Reference Vector Assisted Candidate Search with Aggregated Surrogate for Computationally Expensive Many Objective Optimization Problems

Abstract

Pareto-optimal sets of multiobjective optimization problems with black-box and computationally expensive objective functions are generally hard to locate within a limited computational budget, and this situation gets even worse when more than three objectives are involved. To this end, we present a novel surrogate-assisted many-objective optimization algorithm RECAS. Unlike most prior studies, the proposed algorithm is a non–evolutionary-based method, and it iteratively determines new points for expensive evaluation via a series of independent reference vector assisted candidate searches. Furthermore, to make the number of surrogates to be maintained independent of the number of objectives, in each candidate search, RECAS constructs a surrogate model in an aggregated manner to approximate the quality assessment indicator of each point rather than a certain objective function. Under some mild assumptions, this study proves that RECAS converges almost surely to the Pareto-optimal front. In the numerical experiments, the effectiveness and reliability of RECAS are examined on both DTLZ and WFG test suites with the number of objectives varying from 2 to 10. Compared with six state-of-the-art many-objective optimization algorithms, RECAS generally performs better in maintaining convergent and well-spread approximation of the Pareto-optimal front. Finally, the good performance of RECAS on two watershed simulation model calibration problems indicates its great potential in handling real-world applications. History: Accepted by Antonio Frangioni, Area Editor for Design and Analysis of Algorithms–Continuous. Funding: This work was supported by the National University of Singapore start-up grant to C.A. Shoemaker [Grant R-266-000-109-133]. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplementary Information [ https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2022.1260 ] or is available from the IJOC GitHub software repository ( https://github.com/INFORMSJoC ) at [ http://dx.doi.org/10.5281/zenodo.7243971 ].