A novel hybrid model for hourly PM2.5 prediction considering air pollution factors, meteorological parameters and GNSS-ZTD

With the rapid development of the economy, PM2.5 severely harms human health and social development. In this paper, a novel hybrid hourly PM2.5 prediction model, named CEEMDAN-PE-GWO-VMD-MIF-BiLSTM-AT, is proposed. Firstly, complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the hourly PM2.5 sequence. Secondly, the grey wolf optimizer (GWO) algorithm is utilized to optimize the parameters of variational mode decomposition (VMD). GWO-VMD is employed to decompose the subsequence with the first decomposition's largest permutation entropy (PE). Then, combining multiple impact factors (MIF) including air pollution factors, meteorological parameters, and Global Navigation Satellite Systemderived Zenith Total Delay (GNSS-ZTD), a Bidirectional Long Short-Term Memory network based on attention mechanism (BiLSTM-AT) is established for hourly PM2.5 concentration prediction. Finally, The proposed model predicts hourly PM2.5 in Beijing, Wuhan, Urumqi, and Lhasa to verify its performance. Compared with other models, the present model can predict hourly PM2.5 more accurately.