Leveraging Explainable Extremely Randomized Trees Model for Poisoning Attack Detection in Power Grid Stability Assessment

With the increasing integration of digital technologies and communication networks, the power grid has become more vulnerable to cyber-attacks. Among these threats, false data injection attacks (FDIAs) have emerged as a significant concern. Developing distinctive models using power grid datasets that include poisoning attacks proves to be a complex task, owing to the scarcity of comprehensive data and the tendency of conventional classifiers to be biased towards the majority class. To address this gap, this paper introduces a detection and defense model against poisoning attacks. An extremely randomized tree (ET) model is used to simulate FDIAs targeting the state estimation process, which is crucial for stability assessments. Moreover, an oversampling technique is employed to reduce the effect of the imbalanced dataset. The paper explores the susceptibility of electrical grids to such threats by leveraging the ET model to detect irregularities in data patterns. This paper proposes a machine learning-based detection algorithm that demonstrates high accuracy in identifying FDIAs. The robust state estimation method is proposed to significantly reduce the impact of false data, ensuring more accurate grid stability assessments.