As demands for aerodynamic refinement continue to rise, more design variables are required for parametric modeling, resulting in a dramatic increase in design space dimensions and a corresponding decline in the efficiency of aerodynamic optimization and the predictive accuracy of the surrogate model. To address this issue, an efficient aerodynamic optimization framework for high-dimensional design spaces is proposed. This framework integrates a locally linear embedding technique of manifold learning for nonlinear dimensionality reduction, and an improved multiobjective particle swarm optimization algorithm for global search, enhanced by an ensemble surrogate model. The results indicate that the proposed method successfully extracts low-dimensional manifolds from the original design space, achieving effective dimensionality reduction, which significantly improves optimization efficiency while maintaining high predictive accuracy of the surrogate model. Compared to conventional surrogate-based optimization approaches, the proposed framework reduces training costs and strengthens the performance of the ensemble surrogate model in high-dimensional aerodynamic design.
Yan et al. (Mon,) studied this question.