高速公路动态交通流Elman神经网络模型  

Elman neural network model of freeway dynamic traffic flow

作　　者：梁新荣[1] 刘智勇[2] 毛宗源[1] 

机构地区：[1]华南理工大学自动化学院,广东广州510640 [2]五邑大学信息学院,广东江门529020

出　　处：《交通运输工程学报》2006年第3期92-96,共5页Journal of Traffic and Transportation Engineering

基　　金：广东省自然科学基金项目(06300326)

摘　　要：为了提高高速公路交通流建模的精度,分析了离散的高速公路动态交通流数学模型,基于Elman网络原理,建立了回归神经网络交通流模型。回归神经网络的输入层、上下文层、隐含层和输出层的节点数目分别选为8、30、30和2,采用Levenberg-Marquardt算法对回归神经网络进行训练,并对一条5路段的高速公路进行仿真。结果表明:回归神经网络平均相对误差为8.683 7×10-5,最大相对误差为4.237 1×10-4,与BP神经网络和RBF神经网络相比较,Elman回归神经网络能更好地逼近交通流数学模型,真实地描述交通流基本特性,能准确地建立动态交通流模型,适应交通状况的变化。In order to improve the accuracy of freeway traffic flow modeling, the discrete mathematical model of freeway dynamic traffic flow was analyzed, and a traffic flow model of recurrent neural network was built based on the principle of Elman network. The node numbers of the input layer, context layer, hidden layer and output layer of the recurrent network were selected as 8, 30, 30 and 2 respectively. Levenberg-Marquardt algorithm was used to train the recurrent network, and a freeway with five segments was simulated. Simulation result shows that the average relative error and the maximum relative error for the recurrent network are 8. 683 7 × 10^-5 and 4. 237 1 × 10^-4 respectively, compared with the BP and RBF neural network, the Elman recurrent network can approach the mathematical model of freeway traffic flow more accurately, can better describe the basic properties of traffic flow, and by means of on-line learning from the data measured by sensors on the freeway, the Elman recurrent network can adapt to the change of traffic status. 1 tab, 7 figs, 10 refs.

关 键 词：交通规划 动态交通流 回归神经网络 建模 比较 

分 类 号：U491.112[交通运输工程—交通运输规划与管理]