Parametric analysis and optimization of lift-type vertical axis wind turbines using machine learning techniques

The power coefficients ( C p ) of helical and Φ-type vertical axis wind turbines remain insufficiently explored compared with the H-type ones. A unified machine learning (ML) model was developed to conduct the parametric study of C p across the H-type, helical, and Φ-type turbines with various structural parameters and turbulence intensity ( I u ). Three-dimensional computational fluid dynamics (CFD) simulations, validated against experimental data, were conducted to generate a reliable dataset and elucidate the flow mechanisms. C p of the Φ-type turbine is insensitive to the aspect ratio ( AR ) due to minimal tip loss, whereas a larger AR of the helical turbine expands midspan regions unaffected by tip vortices, improving the maximum torque coefficient ( C m ) and C p . Higher solidity ( σ ) reduces the optimal tip speed ratio and induces larger angles of attack, slightly aggravating the dynamic stall of helical turbines and decreasing C m . Positive pitch angles ( β ), orienting leading edges inward, decrease the negative C m and increase C p . For helical turbines, increased twist angles reduce C p , and the effects of parameters on C p are mutually independent. The curvature ratio of the Φ-type turbine and I u only slightly affect C p . The particle swarm optimization algorithm incorporated with the ML model effectively improves C p for all turbines. • A unified machine learning model is proposed to predict C p for various VAWTs. • Comparative aerodynamics of H-type, helical, and Φ-type turbines are revealed. • Dynamic stall and C m under various parameters are examined via ML model and CFD. • The effects of parameters on C p of the helical turbine are mutually independent. • The XGBR-PSO model outperforms in optimizing parameters of different turbines.