Characterization of constrained continuous multiobjective optimization problems: A feature space perspective

Despite the increasing interest in constrained multiobjective optimization recent years, problems (CMOPs) are still insufficiently understood and characterized. For this reason, selection of appropriate CMOPs for benchmarking is difficult lacks a formal background. We address issue by extending landscape analysis to optimization. By employing four exploratory techniques, we propose 29 features (of which 19 novel) characterize CMOPs. These then used compare eight frequently artificial test suites against recently proposed suite consisting real-world based on physical models. The experimental results reveal that fail adequately represent some realistic characteristics, such as strong negative correlation between objectives overall constraint violation. Moreover, our findings show all studied have advantages limitations, no “perfect” exists. Additionally, effectiveness at predicting algorithm performance demonstrated two algorithms. Benchmark designers can use obtained select or generate CMOP instances characteristics they want explore.