将Kohonen网络用于估计O－D矩阵的算法研究

Ｏ－Ｄ矩阵通常被看成常数．本文将它作为随机矩阵来研究．在Ｏ－Ｄ矩阵有几种可能取值的假设下，用神经网络中的Ｋｏｈｏｎｅｎ网络及相应的竞争学习算法，估计了Ｏ－Ｄ矩阵的值，并用模拟数据作了验证．结果表明，该算法不仅可以估计Ｏ－Ｄ分布的数值．还可以将交通流量按Ｏ－Ｄ分布的不同来分类，从而为交通控制和交通规划的研究提供了有效的途径．

The Algorithm to Estimate O-D Matrix with Kohonen Network

In transportation research O-D matrix is usually considered to be constant. but it should be a random one in fact. In this paper we try to study its probability distribution under some simple conditions. Suppose that an O-D matrix distributes at several points, we use Kohonen network and competitive learning rule to estimate the mutiple values it may take.Result from simulated data shows that we can not only obtain O-D matrix, but also classify input data of traffic flow into groups, each group represents a kind of O-D pattern. This new method will provide an effective way for traffic control and transportation planning.