A Forecast Model To Predict The Next Day’s Maximum Hourly SO2 In The Site Of Priolo (Siracusa)
Free (open access)
U. Brunelli, V. Piazza & L. Pignato
The purpose of this research is to develop a pure predictive model to forecast the next day’s maximum hourly SO2 in the site of Priolo in two representative monitoring stations. We globally propose a Recurrent Neural Network (RNN) with an inherent dynamic memory to forecast the fluctuations of ground concentration of SO2 pollutant. The model uses an available time series which was recorded in the industrial site of Priolo, on the east coast of Sicily, Italy for the period between 1st April 1998 and 12th December 2001. The inputs used to train the neural network are three, where two are adimensional variable, which are obtained by the meteorological data of the site and the third is the stability classes of Thomas. This study has some important implications for health warning systems environmental management in places with high pollution concentration. Keywords: air pollution, neural network, forecasting. 1 Introduction Air pollution has a negative impact on the environmental and public health when it occurs in the lower atmosphere. In this paper the results obtained by an Artificial Neural Network to forecast the daily maximum SO2 ground concentration are presented. A neural network is able to treat information that is uncertain and incomplete like the human brain, and it has been utilised to model complex non-linear functional relationships between predictor variables. The
air pollution, neural network, forecasting.