用相关路段交通流逼近目标路段流量的神经网络方法

Approximation of the Objective Road-Section's Traffic Flow Volume from the Data of Correlative Road-Sections by ANN

为了研究城市道路动态变化的交通流量之间的关系,本文分析了城市道路交通流量分布的特点,提出了用相关路段的流量数据逼近目标路段流量的思想,给出了用于逼近的神经网络模型、具体算法和评价标准,在比较精确的训练样本基础上对网络进行了训练,测试了训练后的网络的性能,并与传统的多项式逼近方法进行了比较。实验表明,采用基于神经网络的非线性动态逼近方法较传统的多项式逼近方法能够更加有效的改善逼近的误差。

In this paper, an effective method of dynamic data-approximation of traffic flow in correlative road-sections by ANN is proposed to precisely research the relations of the varying traffic flow in city road-sections. The distribution of road traffic flow is studied , and accordingly a approximating model is built; the algorithm and the evaluating standard are also given. After training with quite accurate samples, the performance of ANN is tested, and is compared with traditional polynomial approximation. Experimental results have shown that the method of approximation by ANN suits for nonlinear dynamic system, and is able to significantly decrease the error than polynomial approximation.