The significance of periodic parameters for ANN modeling of daily $SO_2$ and NOx concentrations: A case study of Belgrade, Serbia

Özet

In recent decades, artificial neural networks (ANNs) have been used for the prediction of concentration of airpollutants in urban areas. Beside meteorological variables, periodic parameters, such as hour of the day or monthof the year, have been frequently used to improve the performance of ANN models by representing variations ofemission sources. In this paper, different forms of periodic parameters, i.e. smoothed cosines based approximation and normalized historical mean values, were combined with meteorological variables in order to analyze the sensitivity of the ANN model to them. Ward neural network and general regression neural network wereused and compared for the prediction of daily average concentrations of $SO_2$ and NOx in Belgrade, Serbia.Multiple performance metrics have demonstrated that models based on periodic parameters outperform thecorresponding models that used only meteorological variables as inputs. Also, a newly proposed normalizedhistorical mean $MOY_{nmv}$ (month of the year) proved to be more appropriate in majority of cases than thetraditional cosines based approximation ($MOY_{cos}$). A simple rule for the selection of the most efficient MOY formwas defined depending on their mutual correlation (r). Results have shown that if $MOY_{nmv}$ is correlated with$MOY_{cos}$ with r > 0.8, then ANN models what uses $MOY_{nmv}$ provide more accurate predictions.