Three hours ahead prevision of SO2 pollutant concentration using an Elman neural based forecaster

Indoor air quality near the industrial site is tightly joined to pollutant concentration level, since outdoor pollution heavily influences air quality and, consequently, inhabitants health. A pollution management system is essential for health protection. Automatic air quality management systems have became an important research issue with strong implications for inhabitants’ health. In this paper an automatic forecaster based on neural networks for SO2 concentration prevision is proposed. The analyzed area covers different small towns near the industrial site of Priolo, in the south of the world. Among these towns, Melilli was the first town in Italy that was evacuated for high level of pollutant concentrations. In the paper, a traditional stochastic method and several neural models are also compared. Overall, the results of the simulation show that the employment of a neural network forecaster is the most efficient tool to follow the big variations of pollutants concentration when thermal inversion height is taking place. In particular, an Elman neural network shows interesting results in 1, 2, and 3 h ahead forecasting of SO2 concentration, doing the proposed forecaster a powerful tool for both pollution management and health warning systems.