Application of artificial neural network for emission prediction of dust pollutants

Abstract

We study the application of artificial neural network (ANN) for predicting suspended particulate concentrations in urban air, taking into account meteorological conditions. Calculations are based on pollution measurements taken in the city of Radom, Poland, in the period 2001–2002. PM10 emission and primary meteorological data, which were obtained from IEP in Radom, were used to train and test the application of network. Two different methods of emission calculation are applied. Firstly, ANN method based on multilayer perceptron with unidirectional information flow is used. Secondly, a hybrid model based on a modified Gaussian model of Pasquille's type and ANN with radial base function (RBF) is applied. Network architecture and transition function types are described. Statistical assessment of the obtained results is made. In addition, hybrid model results are compared with emission calculations of dust pollution based on the Gaussian model, including various methods of calculation of pollution dispersion coefficients. Copyright © 2006 John Wiley & Sons, Ltd.