Kriging-based multi-fidelity optimization via information fusion with uncertainty

In this paper, a Multi-fidelity optimization method via information fusion with uncertainty (MFOIFU) is proposed. MFOIFU combines prediction uncertainty of kriging and model uncertainty, aiming at reducing computational cost of optimization and guaranteeing reliability of the optima. Firstly, the uncertainty of Low-fidelity (LF) and High-fidelity (HF) models is confirmed, respectively. After that, the optimal estimation theory of Kalman filter is employed to fuse information from LF and HF models. Then, the fused model is optimized and a distinctive updating strategy is presented to supplement feasible solutions. The newly introduced MFOIFU is verified through eight benchmark examples. Results showed that MFOIFU has some advantages over the Single high-fidelity optimization (SHO) method and some of the well-established multi-fidelity methods on computational expense and optimization efficiency. Finally, the MFOIFU method is successfully applied to the shell structure design of an Autonomous underwater vehicle (AUV)