強化学習を用いた摩擦抵抗低減のための壁乱流制御則の開発

抄録

Recently, reinforcement learning has attracted much attention in the area of flow control because of its ability to obtain a long-term optimal policy. So far, it has been applied only to low-dimensional flows with simple actuations. In this study, we apply reinforcement learning to wall turbulence control in order to determine complex spatio-temporal distribution of wall blowing and suction for reducing skin friction drag based on near-wall sensing information. It is demonstrated that the control policy obtained from the present study exhibits unique relationship between sensing information and a control input, and achieves better drag reduction performance than the well-known opposition control.