Global optimization of complex, high-dimensional landscapes remains a fundamental challenge in scientific and engineering domains. To mitigate the inherent limitations of premature convergence and diversity loss, this paper proposes CLGMESC, an enhanced variant of the Escape Algorithm (ESC). The proposed algorithm integrates a dimension-wise comprehensive learning (CL) strategy with a hybrid Cauchy-Gaussian mutation (HCGM) operator. The CL strategy reconfigures the learning paradigm for stagnant individuals, enabling them to construct exemplars from multiple high-quality peers and thereby restore population diversity. Synergistically, the HCGM operator utilizes an adaptive weighting mechanism to dynamically balance heavy-tailed Cauchy mutations for global exploration and thin-tailed Gaussian mutations for local refinement, effectively facilitating escapes from local optima. Comprehensive evaluations on the CEC2017 benchmark suite demonstrate that CLGMESC achieves the top rank among ten advanced metaheuristics (including SBO, BBO, PO, DE, PSO, SMA, CPA, and MGO), with Wilcoxon signed-rank tests confirming its statistical superiority (p < 0. 05) across the majority of test functions. Furthermore, the practical efficacy of CLGMESC was validated through a reservoir production optimization problem using the three-dimensional Egg Model (60 60 7 grid). In determining optimal well controls, CLGMESC achieved the highest Net Present Value (9. 824 10⁸ USD) with the lowest standard deviation, thus substantiating its reliability and robustness in solving computationally intensive real-world engineering problems. The consistently high rankings across diverse benchmarks and the substantial economic gains in the reservoir simulation underscore the algorithm’s pronounced capability to maintain a robust exploration-exploitation balance and dynamically escape local optima in demanding parameter spaces.
Chen et al. (Mon,) studied this question.