Impact of Active Distribution Network Modeling Approaches on Transient Stability Through Sensitivity Analysis

The increasing penetration of distributed energy resources (DERs) poses challenges to power system dynamics, particularly in assessing the impact of active distribution networks (ADNs) on transient stability. The uncertainty of renewable energy and loads, along with modern DER functionalities, complicates this evaluation. This paper proposes a systematic and efficient approach to assess the impact of ADN control parameters, modes, and modeling details under operational uncertainty. The framework is based on SHapley Additive eX-planations, treated here as a sensitivity analysis tool for machine learning (ML) models that predict locational critical clearing times. We propose a method that avoids the violation of power system physical constraints during the sensitivity analysis and is computationally efficient. Its accuracy is validated against established model-free variance-based sensitivity indices, and its robustness is demonstrated under ML-induced variability. The framework is applied to the 39 Bus System for eleven ADN modeling approaches—one static and ten dynamic—reflecting real-world installation settings. Its scalability is verified through targeted tests on a 2000-bus system. Quasi-random sampling is used to capture operational uncertainty by uniformly sampling input factors within realistic bounds. The framework provides scenario-specific explanations and global insights into transient stability margins, enabling operational awareness and decision support for operators.