A novel multi-fidelity surrogate modeling method for non-hierarchical data fusion

Multi-fidelity (MF) surrogate model has been widely used in simulation-based engineering design processes to reduce the computational cost, with a focus on cases involving hierarchical low-fidelity (LF) data. However, accurately identifying and sorting the fidelity of LF models is challenging when dealing with non-hierarchical cases. In this paper, we propose a novel non-hierarchical MF surrogate framework called weighted multi-bi-fidelity (WMBF) to solve this problem. The proposed WMBF has both the advantage of two non-hierarchical frameworks, the weighted sum (WS) and parallel combination (PC) techniques, leveraging an entropy-based weight to include multiple-moments statistical information. It offers not only a weight with more information but also a more individualized scaling function within the weighted-sum framework, additionally a more individualized discrepancy function compared with existing methods. Moreover, it provides the idea of exploiting Kullback–Leibler (KL) divergence (an entropy-based metric) to characterize uncertainty for calculating weight within the WS framework. To validate the performance of the WMBF, we conduct evaluations using several numerical test functions and one engineering case. The result demonstrates that the WMBF achieves both accurate and robust predictions with minimal computational cost.