A hybrid model for short-term air pollutant concentration forecasting

Abstract

The paper focuses on short-term forecasting of air pollutants including SO2, NO2, O3 and PM2.5. A hybrid model of nonlinear autoregressive with exogenous input (NARX) network and autoregressive moving average (ARMA) is applied. The NARX network is used to solve the problem of nonlinear and multidimensional while the ARMA model is aimed to improve the flexibility for different pollutants. The performance of the hybrid model is evaluated by data of pollutant concentration as basic input, and observed/forecast weather condition as exogenous input. The model overcomes the nonlinear and multidimensional problem and shows promising overall results for all the pollutants over traditional method.