Choose Appropriate Subproblems for Collaborative Modeling in Expensive Multiobjective Optimization.

Area of Science:

• Computational Mathematics
• Operations Research

Background:

• Expensive multiobjective optimization problems are often decomposed into single-objective subproblems.
• Current methods can be inefficient by optimizing unnecessary or already-solved subproblems.

Purpose of the Study:

• To propose an adaptive subproblem selection (ASS) strategy for identifying promising subproblems.
• To introduce a novel acquisition function, adaptive lower confidence bound (ALCB), to address exploration-exploitation imbalances.

Main Methods:

• Developed an adaptive subproblem selection (ASS) strategy.
• Employed a collaborative multioutput Gaussian process (CoMOGP) surrogate model for joint subproblem modeling.
• Designed a new acquisition function, adaptive lower confidence bound (ALCB).

Main Results:

• The proposed algorithm demonstrates competitive performance on benchmark problems.
• Effectiveness of ASS strategy, CoMOGP model, and ALCB acquisition function was quantitatively validated.
• The new approach improves efficiency in expensive multiobjective optimization.

Conclusions:

• The proposed ASS strategy, CoMOGP model, and ALCB acquisition function offer a more efficient approach to expensive multiobjective optimization.
• The developed methods effectively address inefficiencies and balance exploration-exploitation in optimization processes.