Neural Modelling of the Tropospheric Ozone Concentrations in an Urban Site

The objective of the present study is to design and develop an Artificial Neural Network (ANN) model for estimations of the ambient ozone concentrations based on meteorological and pollutant parameters. The study focuses on an urban site in the metropolitan area of Athens. The research proves that the optimal ANN is a Modular one that uses the Back Propagation Optimization Algorithm. This ANN includes a Gating Network and it has a single Hidden Layer. Two other Back Propagation ANNs with a simpler architecture reveal a good performance as well. The large amount of data records combined with the good testing results prove the generalization ability of the developed ANN. Statistical analysis techniques, such as combinations of Principal Component and Stepwise Regression Analysis, have been used for the same area in a previous study. Comparing the results of the statistical analysis to the output of the designed optimal ANN reveals that the Neural Network performs more accurately.