Journal of Transportation Systems Engineering and Information Technology ›› 2017, Vol. 17 ›› Issue (5): 68-74.

Previous Articles     Next Articles

Short-term Traffic Flow Prediction Based on CNN-SVR Hybrid Deep Learning Model

LUO Wen-hui, DONG Bao-tian,WANG Ze-sheng   

  1. School of Traffic and Transportation, Beijing Jiaotong University, Beijing 100044, China)
  • Received:2017-06-22 Revised:2017-08-09 Online:2017-10-25 Published:2017-10-30

基于CNN-SVR 混合深度学习模型的短时交通流预测

罗文慧,董宝田*,王泽胜   

  1. 北京交通大学交通运输学院,北京100044
  • 作者简介:罗文慧(1983- ),女,新疆乌鲁木齐人,博士生.

Abstract:

It is very important for intelligent transportation development to realize accurate and fast traffic forecast. However, dominant models for short- term traffic flow forecasting can't extract spatial- temporal characteristics of traffic flow data amply. Moreover, these models are susceptible to outside factors. To resolve these problems, an innovative model based on deep learning is proposed in this paper. Convolutional Neural Network(CNN) and Support Vector Regression(SVR) classifier are combined in this model: feature learning of traffic flow is carried out by using CNN in underlying network, then the extracted results are transmitted to SVR model as input to predict traffic flow. To verify the validity of the proposed model, experiments are conducted on actual traffic flow data of China national highway 103(G103). Experimental results show that the proposed model has higher prediction accuracy than the traditional prediction model, and the prediction performance is improved by 11%, which is an effective traffic flow forecasting model.

Key words: intelligent transportation, traffic flow prediction, convolutional neural network, traffic flow, support vector regression, deep learning

摘要:

精准且快速的短时交通流预测是智能交通发展的重要组成部分.本文针对当前交通流预测模型不能充分提取交通流数据的时空特征、预测性能容易受到外界干扰因素影响的问题,提出一种基于深度学习的短时交通流预测模型,该模型结合卷积神经网络(Convolutional Neural Network, CNN) 与支持向量回归分类器(Support Vector Regression, SVR)的特点:在网络底层应用CNN进行交通流特征提取,并将提取结果输入到SVR回归模型中进行流量预测.为验证模型的有效性,取G103 国道的实际交通流量数据进行试验.结果表明,提出的预测模型与传统的预测模型相比具有更高的预测精度,预测性能提高了11%,是一种有效的交通流预测模型.

关键词: 智能交通, 交通流预测, 卷积神经网络, 交通流, 支持向量回归, 深度学习

CLC Number: