The Research And Application Of Forecasting Technique Based On Secondary Decomposition-Ensemble Learning Paradigm And Combined Optimization Algorithm

Establishing an efficient early warning system for particulate matter(PM2.5)forecasting in the most polluted regions is significant important for environmental watchdog,which can take preventive measures to improve the air quality.Hence,a novel secondary decomposition-ensemble forecasting framework,integrating wavelet packet decomposition(WPD),complementary ensemble empirical mode decomposition(CEEMD),phase space reconstruction(PSR),least square support vector regression(LSSVR)and chaotic particle swarm optimization method combined with gravitation search algorithm(CPSOGSA),is proposed to forecast short PM2.5 concentration in this paper.Firstly,secondary-decomposition(SD)method integrating both WPD and CEEMD decomposition algorithms,is used to decompose the original PM2.5 time series into several intrinsic mode functions(IMFs).Secondly,PSR technique is implemented to choose the optimal input form for each decomposed component,which can reduce the effects of personal selective input on the prediction accuracy.Thirdly,the LSSVR model,which is optimized by CPSOGSA,is utilized to forecast all the reconstructed components.Accordingly,the predicted results of each component can be obtained.Finally,those predicted components are integrated into an aggregated output as the final prediction,using another LSSVR-CPSOGSA.For verification and illustration purposes,the proposed model is used to forecast the PM2.5 concentrations of Chengdu and Shenyang.The empirical study results show that the proposed method can outperform the benchmark methods discussed in this paper by means of the level accuracy and directional accuracy,indicating that the forecasting model based on secondary decomposition-ensemble learning paradigm and combined optimization,is more effective and applicable to the PM2.5 concentration forecasting.