Continuous Optimization through the Transition from Metaheuristic Search to Collaborative Orchestration

Abstract

Continuous optimization research is moving beyond standalone metaheuristic design toward landscape analysis, algorithm selection, cooperative search, and adaptive control. This paper synthesizes these developments through a mechanism-centered taxonomy with four layers: problem characterization, algorithm intelligence, metaheuristic search strategies, and high-level coordination. The analysis shows that existing mechanisms provide important capabilities but remain weakly integrated: landscape descriptors have limited runtime awareness, algorithm selection focuses on recommendation rather than interaction, cooperative metaheuristics rely on predefined structures, and hyper-heuristics mostly regulate operators within single frameworks. Based on these gaps, the paper positions collaborative orchestration as an integration direction for connecting instance-aware modeling, solver interaction, contribution evaluation, shared memory, and dynamic role assignment in continuous optimization.