求解TSP的Hopfield网络权值修正算法分析

针对Ｈｏｐｆｉｅｌｄ神经网络在求解巡回推销员问题（ＴＳＰ）时出现的无效解和局部极小值问题，本文结合网络动态分析，利用权矩阵特征值在网络内部动力特性与外在解表现形式之间的关联作用，通过各特征值间关系协调，对基于原模型的权值修正方案作进一步理论分析与改进，探讨网络收敛于全局有效解的途径，并在全国３１省、市、自治区ＴＳＰ模拟中收到较好效果。

Analysis of New Weight Strategies in Solving TSP by Hopfield Neural Network

In this paper, we study several models of Hopfield neural network to overcome the difficulties of the invalid solution and local minimum in solving the Travelling Saleman Problem(TSP). Using the analytic methods of neural dynamical property, we discover that the set of the eigenvalues of the network connected matrix is the link between the inner dynamic performance and the outside solution form. The modified strategies of the weight matrix was developed theoretically by coordinating the relationship of eigenvalues. These new models have achieved very satisfactory performance in numerical simulation with 31 Chinese cities.