Reinforcement learning applied to power system oscillations damping

This paper investigates the use of reinforcement learning in electrical power system oscillations damping. The approach consists in using temporal-difference learning algorithms to control a FACTS (Flexible Alternative Current Transmission System) so as to damp power system oscillations. The proposed approach is based only on local measurements and frees itself from the knowledge of power system dynamics. An illustration is carried out on a one machine infinite bus system