This paper presents a model of the local thickness of burden layers in the ironmaking blast furnace using neural networks. Local thickness estimates, on which the neural models are trained and verified, have been obtained from stockrods, which measure the burden (stockline) level close to the furnace wall. The relation between the layer thickness and variables, such as stockline level and movable armor settings, has been described by a recurrent network. By incorporating this knowledge in a simplified scheme considering the practical contraints of the charging process, a hybrid model is formed. The hybrid model can be used to yield insight into the dynamics of the layer formation process, since it makes it possible to consider the effect of the stock level, and the histories of the layer thickness and movable armor patterns.
展开▼
