A sequential multi-fidelity surrogate-based optimization methodology based on expected improvement reduction

This paper presents a novel computation-aware multi-fidelity surrogate-based optimization (MFSBO) methodology and a new sequential and adaptive sampling strategy based on expected improvement reduction (EIR). Given a fixed computational budget in each iteration, the EIR-based infill determines the data source and samples of infill by hypothetically interrogating the effect of samples and simulation fidelity on reducing the expected improvement, and enables low-fidelity batch infills within a dynamically varying trust-region to improve exploration as needed to accelerate the MFSBO process. The co-Kriging method is utilized to combine the data from different data sources with varying fidelities and computational costs. The EIR-based infill is then compared with other infill strategies in terms of convergence rate and design accuracy. Results indicate that the proposed method achieves a faster convergence rate and more accurate optimal design during MFSBO for all case studies.