Airborne Dispersion Modelling Based on Artificial Neural Networks

Artificial Neural Networks (ANNs) and airborne dispersion models are two important techniques for predicting air pollution concentrations. The purpose of this paper was to develop an integrated model that can optimise the performances of simple airborne dispersion models. The ANN dispersion model, consisting of the ANN and air dispersion model, was designed and realized. In this new model, the concentration levels produced by the air dispersion model were filtered with an ANN to account for disagreement between the actual and predicted values. The performance of the new methodology was tested by two data sets: the Prairie Grass and Copenhagen when compared with the performance of the simple air dispersion model. Simulation results showed a marked improvement for the ANN dispersion model, which indicated that the use of ANN in order to better the simple air dispersion model could be the reasonable model combination.