Real wind data clustering for optimal design of a 12-slot/10-pole PMSG in wind power applications

Abstract

The growing demand for renewable energy highlights wind power as a key source of sustainable electricity. However, the fluctuating nature of wind poses challenges for turbine operation and generator design, impacting system efficiency. To address this, extensive wind datasets combined with modern analysis methods offer practical solutions. In particular, artificial intelligence (AI) techniques such as machine learning and clustering can model real wind conditions and support improved design strategies. This paper introduces a data-driven approach to wind generator design, moving beyond the traditional single operating point by integrating real-world wind profiles at the early design stage. Several clustering methods are evaluated statistically to determine the most effective representation of wind variability. The chosen method is applied to a 25-kW surface-mounted permanent magnet synchronous generator (PMSG), optimized using surrogate modeling and finite element analysis (FEA). The design is further validated with FEA simulations and experimental tests on a 2-kW prototype.