SRCA-CEC: an adaptive two-phase framework for real-world constrained optimization

Abstract This paper introduces SRCA-CEC, a hybrid two-phase framework based on the Structured Random Cycle-guided Algorithm (SRCA), designed to tackle complex real-world constrained optimization problems. Addressing the limitations of reactive constraint-handling methods in disjoint feasible regions, the proposed framework implements a hybrid search strategy. Phase 1 employs a lightweight Differential Evolution to rapidly identify feasible manifolds, while Phase 2 balances directionally-guided exploration and deterministic exploitation via the SRCA engine and an adaptive Constraint Dead-zone Method (CDM). Furthermore, a Lamarckian repair operator based on Sequential Least Squares Programming (SLSQP), rigorously managed through a memoized oracle, is integrated to refine solutions while strictly adhering to the computational budget. Extensive statistical evaluations on the 57 real-world problems of the IEEE CEC2020 benchmark demonstrate that SRCA-CEC achieves a highly competitive trade-off between precision and robustness. Comparisons with state-of-the-art algorithms, including EnMODE, SASS, COLSHADE, and the more recent SDDS-SABC, alongside performance profile analyses, confirm the reliability of the proposed approach, which achieves a 100% feasibility rate across the entire 57-problem benchmark suite. Finally, complexity metrics indicate a contained computational overhead, validating the framework’s suitability for demanding engineering design applications.