Parallelized multiobjective efficient global optimization algorithm and its applications

Guangyong Sun

doi:10.1007/s00158-019-02417-1

Return

Parallelized multiobjective efficient global optimization algorithm and its applications

2020-01-03

PRE

孙

孙光永 (Guangyong Sun) *

DOI：10.1007/s00158-019-02417-1

Original

Original request for help

Save

Abstract

En 中文

In engineering practice, most optimization problems have multiple objectives, which are usually in a form of expensive black-box functions. The multiobjective efficient global optimization (MOEGO) algorithms have been proposed recently to sequentially sample the design space, aiming to seek for optima with a minimum number of sampling points. With the advance in computing resources, it is wise to make optimization parallelizable to shorten the total design cycle further. In this study, two different parallelized multiobjective efficient global optimization algorithms were proposed on the basis of the Kriging modeling technique. With use of the multiobjective expectation improvement, the proposed algorithm is able to balance local exploitation and global exploration. To implement parallel computing, the Kriging Believer and multiple good local optima strategies were adopted here to develop new sample infill criteria for multiobjective optimization problems. The proposed algorithms were applied to five mathematical benchmark examples first, which demonstrated faster convergence and better accuracy with more uniform distribution of Pareto points, in comparison with the two other conventional algorithms. The best performed Kriging Believer strategy approach was then applied to two more sophisticated real-life engineering case studies on the tailor-rolled blank (TRB) structures for crashworthiness design. After optimization, the TRB hat-shaped tube achieved a 3% increase in energy absorption and a 10.7% reduction in mass, and the TRB B-pillar attained a 10.1% reduction in mass and a 12.8% decrease in intrusion, simultaneously. These benchmark and engineering examples demonstrated that the proposed methods are fairly promising for being an effective tool for a range of design problems.

Keywords:

Parallel computing

Efficient global optimization (EGO)

Multiobjective optimization

Kriging Believer

Multiple good local optima

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.