Abstract (EN):
The air quality models should be able to provide early warnings before the pollution episodes occur, thus contributing for human health and environmental protection. However, the modelling of real-world processes such as air quality is generally a difficult task due to both chaotic and non-linear phenomena and high dimensional sample spaces. Several statistical models were applied to predict air pollutant concentrations, but their performance was always dependent on the time and location of the performed study. Consequently, the models should be optimized concerning their parameters and also their structures. Evolutionary computing encloses two important artificial intelligence strategies: genetic algorithms and genetic programming. Both methodologies use the same principles of the Darwinian Theory of Evolution, enabling the simultaneous optimization of the model structures and parameters. This study aims to present the results of several studies that applied these evolutionary procedures to predict air pollutant concentrations. The studies reported that these methodologies presented good performances and could be used to solve environmental complex problems. They had the ability to select the important variables that mostly influence the concentrations of the studied air pollutants. However, the main disadvantage of these approaches is the high required computation time. In some of these studies, the achieved models presented good performance in prediction of extreme values, which is useful for human health protection, as they can provide more reliable early warnings about high air pollutant concentration episodes.
Language:
English
Type (Professor's evaluation):
Scientific