Deep Reinforcement Learning–Assisted Cubature Kalman Filtering for Robust Multi-Rate Dynamic State Estimation Under False Data Injection Attacks

Multi-rate measurements from heterogeneous sensors such as PMUs and SCADA pose significant challenges for dynamic state estimation (DSE), especially under non-Gaussian noise and false data injection attacks (FDIAs). Traditional Kalman-filter-based estimators with fixed parameters often suffer performance degradation in such scenarios. This paper proposes a DRL-assisted robust cubature Kalman filtering framework for multi-rate DSE, where a TD3 agent adaptively regulates measurement reliability by online scheduling of the effective measurement covariance. The proposed approach integrates asynchronous PMU/SCADA updates within a unified CKF structure and enhances robustness through residual-aware parameter adjustment. Simulation results on the IEEE 39-bus benchmark demonstrate that the proposed method consistently improves estimation accuracy and residual behavior compared with fixed-parameter and heuristic adaptive baselines under clean, non-Gaussian, and FDIA conditions.