Intelligent air quality monitoring using connectionist models

The use of Artificial Neural Networks (ANNs) for emission estimation and forecasting has been recently proven to be promising. For instance, a neural network approach can account for the synergistic effects that arise from the complex interactions of several variables that affect atmospheric dispersion of gases. Unlike other modeling approaches, artificial neural networks are able to provide reliable estimates of emission rates based solely on limited information that is not known with as much certainty. The deterministic models, for instance, would require information regarding atmospheric stability, dispersion coefficients in the lateral and vertical directions, reaction mechanisms, and kinetic data. These information are usually either not well understood or only known with a certain degree of uncertainty. The objective of the present chapter is to offer a background on the use of artificial neural networks in predicting pollutants concentrations. The chapter will start with an introduction to ANNs and will discuss different algorithms used for training them. The chapter will then present several examples that will illustrate the predictive performance of the prepared network models when compared against linear and non-linear regression models, commercial simulators, and measured data.