Dynamic optimal CPS control for interconnected power systems based on SARSA Algorithm

Abstract

In order to get better control performances in a relatively dangerous environment, a novel dynamic optimal CPS control method for interconnected power systems using on-policy reinforcement learning (RL) algorithm-SARSA RL algorithm is introduced in this paper. This controller realizes online learning and optimization of the acceptance rate of CPS values by a reward function which is constructed by the system CPS values and a closed loop which is constructed by CPS control actions. Comparing with off-policy RL algorithm-Q-learning, SARSA is better in convergence ability and safer in selection of policy. It is shown in the simulation experiment that more effective CPS values can be obtained by the controller using SARSA RL algorithm than that by using Q-learning algorithm. © 2014 Springer Science+Business Media New York.