A fitness approximation and on-line variable-fidelity metamodel based multi-objective genetic algorithm

Population-based algorithms can become computationally intractable when applying in practical engineering design optimization involving computational expensive simulation. To address this challenge, this paper proposes an on-line variable-fidelity metamodel (VFM) assisted Multi-Objective Genetic Algorithms (OLVFM-MOGA) approach. In OLVFM-MOGA, the VFM integrates information from low-fidelity (LF) and high-fidelity (HF) models is constructed to replace the simulation model during the optimization process to ease the computational burden. Besides, a novel model updating strategy is developed to address the issues of 1) which individuals will be sent for running simulations. 2) whether the LF model or the HF model should be selected to run for a selected individual. The effectiveness and merits of the proposed OLVFM-MOGA approach are demonstrated on the design optimization problem of a torque arm.