Data-Enhanced mechanistic modeling for Multi-Variable thermal state prediction of the blast furnace process

Accurately predicting the multi-variable thermal state of the blast furnace is crucial for operational stability and high-quality hot metal production, yet it is challenged by the complexity and measurement delays of the process. This study develops a data-enhanced mechanistic modeling framework to address this challenge. The framework synergizes a first-principles mechanistic model, developed in Aspen Plus®, with a data-driven residual correction model. The mechanistic model provides robust but approximate initial estimates of key thermal state parameters. A sophisticated data-driven model, incorporating Gated Recurrent Units and a Multi-Head Self-Attention mechanism, is then employed to learn and correct the complex, non-linear residuals between the mechanistic outputs and actual industrial measurements. Validated on industrial data, the proposed hybrid model demonstrates superior performance in predicting four key variables, i.e., hot metal silicon content, slag basicity, top gas CO utilization rate, and top gas temperature, achieving hit rates exceeding 90% within predefined error margins. The results highlight the potential of this hybrid approach to provide a robust and accurate solution for real-time thermal state prediction in ironmaking processes.