Solving the Chemical Mass Balance Problem Using an Artificial Neural Network

Abstract

A back-propagation artificial neural network (BP ANN) is proposed as a receptor modeling method for solving air pollution source apportionment problems. In order to examine the utility of this method, the simulated aerosol composition data generated by the National Bureau of Standards (NBS) for the EPA workshop on mathematical and empirical receptor modeling held at Quail Roost, NC, in 1982 were examined. Based on the assumption of mass conservation, training sets containing the linear mixing fractions for ambient samples were constructed from input source profiles for the network's learning. The NBS data sets were the prediction sets. Because of the good generalization property of BP ANN, satisfactory prediction results were obtained for NBS data sets I and II. Moreover, when a training set containing as many as possible emission sources some of which were not necessarily active was used, the network was able to identify the actual active sources and quantify their source contributions.