Reinforcement-learning-based control of convectively unstable flows

This work reports the application of a model-free deep reinforcement learning (DRL) based flow control strategy to suppress perturbations evolving in one-dimensional linearised Kuramoto–Sivashinsky (KS) equation and two-dimensional boundary layer flows. The former is commonly used model disturbance developing flat-plate These systems are convectively unstable, being able amplify upstream disturbance, thus difficult control. action implemented through volumetric force at fixed position, performance evaluated by reduction perturbation amplitude downstream. We first demonstrate effectiveness DRL-based KS system subjected random noise. monitored downstream reduced significantly, learnt policy shown be robust both measurement external One our focuses place sensors optimally DRL using gradient-free particle swarm optimisation algorithm. After process for different numbers sensors, specific eight-sensor placement found yield best performance. optimised sensor applied directly Blasius flows, can efficiently reduce energy. Via analyses, mechanism opposition Besides, it that when instability information embedded reward function penalise instability, further improved this unstable flow.