Abstract: In order to deal with the basic ABC algorithm for its slow convergence, tending to get stagnation on local optima, and further to improve its searching efficiency in exploration and exploitation, this paper proposes an improved artificial bee colony algorithm called Enhanced Mutual Learning ABC algorithm（EMLABC）, applying different kind of honey bees with distinguished strategies, firstly for employed bees, by exemplifying mutation perturbation learning frequency and basing on multi comparatively prior neighbors for learning, to enhance global exploration and avoid premature, and then applying onlooker bees with extensive mutual learning strategy, which can enable the new candidate solutions more likely to search in potential better space, thus to achieve fast convergence and accuracy. The experiments are conducted on a benchmark suite of 16 unimodal and multimodal test functions, the results demonstrate significant improvements of EMLABC when compared with the basic ABC algorithm and several recent variants of ABC algorithm.

Key words: artificial bee colony, swarm intelligence, numerical function optimization, mutual learning

摘要： 为了解决基本蜂群算法存在的收敛速度慢、易陷入局部最优等问题，并提高算法在探索和开发方面的寻优性能，提出一种改进的蜂群算法，称为强化互学习的人工蜂群算法（EMLABC），针对不同种类蜜蜂分别采用不同的搜索策略，首先对于雇佣蜂通过采用提高交叉变动学习频率以及同时面向多个较优近邻学习的机制来增强算法的全局探索能力并且避免早熟；其次针对跟随蜂采用深化的互学习策略，使新生子代保持倾向于在潜在更优区域进行搜索，进而提高算法的收敛性能和精度。在16个标准测试集函数和基本蜂群算法以及最近几个变种进行对比测试，结果表明EMLABC在收敛速度、准确寻优能力和稳定性上都有显著的提升。

关键词: 人工蜂群算法, 群体智能, 数值函数优化, 互学习