ABSTRACT
This chapter discusses a novel hybrid neural network training method with the integration of extremal optimization (EO) and LM is presented for BOF endpoint quality prediction. There are many optimization problems in industrial process control, for example, system identification, controller parameter tuning, and online/real-time optimization. The rapid growth and huge progress in manufacturing business environment and technology leads to new requirements for control technology and system science. The primary mission of process control has extended from stability, safety, and quality to constrained optimization control for desired key performance index. The BOF is one of the key processes in iron and steel production. It is a typical complex batch chemical reactor with sophisticated thermal and chemical reaction processes, which converts liquid pig iron into steel with desired grade under oxidation conditions. The main advantage of the proposed EO-LM algorithm is to utilize the superior features of EO and LM in global and local search, respectively.
