Prediction of Ozone Concentration in Ambient Air Using Multilinear Regression and the Artificial Neural Networks Methods

Abstract

This article presents the results of the statistical modeling of the ground-level ozone concentration in the air in the close vicinity of the city of Zrenjanin (Serbia). This study is aimed at defining the dependence of ozone concentration on the following predictors: SO2, CO, H2S, NO, NO2, NOx, PM10, benzene, toluene, m,p-Xylene, o-Xylene and ethylbenzene concentration in the air, as well as on the meteorological parameters (the wind direction, the wind speed, air pressure, air temperature, solar radiation, and RH). Multiple linear regression analysis (MLRA) and artificial neural networks (ANNs) were used as the tools for the mathematical analysis of the indicated occurrence. The results have shown that ANNs provide better estimates of ozone concentration on the monitoring site, whereas the multilinear regression model once again has proven to be less efficient in the accurate prediction of ozone concentration.