基于精英引导自适应差分协同小龙虾优化算法

Elite-Guided Adaptive Differential Cooperative Crayfish Optimization Algorithm

摘要: 针对传统小龙虾优化算法(Crayfish Optimization Algorithm, COA)存在易早熟收敛、全局勘探与局部开发能力失衡、寻优精度不足等固有缺陷，本文提出一种基于精英引导自适应差分协同小龙虾优化算法(Elite-Guided Adaptive Differential Cooperative Crayfish Optimization Algorithm, EGCOA)。该算法首先采用精英反向学习机制完成种群初始化，提升初始解在解空间的分布均匀性与种群多样性，为算法寻优奠定优质起点；其次引入融合差分进化的繁殖策略实现个体位置迭代更新，同步兼顾算法的全局搜索速率与求解精度；最后提出带自适应扰动的精英保留策略，有效增强算法的局部深度开发能力，规避迭代过程中的早熟收敛问题。基于CEC2020基准测试函数集的仿真对比实验结果表明，所提EGCOA在收敛速度、寻优精度与运行鲁棒性上均显著优于原始COA及其他主流群智能优化算法，有效解决了原始算法的固有缺陷，具备更优异的综合优化性能。

Abstract: The Crayfish Optimization Algorithm (COA) suffers from three inherent drawbacks. It tends to converge prematurely, faces an imbalance between global exploration and local exploitation, and exhibits unsatisfactory optimization accuracy. To solve these problems, this paper proposes a Elite-Guided Adaptive Differential Cooperative Crayfish Optimization Algorithm (EGCOA). First, the algorithm adopts an elite opposition-based learning mechanism for population initialization. This mechanism improves the uniform distribution of initial solutions in the solution space and enriches population diversity, laying a high-quality starting point for the optimization process. Second, a breeding strategy integrated with differential evolution (DE) is introduced to update individual positions iteratively. This strategy balances the global search speed and solution accuracy of the algorithm simultaneously. Finally, an elite retention strategy with adaptive perturbation is designed. It effectively enhances the algorithm’s local deep exploitation capability, and avoids premature convergence during the iteration process. We carried out simulation comparison experiments based on the CEC2020 benchmark test function suite. The results show that the proposed EGCOA significantly outperforms the original COA and other mainstream swarm intelligence optimization algorithms in convergence speed, optimization accuracy, and operational robustness. It effectively overcomes the inherent defects of the original algorithm, and presents more excellent comprehensive optimization performance.