Assessment of the effectiveness of reinforcement learning in control of wall turbulence

Abstract

In recent years, reinforcement learning has been attracting attention as a framework for developing a new control strategy in a non-linear system by relating limited sensing information to actuation through deep neural network. So far, although reinforcement learning has been applied to relatively simple flow fields with limited degrees of freedom of a control input, its application to turbulent flows with large degrees of freedom has not been reported. In this study, we apply reinforcement learning to wall turbulence control for drag reduction in order to assess the effectiveness of reinforcement learning in a turbulent flow. It is shown that the current framework successfully yield a new control law which is more effective than the existing opposition control.