Integral Reinforcement Learning-Based Adaptive Optimal Automatic Generation Control

The growing penetration of renewable energy sources highly increases uncertainties in power systems, posing great challenges to system frequency regulation. Automatic generation control (AGC) serves as an essential frequency regulation approach, which is important to maintain frequency security. In this paper, the AGC is improved by using a novel adaptive optimal control algorithm. The proposed adaptive optimal control method uses policy iteration (PI) algorithm with an actor-critic neural network scheme. Considering the timevarying power system operation state, integral reinforcement learning technique is embedded in the proposed control method, which could provide feasible approach for adaptive optimal control without accurate dynamic models of power systems. An excitation relaxed adaptive law is applied in the critic neural network design to guarantee the learning efficiency. Simulation results validate the effectiveness of the proposed method.