Reinforcement-learning-based actuator selection method for active flow control

This paper addresses the issue of actuator selection for active flow control by proposing a novel method built on top reinforcement learning agent. Starting from pre-trained agent using numerous actuators, algorithm estimates impact potential removal value function, indicating agent's performance. It is applied to two test cases, one-dimensional Kuramoto–Sivashinsky equation and laminar bidimensional around an airfoil at $Re=1000$ different angles attack ranging $12^{\circ }$ $20^{\circ , demonstrate its capabilities limits. The proposed actuator-sparsification relies sequential elimination least relevant action components, starting fully developed layout. relevancy each component evaluated metrics based function. Results show that, while still being limited this intrinsic paradigm (i.e. elimination), patterns obtained policies performances allow us draw accurate approximation Pareto front versus budget.