一种基于可变多簇结构的动态概率粒子群优化算法

针对传统粒子群优化算法中全连接型拓扑和环形拓扑的特点,引入了一种粒子群信息共享方式——多簇结构,进而基于多簇结构提出了动态可变拓扑策略以协调动态概率粒子群优化算法的勘探和开采能力,并从理论上分析了最优信息在各种拓扑中的传播,同时从图论角度分析了几种经典拓扑以及动态可变多簇结构的统计特性.通过典型的Benchmark函数优化问题测试并比较了几种经典拓扑以及可变拓扑在高斯动态粒子群优化算法中的性能.实验结果表明,基于多簇结构的可变拓扑策略在求解复杂优化问题时优势明显,可以有效地避免算法陷入局部最优,在保证收敛速度的同时增强了算法的全局搜索能力.

Dynamic Probabilistic Particle Swarm Optimization Based on Varying Multi-Cluster Structure

Regarding to the characteristic of Gbest model and Lbest model in original particle swarm optimization, a neighborhood topology structure is developed, called multi-cluster structure. Moreover, a varying neighborhood strategy based on multi-cluster is proposed to coordinate exploration and exploitation. Furthermore, the information dissemination of several topologies is analyzed theoretically, and the statistical properties of canonical topologies and varying neighborhood topology are analyzed from graph theory. Gaussian dynamic particle swarm with several canonical topologies and varying topology are tested on five benchmark functions which are commonly used in the evolutionary computation. Experimental simulation results demonstrate that dynamic probabilistic particle swarm optimization with the varying neighborhood topology can solve complex optimization problems and escape from local optimal solutions efficiently. The results also reveal that the proposed method enhances the global search ability obviously.