Abstract
A short-term prediction of air pollution concentration by a neural network is described. The neural network used in this paper can identify a nonlinear system whose structure is very large and complex. By using the time series data of the SO2 concentration in Tokushima, Japan, we intend to find a suitable model for predicting SO2 concentration a few hours in advance. Eight different prediction models obtained by the neural network are compared to find a suitable neural network structure and suitable input variables in the prediction model. The predicted results obtained by the neural network are compared with the results obtained by linear and nonlinear statistical models, such as a linear regression model, a linear auto-regressive model and a nonlinear GMDH model. It is shown that the neural network developed in this paper gives a better prediction result as compared with the statistical models.
