@Article{ee.2026.077417,
AUTHOR = {Hao Yang, Shuo Feng, Bo Jin, Fang Shi, Zhenglong Sun, Xiaohan Shi},
TITLE = {Real-Time Adaptive Load Shedding Using Dual DDQN toward Criterion-Aware Post-Fault Frequency Recovery},
JOURNAL = {Energy Engineering},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/energy/online/detail/26087},
ISSN = {1546-0118},
ABSTRACT = {To ensure secure post-fault frequency ride-through in modern power systems, the grid codes specify strict frequency recovery criterion (FRC) for different regional power systems. In response to the FRC, this paper proposes a decentralized adaptive load shedding strategy to enhance frequency recovery. First, a Markov decision process (MDP) is constructed to characterize the load shedding control procedure for frequency control. The frequency evolution status used to decide load shedding actions is extracted based on the FRC, which includes the magnitude deviation and the recovery time deviation of bus frequency. The load shedding amount and time delay for each shedding round are taken as the MDP actions. Reward functions are designed by considering both the load shedding cost and the FRC requirement for safe reinforcement learning (RL). Second, a framework of offline centralized learning and real-time decentralized control based on dual agents consisting of a load-shedding agent and a time-delay agent is proposed. By utilizing an RL-based double deep Q network (DDQN) algorithm, the dual agents learn the optimal load shedding policy that responds to the FRC through interactions with the MDP. Finally, a decentralized load shedding control strategy is constructed by deploying the well-trained dual agents at load bus stations. According to the real-time frequency evolution, the strategy can adaptively determine the time, location, amount, and round of load shedding to facilitate frequency recovery while meeting the FRC. Simulations on a modified IEEE 39 test system validate the effectiveness and adaptability of the proposed method.},
DOI = {10.32604/ee.2026.077417}
}