Gaining Insight into Short-Term Voltage Stability Assessment Utilizing Model-Agnostic Explanations Algorithm

With the increasing penetration of renewable energy and dynamic loads, short-term voltage stability (STVS) in modern power systems has become a growing concern. To address this challenge, data-driven methods have been widely employed for rapid and accurate STVS assessment. However, these models often lack transparency, hindering their adoption in real-world operations. This paper proposes an interpretable STVS assessment framework that integrates the eXtreme Gradient Boosting (XGBoost) classifier with Shapley Additive Explanations (SHAP) to provide both global and local insights into model predictions. First, an STVS assessment model is developed based on XGBoost. In order to improve the interpretability of the model, a SHAP-based interpretability framework is developed for STVS assessment. By quantifying feature contributions, the framework reveals the physical factors influencing voltage stability and enhances model transparency. Case studies on a modified 33-bus distribution network validate the effectiveness of the proposed method.