Application Of The Nonlinear Black-Box Model Transparency In Iron And Steel Industry

Generally, the data-driven models applied in blast furnace are black-box models. However, the results obtained from a black-box model are incomprehensible from the application point of view. It is difficult to apply the results to the actual control. Therefore, apply the black-box model transparent technology on the blast furnace has good practical significance.Firstly, this paper has clustered the silicon content into five clusters to determine the controlled bound of the silicon content with the cluster centers and developed a fuzzy-based SVMs multi-classifier to address a3-class classification problem of the hot metal silicon content from the control viewpoint. In order to prevent over-fitting and simplify the model, the idea of fuzzy entropy-based feature selection is introduced in our paper. It is proved that feature selection is valid to simplify the model and even the model accuracy has improved to some extent. However, despite its kinds of advantages, the opaqueness of the trained nonlinear SVM model is an unbreakable barrier. Therefore, some practicable blast furnace rules are extracted from the constructed SVMs model with the CART algorithm. The encouraging agreements between the predicted values and the real ones reveal the extracted rules can work for the blast furnace system well. Compared with the original SVMs black-box model, the current extracted rules can yield not only the label of classifying the silicon content, but the reason of producing it, to some extent, they can provide some referenced control span guidance in a comprehensible way for blast operators together with their rich practical experience. Wholly speaking, these results can better serve for making control decision on the blast furnace operation.