 |
REINFORCEMENT LEARNING STRATEGIES FOR CLOSED-LOOP CONTROL IN FLUID DYNAMICS
1 : Institut P' CNRS - Université de Poitiers - ENSMA UPR 3346 SP2MI
(Pprime)
* : Auteur correspondant
CNRS : UPR3346, Université de Poitiers, ENSMA
Téléport 2 - 11 Boulevard Marie et Pierre Curie BP 30179 F86962 FUTUROSCOPE CHASSENEUIL Cedex -
France
This work discusses a closed-loop control strategy based on a pure-driven approach relying on scarce and streaming data. A continuous reinforcement learning algorithm is applied to the sensor measurements from which a Markov process model is derived, approximating the system dynamics. An implicit model of the system at hand is learned from the data, allowing for gradients evaluation and leading to quick convergence to an efficient control policy. This method is illustrated on the control of the drag of a cylinder flow.
