An online variable-fidelity optimization approach for multi-objective design optimization

Multi-objective genetic algorithms (MOGAs) are effective ways for obtaining Pareto solutions of multi-objective design optimization problems. However, the high computational cost of MOGAs limits their applications to practical engineering optimization problems involving computational expensive simulations. To address this issue, a novel variable-fidelity (VF) optimization approach for multi-objective design optimization is proposed, in which a VF metamodel is embedded in the computation process of MOGA to replace the expensive simulation model. The VF metamodel is updated in the optimization process of MOGA, considering the cost of simulation models with different fidelity and the influence of the VF metamodel uncertainty. A normalized distance constraint is introduced to avoid selecting clustered sample points. Four numerical examples and two engineering cases are used to demonstrate the applicability and efficiency of the proposed approach. The results show that the proposed approach can obtain Pareto solutions with good quality and outperforms the other four approaches considered here as references in terms of computational efficiency.