A dynamic fuzzy q-learning controller to improve power system transient stability

This paper implements a dynamic fuzzy Q-learning (DFQL) controller to improve the transient stability of both single and multi-machine power system. The DFQL essentially combines the Q-learning-, a model free reinforcement learning strategy, with the classical fuzzy controller. The proposed controller can effectively adjust fuzzy inference parameters with the help of Q-learning. The output of the controller is being used to modify the value of a series compensator which is placed at various locations of the power system. The effectiveness of the controller has been demonstrated considering a multi-machine power system and has been compared with a bang bang controller(BBC). It has been shown that the DFQL could damp out the oscillations at a faster rate and improves the transient stability of the system.