A Neural Network based Reinforcement Learning Controller for Automatic Generation Control

It has been proposed recently that the control policy for Automatic Generation Control (AGC) of an interconnected power system can be learnt through a Reinforcement Leraning (RL) approach [7]. In this method, which is the first application of RL techniques for power system control problems, it was assumed that the input variables to an AGC take only discrete values. However, in a power system, quantities of interest for an AGC such as ACE, frequency deviation etc. are all continuous variables. In this paper, the earlier RL approach is extended to take care of the case of continuous state variables. We propose a Radial Basis Function (RBF) neural network to represent the control policy and propose a learning algorithm so that optimal control law for the AGC can be learnt through a training phase. The RL approach to AGC design does not need any information regarding power system model and it is simple to implement. It is shown through simulations that this new AGC performs as well as or better than other existing methods. Keywords—Power System Control, AGC, Reinforcement Learning, RBF Neural Networks.