Modeling of high dimensional blast furnace system by manifold learning

The blast furnace ironmaking process is one of the most energy intensive industrial processes that consume large quantity of materials every day. Thus the modeling and control problem of blast furnace system becomes extremely important. In this paper, a two step method based on manifold learning is proposed to model the high dimensional blast furnace system. The first step is the dimension reduction step by locality preserving projection (LPP). Through LPP, the original input space is projected onto the low dimensional space and the nonlinear relationship between the low dimensional space and the output space is approximated by flexible least squares (FLS). FLS takes the time varying characteristics of blast furnace system into account by considering both the measurement error and the dynamic error. Simulation results show that the proposed method has good prediction accuracy.