S-CoEA: Subproblems Co-Solving Evolutionary Algorithm for Uncertain Optimization.

Area of Science:

• Optimization
• Computational Intelligence
• Operations Research

Background:

• Existing methods for uncertain optimization problems (UOPs) often require decision-maker preferences or probability distributions, which are difficult to obtain.
• Making assumptions about these factors can be risky when knowledge is limited.

Purpose of the Study:

• To propose a novel evolutionary algorithm (EA) for UOPs that treats them in an a posteriori manner.
• To develop an algorithm that does not rely on explicit decision-maker preferences or probability distributions.

Main Methods:

• A subproblem co-solving evolutionary algorithm (S-CoEA) is proposed, decomposing UOPs into correlated subproblems.
• An ordered weighted-sum (OWS) operator is used for decomposition, with subproblems co-solved in parallel using neighboring information.
• A sampling strategy gathers distribution information and mitigates uncertainty effects, enhanced by a sample-updating scheme.

Main Results:

• S-CoEA was compared against state-of-the-art competitors, including EA with exponential sampling (E-sampling) and population-controlled covariance matrix self-adaptation evolution strategy (pcCMSA-ES).
• Numerical experiments on continuous and discrete test instances demonstrated S-CoEA's superior or competitive performance across most cases.

Conclusions:

• S-CoEA offers an effective approach for handling UOPs, particularly when decision-maker preferences or probability distributions are unknown or difficult to ascertain.
• The algorithm's ability to decompose and co-solve subproblems, coupled with its sampling strategy, provides robust performance under uncertainty.