引用本文:费春国, 韩正之, 唐厚君.基于连续Hopfield网络求解TSP的新方法[J].控制理论与应用,2006,23(6):907~912.[点击复制]
FEI Chun-guo, HAN Zheng-zhi, TANG Hou-jun.Novel method to solve TSP by continuous Hopfield neural network[J].Control Theory and Technology,2006,23(6):907~912.[点击复制]
基于连续Hopfield网络求解TSP的新方法
Novel method to solve TSP by continuous Hopfield neural network
摘要点击 3180  全文点击 1788  投稿时间：2005-04-27  修订日期：2005-12-05
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  
  2006,23(6):907-912
中文关键词  连续Hopfield网络  能量函数  组合优化  旅行商问题(TSP)  全局最优
英文关键词  continuous Hopfield neural network  energy function  combination optimization  traveling salesman problem(TSP)  global optimization
基金项目  
作者单位
费春国, 韩正之, 唐厚君 中国民航大学工程技术训练中心, 天津300300
上海交通大学自动化系, 上海200030
上海交通大学电气工程系, 上海200030 
中文摘要
      当连续Hopfield网络及其能量函数同时具有自反馈或不具有自反馈时, 称之为一致连续Hopfield网络. 在分析了一致连续Hopfield网络能量稳定性的基础上, 进一步研究了当网络有自反馈, 而其能量函数无自反馈的情况下,网络能量变化的性质, 分别给出了使能量函数上升、下降和不变的条件. 利用这一理论, 可以克服由于梯度下降法所导致的网络能量函数总是下降, 从而使网络陷入局部极小值或不可行解的现象. 最后在这个理论的基础上我们给出了一种新的求解TSP(traveling salesman problem)的方法, 仿真研究表明此方法对于求解TSP问题是很有效的.
英文摘要
      Whether both continuous Hopfield neural network (CHNN) and its energy function have self-feedback or not, it is called uniform CHNN. Firstly, convergence of uniform CHNN is analyzed. Secondly, the character of energy function variation are studied when the CHNN has self-feedback while its energy function has not self-feedback. Thirdly, conditions are proposed to ensure that the energy function can increase, decrease or not change respectively. This principle eradicates local minima or invalid solutions caused by consistent reduction of the energy function via the usual gradient descent method. Furthermore, a new approach to solve TSP (traveling salesman problem) is proposed according to this principle. Finally, simulations show the new approach can provide very good results when it is used to solve TSP.