精英学习的多维动态自适应人工鱼群算法

Elite Learning-based Multi-dimensional Dynamic Adaptive Artificial Fish Swarm Algorithm

针对人工鱼群算法存在易陷入局部最优、鲁棒性差以及寻优精度低的问题,提出了精英学习的多维动态自适应人工鱼群算法.传统人工鱼群用欧式距离度量视野、步长,无法体现不同维度上鱼群的搜索进度.提出的算法为每个维度设定独立的视野和步长,从而定义了视野向量、步长矩阵及多维邻域,以此改进了鱼群的4种基本行为,使人工鱼个体能够根据鱼群分布情况自适应调整寻优范围.为了增加鱼群的全局性,降低人工鱼陷入局部最优的可能性,提出了一种人工鱼精英学习策略.仿真实验结果表明,该算法能有效地提高人工鱼群的寻优精度、寻优质量及鲁棒性,且提高了人工鱼群的全局搜索能力.

The Artificial Fish Swarm Algorithm has some disadvantages such as falling into local optimum, poor robustness and low search accuracy. To solve these problems, this paper proposed an elite learning-based Multi-dimensional dynamic adaptive artificial fish swarm algorithm. The Artificial Fish Swarm Algorithm uses Euclidean distance to measure visual and step that can＇t reflect searching progress in different dimensions. The improved algorithm set independent visual and step for each dimension, and defined visual vec- tor, step matrix and multi-dimensional neighborhood, then improved 4 basic behaviors. Thereby, the artificial fish can adjust their own searching range adaptively according to their distribution. Otherwise, in order to improve fish＇s global search performance and decrease the probability of falling into local optimum, this paper proposed an elite learning strategy. The simulation results show that improved artificial fish swarm algorithm has good searching quality, better accuracy and robustness. Meanwhile, the algorithm improves artificial fish＇s global searching ability compared with other AFSAs.