邻域结构为复杂网络的粒子群算法

PSO algorithm with spatial structure based on complex network

为避免早熟收敛和局部最优,设计了一种基于复杂网络进行个体交互的粒子群算法(CNS-PSO)。该算法在粒子与网络节点间建立映射关系,并根据节点的邻居集合,获得粒子的动态飞行邻居。每个飞行邻居集合是一个独立又彼此联系的进化小社会。在CNS-PSO中,每个粒子的位置更新策略不仅考虑了认知部分及社会的信息共享,还考虑了小社会内和小社会间的信息交流。在八个测试函数上,将CNS-PSO与标准PSO进行了对比,在不同维度的大多数函数上,前者的性能均优于后者,说明具有无标度网络邻域结构的CNS-PSO算法具有较强的避免早熟和逃逸局部最优的能力。

In order to avoid premature convergence and local optimum,this paper designed a novel method named CNS-PSO.The individuals in this algorithm interacted with each other based on a given complex network. CNS-PSO established the mapping between particles and network nodes,and obtained the particles＇ flying neighbors dynamically according to their nodes＇ neighbors on a given complex network. Each set of flight neighbor was an independent but contacted evolutionary small community. In the algorithm,the particle＇s update strategy for its position taken into account not only the cognitive part and the information sharing of the whole society,but also the information exchange within a small community or between two small ones. To validate CNS-PSO,this work conducted a series of experiments on eight test functions with different dimensions. On most of functions,CNS-PSO is better than standard PSO. The results show that the proposed algorithm with scale-free spatial structure CNS-PSO algorithm has stronger ability to avoid premature and escape from local optimal.