New Feature Selection Method and Its Applications on Prediction of Hot Metal Silicon Content in Blast Furnace

In the related studies of data-driven models for predicting hot metal silicon content in blast furnace, the prediction accuracies are always depended on the feature selection methods. In this paper, a novel feature selection method based on rough set is thus proposed. First, we suggests the notion of attribute distance to analyze the differences between samples to avoid the discerning marginal problem On this basis, a new discernible matrix and the related attribute reduction algorithm are proposed. Next, we trained the hot metal silicon content prediction models based on deep learning and compared the proposed algorithm with the traditional algorithms based on the Rough set, Pearson correlation coefficient, and Grey correlation coefficient. These experimental results show the effectiveness of the proposed algorithm.