Modeling of hourly NOx concentrations using artificial neural networks

Abstract

Modeling of ambient air quality is an important component of urban air quality management. Artificial neural network (ANN) modeling may offer advantages in understanding processes that follow nonlinear, complex relationships. Artificial neural network modeling is a black-box method where relationships describing complex situations are not necessarily known. The ANN models learn patterns based on historical data, and then conduct simulations based upon these patterns. The objective of this study was to evaluate the feasibility of using an ANN to predict ambient hourly concentrations of oxides of nitrogen (NOx) in an industrial corridor adjacent to Edmonton, Alberta. A standard 4-layer back-propagation network was used to predict ambient hourly NOx concentrations using industry stack emission rates, meteorological data, and traffic counts as input variables. The resulting model fit (R2 of 0.63) and precision of model prediction (root mean square error of 1.8 × 10–3 ppm as NO2) suggested that ANN modeling shows promise for predicting NOx behaviour; however, further work is necessary to improve its forecasting ability.Key words: urban air quality, airshed, modeling, artificial neural network.