Artificial Neural Network Modeling and Forecasting ofHourly PM2.5 in urban air in Edmonton

摘要

Particulate Matter 2.5 (PM2.5) is formed due to anthropogenic (man-made) or biogenicsources (natural). Studies have shown a strong correlation between PM2.5 and healtheffects such as respiratory and cardiovascular problems. Artificial Neural Networks(ANNs), an artificial intelligence (AI) technique, has the ability to generalize historicaldata patterns and make inferences about future trends. In this research, an ANN techniquewas used to model PM2.5 for the city of Edmonton, Alberta, Canada. Two models werebuilt, one for virtual monitoring and the other for forecasting. Because of the seasonalvariation of PM2.5 in Edmonton, season-specific versions of the ANN models weredeveloped for both virtual monitoring and forecasting purposes. Virtual monitor modelswere able to make good predictions for all the seasons, with the exception of summer.The addition of one-hour lagged input PM2.5 data improved the prediction capability ofthe virtual-monitor ANN models. However, the addition of one-hour lagged input PM2.5data did not improve the predictive capability of the forecast ANN models except for thespring season. Overall the models developed using a systematic approach performedbetter than the other models in the literature.