Online Data-Driven Evaluation Method of Static Voltage Stability Margin for AC/DC Hybrid Power Systems Considering Uncertainty of Renewable Energy

The increasing integration of renewable energy and HVDC technology has significantly heightened the operational complexity of large-scale power systems and amplified the risk of voltage collapse. To enable rapid and effective voltage stability assessment of such highly uncertain AC/DC hybrid systems, a data-driven online probabilistic predictive framework for the static voltage stability margin (SVSM) is proposed. A LightGBM–SHAP–KDE algorithm is developed within this framework, where SHAP is employed for feature selection and explanatory analysis, LightGBM is utilized to ensure prediction efficiency, and KDE extends point predictions to probabilistic distributions, thereby enabling accurate, fast, and interpretable probabilistic SVSM prediction. Case studies on the modified IEEE 39-bus system and the Polish 2383-bus winter-peak system demonstrate that the proposed approach achieves superior accuracy and computational efficiency compared with EXtreme Gradient Boosting (XGBoost), random forest (RF), and support vector machine (SVM), and also achieves better real-time performance compared with the traditional Cumulant-based interval calculation method. Additional evaluations under PMU/WAMS data latency and asynchrony, topology changes, measurement noise and missing data further confirm the robustness and scalability of the proposed framework under complex operating conditions.