An active learning variable-fidelity metamodelling approach based on ensemble of metamodels and objective-oriented sequential sampling

Abstract

ABSTRACTComputational simulation models with different fidelity have been widely used in complex systems design. However, running the high-fidelity (HF) simulation models tends to be very time-consuming, while incorporating low-fidelity (LF), inexpensive models into the design process may result in inaccurate design alternatives. To make a trade-off between high accuracy and low expense, an active learning variable-fidelity (VF) metamodelling approach aiming to integrate information from both LF and HF models is proposed. In the proposed VF metamodelling approach, a model fusion technology based on ensemble of metamodels is employed to map the difference between the HF and LF models. Furthermore, an active learning strategy based on a generalised objective-oriented sequential sampling strategy is introduced to make full use of the already-acquired information of difference characteristics between the HF and LF models. Several numerical and engineering cases verify the applicability of the proposed VF metamodelling approach. Different types of test cases, sample sizes, and metamodel performance evaluation measures including accuracy and robustness are considered.