In the optimization task of the control parameters of some industrial processes, it is necessary to explore unknown response landscape of the system by performing plural sampling of the output for input parameter combination of the system. Skilled operators has been conducting such tasks based on their experience and knowledge. In this study, the authors had formulated this problem as a machine learning process, and had developed an algorithm that sequentially selects quasi-optimal next sampling. Experimental results discovered a case that the algorithm can reduce the control parameter optimization task which requires 3 months duration into about 1 week.
