An aerodynamic optimization approach based on surrogate models and dynamic genetic algorithm for a guided dual-rotor Savonius turbine

The development of distributed energy systems highlights the challenges of harnessing wind power in complex urban environments. The integrated design of airflow guides and multi-rotors, with their self-starting and augmentation capabilities, is particularly well suited for urban environments. However, the system exhibits strong coupling between aerodynamic characteristics and multiple design parameters, resulting in enormous computational demands for traditional multi-parameter optimization based on computational fluid dynamics (CFD). Thus, this study proposes a V-shaped guided dual-rotor Savonius turbine as the urban energy harvesting device. Meanwhile, an integrated surrogate-assisted optimization framework is developed to capture the complex aerodynamic interactions and adaptively expand the search space beyond fixed boundaries. Using high-fidelity CFD data, multiple machine learning surrogate models are trained and compared, and a hybrid optimization strategy is constructed to jointly optimize rotor geometry, inter-rotor configuration, and guide-plate parameters. The results show a 20.4% improvement in the power coefficient (Cp) compared to the prototype turbine, validating the synergy between structural integration and data-driven optimization. This work not only advances the high-performance design of Savonius turbines for turbulent urban wind fields but also provides a generalizable methodology for accelerating the optimization of complex distributed energy systems.