Cyber resilient machine learning models for estimating synchronous machine excitation current: Mitigating the impact of DoS attacks

The accurate estimation of excitation current in Synchronous Machines (SMs) is essential for maintaining the stability, efficiency, and power factor correction of power grids. With the increasing reliance on Machine Learning (ML) models for this task, the vulnerability of these models to cyber attacks, particularly Denial of Service (DoS) attacks, has become a critical concern. This study explores eight commonly used ML models namely Gradient Boosting, Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Random Forest, Long Short-Term Memory (LSTM), Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and ridge regression in estimating excitation current under normal, DoS attacked and defense scenarios. The evaluation uses key metrics, including Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination ( R 2 ) to assess model performance. The results demonstrate that the tree-based models, especially Gradient Boosting and Random Forest, deliver superior performance with normal data, outperforming other models in terms of accuracy. However, under attack condition, these ML models exhibit significant performance degradation increasing the RMSE up to 6295%. A defense strategy has been proposed to mitigate this attack’s effect, incorporating a detection system to identify and replace attacked data with predicted values. This approach not only restores model performance but also improves the reliability of estimation systems under attack condition, reducing RMSE by up to 2725%. This study contributes to the development of robust and cyber resilient estimating methodologies, offering practical implications for the advancement of ML applications in power grid systems and beyond. • Comparative study of 8 ML models for accurate SM excitation current estimation. • Employing DoS attacks to evaluate the ML models’ vulnerabilities and robustness. • Proposing a defense strategy to reduce the impact of DoS attacks.