ABSTRACT
Reinforcement Learning (RL) approaches are designed to select the policies that minimize the objective function in dynamic learning environment. This chapter aims to investigate how distributed intelligence can be used to control modern power systems. Traditional control systems for power grids were designed to handle large production units operating under central control. The adaptive critics implementation capitalizes on simple layers of neural network structures and tuning laws, which makes it more attractive than other complicated solution structures used for other differential graphical games. The constrained graphical game is a special type of the standard game, where the policies of the nodes are constrained and the communications between the nodes are done via a communication graph topology. An off-policy RL algorithm is proposed to solve a cooperative control problem using a game’s theoretic frame-work, where a behavioral policy is used for learning purposes.
