengthen the prevention,control and management of air pollution.Timely and effective air quality forecasts not only play an important role in the prevention and control of pollutants,but also provide early prediction information for human outdoor travel and industrial production,etc.Air Quality Index(AQI)is a comprehensive index for assessing air quality.Carbon Monoxide(CO)is one of the important pollutants in the air.Therefore,this research proposes an optimal-combined adaptive model based on discrimination to model and predict AQI and CO,which is called CEEMD-IOASVR-CSCA adaptive model.At present,among the various research results of air quality prediction,some directly used individual model to predict and analyze,and others chose to use combined model,but some researchers were blind when using individual or combined models and cannot achieve better prediction results.In fact,there is no universal optimal model,and there is no universal combination forecasting model.That is,sometimes the combined model will produce worse prediction result than the optimal single model.In order to solve the above problems,this research proposes CEEMD-IOASVR-CSCA adaptive model based on Complementary Ensemble Empirical Mode Decomposition(CEEMD),Intelligent Optimization Algorithm(IOA),Support Vector Regression(SVR),Cuckoo Search(CS),Classification Algorithm(CA).The specific process of CEEMD-IOASVR-CSCA adaptive model is as follows:Divide the AQI or CO time series data with 346 observations into 100 groups,use CEEMD-PSO-SVR,CEEMD-PSOGSA-SVR,CEEMD-GWO-SVR and other technologies to establish six individual models for each group,and use CS optimizes the weight of the single model,and the single model is weighted to obtain the Optimal Combined Model(OCM).By comparing the effective measure of OCM and optimal single model,100 groups of classified samples are obtained.SVM-bagging and Random Forest(RF)are used to establish discriminant classifier.For the new regression data,it is determined whether to use OCM or optimal single model for prediction based on the discriminant classifier.The prediction results show that the proposed model has high accuracy and stability.For example,in the prediction of the last group of Chongqing CO,the MAPE of the CEEMD-IOASVR-CSCA adaptive model is 6.097%,which is 2.301% lower than the 8.398% of Simple Weighted Model(SWM).