A reinforcement learning approach to dynamic optimization of load allocation in AGC system

A Reinforcement Learning (RL) method applied to the dynamic load allocation in AGC system is presented. The problem can be modeled as a Markov Decision Process (MDP). The Q-learning algorithm as a model-free learning algorithm is introduced. It learns an optimal action strategy by experience from exploring an unknown system and getting rewards. Rewards are chosen to express how well actions control the system. The applications of the Q-learning algorithm to the two-area power system model and China Southern Power Grid model are presented. The case study shows that the Q-learning algorithm enhances the performance of AGC system under CPS.