Abstract Optimizing process parameters is essential for maintaining stable and efficient blast furnace (BF) operations under varying conditions. However, most existing methods focus on overall performance and overlook the internal states of the furnace that govern BF behavior, which can result in abnormal or suboptimal operations. This paper proposes an optimization model that incorporates AI-predicted internal states into the optimization process. The model is solved by integrating a metaheuristic algorithm with a recently developed prediction model that combines a mechanistic model and a data-driven model. Two key innovations are introduced. First, the AI-predicted shape of the cohesive zone, which largely governs BF performance, is embedded as constraints to filter out candidate solutions associated with unexpected internal states. Additionally, due to the high computational cost of mechanistic data, an optimization algorithm with a controllable search domain is proposed to guarantee prediction accuracy under sparse and uneven mechanistic sampling. Comparative analyses demonstrate that the proposed method effectively selects optimal parameters within a predefined distance. The obtained parameters are not only globally optimal in mathematics but also physically achievable, as verified by the mechanistic model. The model’s flexibility supports its use in diverse multi-objective tasks, suggesting great potential for reliable parameter optimization in an industrial environment.
Wu et al. (Thu,) studied this question.