Predicting chaotic sequences in a blast furnace by a generalized radial basis function network

Nonlinear forecasting is applied to characterize the complexities in time series of temperature fluctuations and pressure fluctuations observed in a blast furnace. A generalized radial basis function network and a Sugihara-May predictor are used as the predictive models. The temperature sequence may be diagnosed as low-dimensional chaos according to the scaling property of the prediction error as a function of the prediction-time interval. Short-term forecasts about the temperature sequence are successfully made by the network that has learned the underlying dynamics embedded in three-dimensional delayed space.