A novel hourly PM2.5 concentration prediction model based on feature selection, training set screening, and mode decomposition-reorganization

Abstract

Accurate prediction of PM2.5 and other air pollutants concentration can provide early warning information for sustainable urban pollution control, urban construction and travel planning. In this paper, combined with feature selection, training set selection, mode decomposition and reorganization, machine learning, a new PM2.5 concentration hybrid prediction model is established. Firstly, historical data were screened by random forest (RF) and grey system approximation model (GSA). Secondly, the processed data is decomposed by time varying filtering based empirical mode decomposition (TVFEMD). Then, the extreme learning machine (ELM) optimized by moth flame optimization algorithm (MFO) is used for prediction. Based on the data of four cities in Beijing Tianjin Hebei region, the following conclusions can be drawn: (1) The effectiveness and robustness of the proposed model are verified, and the evaluation indexes are the best. (2) RF-GSA can effectively improve the quality of training set. (3) Mode decomposition and reorganization can effectively improve the prediction accuracy. The model can provide a reference for government policy-making and residents travel.