Abstract: As Particle Swarm Optimization (PSO) may easily get trapped in a local optimum, an improved PSO based on adaptive dynamic neighborhood and comprehensive learning named ADPSO was proposed. In ADPSO, the neighbors of each particle were dynamically constructed in terms of the best performing particle among the current particle neighborhood. Then the learning mechanism of each particle was separated into three parts: its own historical best position, the best neighbor and the global best one. At the position of the new particle, a random position around itself was added to increase the probability for the particle to move to that promising region. The test results on benchmark functions show that ADPSO achieves better solutions than other improved PSO, and it is an effective algorithm to solve multi-objective problems.

Key words: adaptive, Particle Swarm Optimization (PSO), dynamic neighbor, comprehensive learning, multimodal function

摘要： 为了克服粒子群算法在求解多峰函数时极易陷入局部最优解的缺陷, 提出一种基于自适应动态邻居广义学习的改进粒子群算法(ADPSO)。在ADPSO算法中, 根据每个粒子邻居中最好运行粒子的状态动态地调整邻居拓扑结构；每个粒子的学习样本包括全局最优粒子、自身最优粒子和粒子邻居中最优运行粒子；并且在新产生的粒子位置上, 加上一个随机位置以增加粒子向全局最优解移动的概率。在基准函数的测试中, 结果显示ADPSO算法比其他PSO算法有更好的运行效果，是求解多峰问题的一种有效算法。

关键词: 自适应, 粒子群算法, 动态邻居, 广义学习, 多峰函数