Applying ICA on neural network to simplify BOF endpiont predicting model

This paper proposes an improved method to modeling the dynamic process of basic oxygen furnace (BOF) and the main idea is simplification. Aiming at the problem that it is usually difficult to build a precise endpoint dynamic model because of the high dimensional input variables which affect the final results - carbon content and temperature, this paper builds endpoint carbon content prediction model and endpoint temperature prediction model separately. First, the more important variables are chosen for two models by analyzing the mechanism. The independent component analysis (ICA) is applied to reduce the input dimension for temperature prediction model. Results show that the model simplification is essential and effective.