Detection and Recovery of Malicious Power System Measurements Under False Data Injection Attacks

#foothill #grid #pes #power #power-electronics #energy #converters

False Data Injection Attacks (FDIAs) have emerged as a major cyber threat to modern power systems, jeopardizing the accuracy of critical measurement data and potentially destabilizing system operations. This paper proposes a robust detection and recovery framework tailored to identify and reconstruct malicious measurements under FDIAs. Unlike traditional learning-based approaches that suffer from poor generalization under unseen conditions, our method leverages a two-stage architecture combining anomaly detection and data imputation to ensure reliable performance across diverse operating states. In the first stage, a classifier-guided detection module identifies anomalous measurements using temporal and spatial dependencies. The second stage employs a diffusion-based imputation model that reconstructs the corrupted data while accounting for cyber-physical uncertainties, such as data loss, renewable variability, and nonlinear system dynamics. The recovery process is further enhanced by a dynamic neural network adapter, which generates context-aware parameters to adapt to current system conditions in real-time. Extensive simulations validate the effectiveness of the proposed method, showing improved accuracy, robustness, and computational efficiency over conventional techniques. The framework demonstrates strong resilience even when trained and tested under different system scenarios, making it a promising solution for enhancing measurement security in future power grids.