The application of neural techniques to the modelling of time-series of atmospheric pollution data

Predicting atmospheric pollutant concentrations in both urban and industrial areas is of great significance for decision-making. This paper considers the possibility of using neural techniques to identify models for atmospheric pollutant prediction. It gives the results of short- and medium-range prediction of concentrations of O3, NMHC, NO2 and NOx, which are typical of the photolytic cycle of nitrogen. The results obtained show that neural techniques have a good capacity for modelling the phenomena under investigation as compared with the traditionally used autoregressive prediction models. The possibility of using neuro-fuzzy networks also allows the features of neural networks to be combined with fuzzy logic, thus providing automatic extraction of rule bases in the usual ‘if…then…’ form; this represents a transparent form of modelling which provides useful indications for analysis of the phenomena in question or integration with already acquired knowledge.