Improved Optimization Method for Blast Furnace Based on Robust Soft Sensor and Process Knowledge

The optimization of blast furnace ironmaking process will bring great economic and ecological benefit. However, due to the complicated industry environments and limits on production situations, conventional optimization methods have the difficulty to obtain optimal performance indexes. To optimize conflicting objectives of blast furnace, production quality and energy consumption, an improved optimization method based on robust soft sensor and process knowledge is proposed. This method firstly develops a robust performance index soft sensor based on the gated recurrent unit and huber loss function, which can capture the dynamics of blast furnace data and predict the points with huge fluctuation more accurately. Meanwhile, for the sake of improving the optimization efficiency, a process knowledge base is constructed to mine the knowledge of optimal cases. It is the basis of the innovative knowledge repair operator, which is embedded with the optimization algorithm to reduce the infeasible solutions in the primary population evolution and accelerate the convergence speed. Comprehensive experiments are performed in the blast furnace ironmaking process by means of the AICS platform of Alibaba cloud, resulting in the improvement of robust prediction performance and optimization convergence compared with other widely-used methods and a successful meet with industry demands of blast furnace process.