Аннотация
Accurate modeling of wind speed probability density (WSPDs) distributions is important for assessing renewable energy potential. In this paper, we introduce the Shape-aware Hybrid Adaptive-bandwidth KDE (SHA-KDE) method, a new nonparametric modeling approach. It is designed to preserve local detail around peaks of a distribution while ensuring stability in data-sparse regions. SHA-KDE combines curvature-based bandwidth and nearest-neighbor bandwidth adaptation strategies using an adaptive mixing weight. This enables adaptive local smoothing to fit the data’s real features, reducing the need for manual tuning. Experiments on datasets from 10 meteorological stations in Vietnam show that SHA-KDE performs better than traditional Weibull, fixed-bandwidth KDE (KDE), and adaptive-bandwidth KDE (AKDE) models as measured by Cramér-von Mises (CvM), Anderson-Darling (A-D), and Log-likelihood. When combined with the power curve of the Vestas V110-2.0 turbine, the proposed framework yields screening-level wind energy estimates, with an indicative net capacity factor (CF) of approximately 40% and an indicative annual energy production (AEP) of approximately 7.0 GWh per turbine under the assumed 15% loss factor. This study provides a reliable statistical tool for planning and optimizing wind power projects in Vietnam and other regions with complex wind patterns.