三态层叠PCNN原理及在最短路径求解中的应用  

Three-state cascading pulse coupled neural network and the application in finding shortest paths

作　　者：赵荣昌[1] 马义德[1] 绽琨[1] 

机构地区：[1]兰州大学信息科学与工程学院电路与系统研究所,甘肃兰州730000

出　　处：《系统工程与电子技术》2008年第9期1785-1789,共5页Systems Engineering and Electronics

基　　金：国家自然科学基金(6057201160872109);新世纪人才支持计划(NCET-06-0900)资助课题

摘　　要：为实现神经网络的流水线操作,将电路设计中的三态思想和层叠流水线思想运用到神经网络中,通过建立具有抑制、亚点火和点火三种状态的神经元,提出了三态层叠脉冲耦合神经网络模型。通过三态神经元,将点火过程分解成三阶段以便实现流水线操作,成功解决了神经网络在自动波传播方向上的并行处理问题,在自动波传播的横向和纵向都实现了并行处理,极大地提高了算法的运算速度和准确性。将此模型运用到最短路径的求解问题中,通过实验表明,该算法在保证全局搜索的同时提高了搜索速度,且其对初始条件和参数的依赖性很小。A new neural network-three-state cascading pulse coupled neural network （TCPCNN） is presented via taking the ideas of three-state and pipelining used in circuit designing into neural network, and creating new neurons which have three states： restraining, sub-firing and firing. The three-state fired process is applied and is divided into three stages in order to pipelining. The proposed algorithm successfully solves the parallel processing problem in the direction of auto-wave spreading in neural network and realizes parallel processing in both the spreading direction and the transverse direction. Consequently, the algorithm has higher computing speed and stronger correctness. And then, this idea is applied to the shortest path problem. Experiment results indicate that this algorithm can not only ensure full-scale searching, but also increase processing speed; further- more, it has little dependence on initial conditions and parameters.

关 键 词：组合优化 脉冲耦合神经网络 三态层叠流水线操作 自动波 最短路径 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]