Optimization of geometric parameters for high-head turbine runner using CFD and machine learning techniques

To enhance the design efficiency and operational performance of high-head hydraulic turbine runners, this study proposes an integrated and efficient framework that combines inverse three-dimensional (3D) design, computational fluid dynamics (CFD) simulations, and machine learning (ML) techniques. The framework begins with inverse 3D modeling of the runner using CFturbo software. Five geometric parameters are selected as optimization variables, including blade count, axial crown profile, and lower loop curvature controls. CFD simulations are then performed to evaluate turbine efficiency under varied parameter combinations. A backpropagation (BP) neural network-based surrogate model replaces costly CFD simulations, achieving an RMSE of 0.61% and an R2 of 0.9374, and achieves high prediction accuracy. Finally, an enhanced artificial hummingbird algorithm (CHSAHA), which incorporates chaotic mapping and spiral search strategies, is applied to optimize the runner parameters to avoid local optima. The results demonstrate that the optimized runner achieves a remarkable efficiency improvement of 10.2% to 13.7% compared with the original design, which validates the effectiveness of the optimization process. This integrated frame provides a cost-effective solution for high head turbine design, showing promise for renewable energy systems that require rapid performance optimization.