Application of subspace methods in predicting silicon content in blast furnace ironmaking

Blast furnace ironmaking is a highly energy-intensive industrial process, and recent advancements focus on reducing energy consumption by maintaining low silicon content in hot metal. This study aims to accurately predict silicon content to facilitate timely corrective actions by blast furnace operators. We utilize a nonlinear Wiener model that comprises a linear time-invariant system coupled with a static nonlinearity, enabling precise approximation of nonlinear dynamics within the ironmaking process. By applying subspace methods for model identification, which are effective in describing industrial systems, we leverage numerical linear algebra techniques such as singular value decomposition and QR factorization. This paper discusses the integration of both linear and nonlinear dynamics into the Wiener model framework by implementing subspace base identification. Results demonstrate the efficacy of combining linear filtering and nonlinear processing, showcasing the performance benefits of this approach in predicting silicon content reliably during the ironmaking process. Our findings underscore the potential of advanced identification methods in optimizing blast furnace operations and improving energy efficiency in iron production.