強化学習を用いたシステムのモデリングと制御(<小特集>制御工学への知能科学からの接近)  [in Japanese] System Modeling and Control Applied by Hierarchical Reinforcement Learning(<Special Feature>Intelligent Systems Science Approach to Control Systems Science)  [in Japanese]

In this paper, it is described that the system modeling and predictive control method based on the hybrid type reinforcement learning and its application results. The hybrid type reinforcement learning is consists of two learning layers. The linear combination of the state-value functions in the lower layer is taken into the secondary reward of the upper layer. It is applied to make a force model of the plate rolling in the steel making process. The constructed model can predict rolling force precisely. In the predictive control, the optimum control input is chosen from the provided inputs by changing the plural reinforcement learning models. The control system using the reinforcement learning is applied to control a dynamical system. Effectiveness is confirmed by numerical simulation.