业务分离的一种神经网络方法

A Neural Network Method for Traffic Dispersion

在分组交换网络中，源目节点对（ｓｏｕｒｃｅｄｅｓｔｉｎａｔｉｏｎｐａｉｒ，下称ＳＤ对）之间的业务常常需要在多条路径上并行传输．本文研究如何将业务流有效地分布于一组路径上，以使网络具有最小的分组平均时延．由于这个问题在很多情况下需要实时处理，因此我们提出了一种神经网络方法来加以解决．在一个典型网络上的试验表明，我们的方法获得的结果优于前人用各种严格的数学分析方法得到的好的结果．

In packet switched communication networks,traffics between any source destination(SD) pair are expected to be transmitted over multiple paths.In this paper,we focus on the problem how to optimally distribute the load on a set of alternative paths so that the average time delay of a packet can be minimized.Since real time processing of this problem is needed in most cases,we propose a neural network method to solve it.Experiments on a typical communication network demonstrate that the result obtained using our method is better than the best result yielded by other rigorous mathematical algorithms.