Abstract: A Particle Swarm Optimization based on Self-adaptive Multi-Swarm（PSO-SMS） algorithm is proposed to balance the exploration ability and development ability of the algorithm and improve its comprehensive performance on different problems. It consists of three modules, including the recombination, adjustment of sub-swarm size and detection. In the initial stage of evolution, the entire swarm is divided into many sub-swarms. The recombination module enables the different sub-swarms to share advantageous information, which is beneficial to the optimization of uni-modal and multi-modal functions. When the swarm is trapped in a potential local optimum, the detection module can help the swarm jump out of the current local optimum based on certain historical information from the search process. Through the adjustment to sub-swarm size, the size of each sub-swarm gradually increases during evolution, which will facilitate the improvement of exploration ability in the initial stage and the later development ability of the algorithm. The comparison between CEC2013 test suite and other seven PSO algorithms shows that the PSO-SMS algorithm has outstanding performance in solving the optimization problems of different functions.

Key words: particle swarm optimization, global optimization, self-adaptive, multi-swarm

摘要： 为了平衡算法的探测能力和开采能力，提高粒子群算法在不同类型问题上的综合性能，提出了一种基于自适应多种群的粒子群优化算法（PSO-SMS）。算法包含重组、子群规模调整和探测三个模块。在演化初始阶段，整个种群被划分成许多子种群。重组模块使不同子群间可以共享优势信息，有利于单峰和多峰函数的优化。当种群陷入潜在的局部最优时，探测模块可基于搜索过程的一些历史信息，帮助跳出当前的局部最优。通过子群规模调整，每个子种群的大小随着进化的过程而逐渐增加，有利于提高算法在初始阶段的探测能力和后期的开采能力。通过CEC2013的测试集与其他七种PSO算法的比较表明，PSO-SMS算法在解决不同类型的函数优化问题上有着突出的性能表现。

关键词: 粒子群算法, 全局优化, 自适应, 多种群