Detection and Classification of False Data Injection Attacks in Power Grids Using Machine Learning and Hyperparameter Optimization Methods

False Data Injection Attacks (FDIA) are a significant threat to power systems, compromising system integrity through data corruption and false information injection, leading to severe impact on system state estimation and potential extensive damage. This paper proposes multiple machine learning models for stealthy FDIA detection, accounting for various attack types that a system may encounter. Hyperparameter optimization is introduced to mitigate the impact of imbalanced training data on model performance. The proposed models are comprehensively evaluated against different FDIA types, considering the heterogeneity of these attacks. Experimental results demonstrate the effectiveness of hyperparameter optimization in improving FDIA detection and preventing potential undetected attacks. Additionally, an analysis is conducted to assess the likelihood of different attack types (ramp, step, poison, and freeze), evading detection by the models. This analysis aids in understanding the magnitude of threat posed by each attack type and identifying the models that excel in detecting specific FDIA types.