Neural networks and periodic components used in air quality forecasting

Forecasting of air quality parameters is one topic of air quality research today due to the health effects:: caused by airborne pollutants in urban ar eas. The work presented here aims at comparing two principally different ne ural network methods that have been considered as potential tools in that a rea and assessing them in relation to regression with periodic components. Self-organizing maps (SOM) represent a form of competitive learning in whic h a neural network learns the structure of the data. Multi-layer perceptron s (MLPs) have been shown to be able to learn complex relationships between input and output variables. In addition, the effect of removing periodic co mponents is evaluated with respect to neural networks. The methods were eva luated using hourly time series of NO2 and basic meteorological variables c ollected in the city of Stockholm in 1994-1998. The estimated values for fo recasting were calculated in three ways: using the periodic components alon e, applying neural network methods to the residual values after removing th e periodic components, and applying only neural networks to the original da ta. The results show ed that the beat forecast estimates can be achieved by directly applying a MLP network to the original data, and thus, that a com bination of the periodic regression method and neural algorithms does not g ive any advantage over a direct application Of neural algorithms. (C) 2001 Elsevier Science Ltd. All rights reserved.