Interpretable power grid overload detection with information flow-based fuzzy cognitive maps

The increasing integration of renewable energy sources (RES) brings significant challenges, such as grid overloading, which can escalate operational costs and brings a risk of system-wide blackouts. Traditional AI-based solutions often fall short due to their "black-box" nature, while many existing eXplainable AI (XAI) methods either provide unreliable post-hoc explanations or present compromised reliability and generalizability due to their vulnerability to spurious correlations. This study introduces a novel framework based on recently introduced Information Flow-based Fuzzy Cognitive Maps (IF-FCMs), a causal eXplainable Artificial Intelligence (CXAI) approach designed to overcome these limitations. By employing a two-layered architecture, this framework not only detects grid overloads but also identifies both local and global factors causing these events. Tested on the Medium Voltage Oberrhein network, IF-FCMs demonstrated predictive performance comparable to state-of-the-art models such as Explainable Boosting Machines (EBM) and Logistic Regression (LR), while offering superior interpretability. These strengths establish IF-FCMs as a reliable and powerful tool for future energy management systems.