基于分区个体排名的非线性种群缩减的人工蜂群算法.

Aiming at the problem that ABC has strong exploration but weak exploitation, which leads to slow convergence speed, this paper proposed an unlinear population size reduction strategy based on cluster individual rank (UPSR-CIR). Firstly, the strategy designed the long-tail unlinear population size reduction function which maintained a large population to explore fully in the early stage, and reduced the population size rapidly in the middle stage, so as to maintain a small population to strengthen exploitation in the late stage, while allocating relatively more computing resources for the late stage to accelerate convergence. Secondly, to ensure the diversity of the population, it used K-means clustering dynamically to divide the population into clusters every a certain number of generations, and carried out the population size reduction in the unit of cluster. At the same time, when the population size reducing in the unit of cluster, it determined the number of individuals deleted according to the rank of the best individual in the cluster, so as to reserve relatively more computing resources for the potential cluster with higher rank to further strengthen exploitation. This paper used 22 benchmark test functions to compare and analyze the UPSRCIR on ABC and its variants. The results show that the UPSR-CIR exhibits higher solution accuracy, stability and convergence speed. It is also universally applicable to ABC variants. Finally, this paper also used 12 classical TSP cases to validate the practicality and superiority of the UPSR-CIR strategy on real application problem. [ABSTRACT FROM AUTHOR]