Application of Neural Network to Predict Sulphur Content in Hot Metal

A model for predicting the sulphur content in hot metal based on neural networks is introduced. Blast temperature, blast flux, top temperature, burden, coal injection rate, sulphur content in ore, sulphur content in coke, sulphur content in coal and silicon content of last tap were selected as inputs. The inputs were treated with time lag to improve prediction. Some methods were adopted to resolve the problem of local convergence and long learning time of BP neural network. The predicted results indicated that the hitting rate was 70.7% when the absolute error was less than 0.001, and the hitting rate was 90% when the absolute error was less than 0.005. Thus the validity of the model was proved.