Multi-stage accelerated structural optimization framework integrating multi-fidelity neural networks and evolutionary strategies

To overcome the high computational cost of finite element analysis in optimizing complex structures, this article proposes a multi-fidelity evolutionary framework integrating hierarchical decision making and collaborative exploration. At its core is a multi-fidelity neural network that fuses low- and high-fidelity data to predict performance efficiently. A dual-loop optimization mechanism is employed: a main loop ranks and selects elite solutions for expensive high-fidelity evaluation, while a virtual loop explores promising regions using surrogate models, guided by the main loop’s best solutions. An online incremental learning strategy continuously refines the predictive model, enhancing adaptability. Experimental results demonstrate that the proposed method significantly improves convergence speed and solution quality compared to traditional evolutionary strategies, providing a viable technical approach for computationally expensive engineering optimization problems.