PM_2.5 Forecasting Using A New Hybrid Model Based On Decomposition Extraction Prediction And Ensemble

Particulate matter(PM_2.5)is fine particle with an aerodynamic equivalent diameter less than or equal to _2.5 microns.High concentrations of PM_2.5 can cause serious problems because PM_2.5 is reached the critical value of alveolar and carry many harmful organic and inorganic molecules.Thus,forecasting of PM_2.5 concentrations plays a significant role in the reduction and control of PM_2.5 emission.Based on the princip le of “decomposition-extraction-prediction-ensemble”,this paper proposes a new FEEMD – KPCA – ELM（Fast ensemble empirical mode decomposition-Kernel principal component analysis-Extreme learning machine）model for forecasting PM_2.5 concentrations.The FEEMD part decompose original PM_2.5 series into several intrins ic mode functions（IMFs）,while the KPCA part extract the common latent factor for simplifying the complex data.This model predicts each principal component with ELM individually and then integrating all predicted principal components for the ensemble result as the final prediction by another ELM.Ljung-Box Q-Test,a method for white noise test,is utilized to identify the integrity of information applying which based on extracting data series.The proposed hybrid method is examined by forecasting the daily data of PM_2.5 in two major cities of China（Shanghai and Taiyuan）.The experime nt a l result indicates that the developed hybrid model with fast and accurate prediction is superior to other benchmark models and suggests it can be used to develop advanced warning systems of PM_2.5.