MetaHeuristics
Previous Projects

Neural modeling of blast furnace control



Stabilization of hot metal outlet temperature is the most important objective of blast furnace control actions. Because of large lags in the system control actions must be taken well in advance, so good predictions are needed. Typically 13 input parameters are used in the Metacomp model that predicts hot metal temperature. Eight of these are tuyere inputs and the charge fed at the top. The rest give the composition of the top gas. Sucessful predictions were possible 4 hours in advance with generalized, multilayered feedforward networks. Pruning was required to improve generalization, and the time history of certain variables were also needed for good network performance.