Optimal-Combined Model For Air Quality Index Forecasting

Air quality prediction is one of the great significance tasks for public health and pollution control.Air quality prediction can guide residents’ outing activities and the production of highly polluting enterprises.Statistical methods are commonly used in air quality prediction.At present,China has adopted AQI(Air Quality Index)for air quality forecasting which is more comprehensive and scientific.Therefore,modeling AQI series has practical significance.However,current research on AQI prediction is scarce,especially on combined prediction.For the high complex,volatile and nonlinear AQI series,this research presents a novel Optimal-Combined Model based on CEEMD(Complementary Ensemble Empirical Mode Decomposition),PSOGSA(Particle Swarm Optimization and Gravitational Search Algorithm),PSO(Particle Swarm Optimization)and combined forecasting method.The Optimal-Combined Model effectively solves the problems in combined forecasting.When the model is combined,the proposed model will first judge the single models weather they are suitable for combined forecasting.If the predictive ability of the simple model is very different,that is,there is a single model with poor prediction effect in the simple model groups,then the simple model with the best prediction performance will be selected to predict AQI;if the prediction performance of the several simple models are basically the same,the combined prediction will be used for AQI forecasting,and the weight of each simple model in the combined model will be obtained by the PSO optimization algorithm.In the modeling process,CEEMD is used for data preprocessing of AQI time series in the first place,SVR-PSOGSA and GRNN are used for modeling and prediction of the decomposed sequences in the second place,and the optimal single model or combined model is used through discriminating to predict AQI at last.AQI series forecasts of five cities in North China show that the proposed model has the highest correct rate of forecasting classifications,smaller prediction error and can be applied to the prediction of AQI in different cities with stronger generalization ability compared with the candidates.