Forecasting traffic-related nitrogen oxides within a street canyon by combining a genetic algorithm-back propagation artificial neural network and parametric models

Abstract

A human-built urban street canyon is typically characterized by the presence of buildings on both sides of a road in which the air pollutants, especially those resulting from road traffic, may pose a potential threat to human health. Artificial Neural Network (ANN) differs from parametric model in that it is trained to learn a solution rather than being programmed to model a specific problem with a normal way. Therefore, they would be able to offer better practical skill to predict pollutant. Because pollutant distribution is often greatly influenced by the terrain and meteorological factors, establishing ANN has to consider effective parameters as the input neurons. In this study, traffic-related nitrogen oxides was predicted by combining a genetic algorithm-back propagation ANN and two parametric models (STREET model and OSPM model). This study took those independent parameters or components in the two parametric models into account as the input neurons. These input neurons are likely to enable ANN to reach desire simulation accuracy. Results indicated that ANN had better performance than the parametric models. Further study illustrated that the simulation had higher squared Pearson correlation coefficient (R2) up to 0.73 for validation, less simulation error, and better trend description in measured data than measured mean. Overall, this study allowed the use of ANN as a viable option for forecasting the traffic-related air pollutants within a street canyon.