Abstract: To solve the problems of inferior local search ability, slow convergence and easily trapping into the local optimization existing in the artificial bee colony algorithm, this paper proposes a gradient search strategy based on population by introducing gradient information and using central difference schemes for gradient approximation processing. The novel search strategy used by employed bees improves the local search ability while the scout bees still employ global random searching strategy to avoid premature phenomenon led by fast convergence in solving multimodal problems. Simulation results in six standard test functions show that the proposed searching mechanism gives a good balance between local solution and global exploration and improves the optimization ability of different kinds of optimization problems.

Key words: artificial bee colony algorithm, approximate gradient, local search, cooperation and sharing

摘要： 针对人工蜂群算法自身存在的局部搜索能力较差，收敛较慢，易受到局部最优束缚的问题，在种群搜索过程中引入梯度信息，并利用中心差分格式对梯度做近似处理，提出了一种基于种群的梯度搜索策略，并用于人工蜂群算法采蜜蜂阶段的搜索，提高算法的局部搜索能力。同时，侦察蜂采用了全局随机搜索策略，以避免在解决多峰问题时，由于快速收敛而导致的早熟现象。在6个标准测试函数上的仿真实验结果表明，这种新的搜索机制在局部求解与全局探索之间取得了较好的平衡，使得改进后的算法在不同类型问题上的优化能力有了明显改善。

关键词: 人工蜂群算法, 近似梯度, 局部搜索, 合作与共享