Load frequency control based on reinforcement learning for microgrids under false data attacks

Load Frequency Control (LFC) operations in microgrids rely heavily on cyber networks, significantly increasing their vulnerability to cyber-attacks. These attacks pose a serious risk by compromising the security of microgrid frequency. This paper investigates the estimation and mitigation of simultaneous false data injection attacks (FDIAs) on both control and measurement signals, considering the challenges posed by the uncertainty of microgrid parameters, sudden load changes, and the intermittency of renewable energy resources. First, the problem of FDIAs on LFC has been formulated. Then, an unknown input observer (UIO) based on augmented representation is designed for the concurrent estimation of real system states, FDIAs, as well as load/generation disturbances. Afterward, a resilient LFC controller is designed to mitigate frequency discrepancies in the microgrid. The controller integrates UIO attack estimations, combined with the Reinforcement Learning-based Deterministic Policy Gradient (RL-DDPG) algorithm. RL-DDPG is trained to adaptively optimize LFC performance under varying microgrid uncertainties, FDIA, and load disturbances. This deep learning algorithm employs an actor- critic approach, combining the benefits of both value-based and policy-based reinforcement learning. The developed control ensures the frequency security of the microgrid against synchronized FDIA and internal disturbances, along with diverse operating conditions. The efficacy of the proposed control strategy is validated through various scenarios, underscoring its effectiveness and resilience in addressing the challenges posed by cyber-attacks, uncertainties in the microgrid, and fluctuations in load and renewable energy generation.