基于神经网络的路口交通流转向比预测  

ANN-Based Prediction of Turning Rate of Traffic Flows at Intersection

作　　者：李瑞敏[1] 陆化普[1] 史其信[1] 

机构地区：[1]清华大学交通研究所,北京100084

出　　处：《西南交通大学学报》2007年第6期743-747,共5页Journal of Southwest Jiaotong University

基　　金：科技部"十五"科技攻关项目(2002BA404A20B)

摘　　要：为了预测路口交通信号控制所需的转向交通流量，提出了基于改进BP（back-propagation）神经网络的路口交通流转向比预测模型，给出了相应参数的计算方法；采用自适应学习率和动量梯度下降法以提高神经网络的学习速度和算法的可靠性，并用调查数据对模型进行了检验．研究结果表明，与传统的平均值法相比，用所提出的模型，平均绝对相对误差减小约1％～3％．Based on an improved back-propagation neural network, a predication model for the turning rate of traffic flows at intersections was proposed to predict traffic flows for the signal control of intersections. The corresponding method to determine necessary parameters in this model was given. improve the learning rate and reliabihty of neural network algorithms, approach and the gradient descent with momentum method were adopted. carried out to prove the correctness of the proposed the self-adaptive learning rate In addition, a simulation was model. The research result shows that compared with the average value method, the proposed model can decrease the mean absolute relative error by 1% -3%.

关 键 词：交通流转向比 预测模型 神经网络 自适应学习率 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]