Empirical Mode Decomposition Based Deep Neural Networks for AQI Forecasting

Wei Jiang, Yuxia Fu, Fabing Lin, Jing Liu, Choujun Zhan

Abstract: The Air Quality Index (AQI) is a significant indicator that can intuitively reflect the levels of air pollution. Accurate forecasting of AQI will help governments control air pollution problems and prevent citizens from a smoggy environment. In this research, we propose a general hybrid model for short-term AQI forecasting. First, we adopt the Empirical Mode Decomposition (EMD) method to decompose historical AQI time series for extracting decomposed components as features. Then, the decomposed components of AQI and the concentration of other air pollutants, such as PM2.5, PM10 and etc., are utilized as input features to train 2 parallel 1D Convolutional neural networks (1DCNN). Finally, the output of the 1DCNN is adopted as input features for train a Long short-term memory (LSTM) network. Experimental based on datasets from 2 observation stations demonstrated that the proposed hybrid model performs the best results.