Forecasting of ozone pollution using artificial neural networks

Purpose: The objective of this study is to develop and validate a neural-based modelling methodology applicable to site-specific short- and medium-term ozone concentration forecasting. A novel modelling technique utilizing two feed forward artificial neural networks (FFNN) is developed to improve the performance of time series predictions. Design/methodology/approach: Air pollution and meteorological data were collected for one year in two locations in Kuwait. The hourly averages of the data were processed to generate a covariance matrix and analyzed to generate the principal component method. A two-FFNN model is then used to predict the actual data. Findings: The newly developed model improves the prediction accuracy over the conventional method. Owing to the presence of noise and other minor disturbances in the data, shorter-range modelling gives better modelling results. Originality/value: A novel modelling technique is developed to predict the time series of zone concentration.