Physics-Aware and Storm-Adaptive Machine Learning Clustering of Offshore Wind Regimes

Resumen

Assessing the prevailing wind behaviors at an offshore site is important to optimizing energy capture and overall performance of future sustainable wind power installations. This manuscript presents an Adaptive Weighted Storm-Aware Clustering (AWSC) framework, which uses physical insights to perform unsupervised learning for offshore wind regime identification. The AWSC framework uses thermodynamic feature weighting, dual-path storm separation and adaptive density refinement, thereby generating physically interpretable and statistically compact wind regime clusters. Offshore sites in four distinct locations around the world were examined. The locations included the Pacific Trades/Tropics, the North Sea and the Arctic regions. Results were generated using ERA5 reanalysis data at a 0.25° spatial resolution. The validation results show that AWSC achieves equal performance across physical and statistical evaluation metrics, with Silhouette coefficients ranging from 0.54 to 0.63 and Davies–Bouldin indices ranging from 0.43 to 0.64, indicating well-defined clusters that preserve their aerodynamic and thermodynamic properties. The method successfully separated storm patterns from background data without compromising the overall structure, and it achieved equal variable importance through feature re-weighting ( ≈ 1.05). The research demonstrates that AWSC delivers a method for wind regime classification which combines physical understanding with efficient computational processing to link data analytics with atmospheric science. The developed framework provides a foundation for worldwide meta-clustering operations, digital twin systems, and energy and exergy integration, which will lead to AI-based sustainable offshore wind energy systems.