Modeling for prediction steel mechanical property based on KFA-KPLS

Based on Kernel Factor Analysis (KFA) and Kernel partial least square (KPLS), a modeling method for the prediction of steel mechanical property is proposed in this paper. In order to eliminate the heterogeneity among variables in the hot rolling process, the KFA is used for latent factor load vectors, then the variables with bigger factor load are clustered into sub-clusters, which are reorganized by KPLS with the objective variable. Finally, the results of all KPLS were used as the input of the KPLS model to predict the steel mechanical property. An application study is carried out on the real production data acquired from a steel-making plant. The experimental result shows that the precision of the presented method is greatly improved.