A novel automatic generation control method with hybrid sampling for multi-area interconnected girds

Introduction: The emerging "net-zero carbon" police will accelerate the large-scale penetration of renewable energies in the power grid, which would bring strong random disturbances due to the unpredictable power output. It would affect the coordinated control performance of the distributed grids.Method: From the quadratic frequency modulation perspective, this paper proposes a fast Q-learning-based automatic generation control (AGC) algorithm, which combines full sampling with full expectation for multi-area coordination. A parameter sigma is used to balance the state between the full sampling update and only the expectation update so as to improve the convergence accuracy. Meanwhile, fast Q-learning is incorporated by replacing the historical estimation function with the current state estimation function to accelerate the convergence speed.Results: Simulations on the IEEE two-region load frequency control model and Hubei power grid model in China have been performed to validate that the proposed algorithm can achieve optimal multi-area coordination and improve the control performance of frequency deviations caused by the strong random disturbances.Discussion: The proposed Q-learning-based AGC method outperforms the convergence accuracy, speed, and control performance compared with other reinforcement learning algorithms.