DOI: 10.11908/j.issn.0253-374x.20242

作者:

作者单位: 1.同济大学 道路与交通工程教育部重点实验室，上海 201804;2.连云港杰瑞电子有限公司 智能交通事业部，江苏 连云港 222061


作者简介: 唐克双（1980—），男，教授，工学博士，主要研究方向为智能交通系统、信号控制、驾驶行为。 E-mail: tang@tongji.edu.cn


通讯作者: 张锋鑫（1982—），男，高级工程师，工程硕士，主要研究方向为智能交通系统、软件系统架构、大数据。 E-mail：zhangfengxin3@163.com

中图分类号: U491.14


基金项目: 国家自然科学基金（61673302）




Short-Term Travel Speed Prediction for Urban Expressways Based on Convolutional Neural Network with Inception Module
Author:

Affiliation: 1.Key Laboratory of Road and Traffic Engineering of the Ministry of Education， Tongji University， Shanghai 201804， China;2.Intelligent Transportation Department， Lianyungang Jari Electronics Co.， Ltd.， Lianyungang 222061， Jiangsu， China


Fund Project:




摘要
| 图/表
| 访问统计
| 参考文献
|相似文献
| 引证文献
| 资源附件

摘要:为了高效捕捉城市快速路复杂的交通拥堵特征，提升短时行程速度预测的准确性，以卷积神经网络为基础，结合Inception模块，构建行程速度短时预测模型。将行程速度信息按照时空关联关系组织为二维数据矩阵，以图像为特征学习对象，自动提取交通数据高维特征并学习多粒度复杂交通拥堵模式，通过系统的网络设计与测试训练得到模型最优结构参数和优化参数，结合回归分析方法与梯度幅度相似性偏差指标，综合评价模型性能。实证结果表明，模型提取行程速度数据时序特征和时空演化特征能力较强，预测准确性较高，可进一步应用于其他交通参数的短时预测。



Abstract:In order to effectively learn the mixed traffic congestion patterns from the urban expressways and improve the accuracy of short-term travelling speed prediction， based on the convolution neural network， and incorporated with the Inception Module， a short-term travelling speed prediction model was established. The travelling speed information was arranged into two-dimensional matrices which could represent the traffic states， and the features represented by the input time-space travel speed images were learnt. The optimum model was obtained as the result of a systematic neural network design and training process， with the ability to automatically recognize multi-scale mixed traffic congestion patterns and extract high-dimensional features of the traffic data. Besides the regression analysis method as well as the gradient magnitude similarity deviation indicator was introduced to conduct a comprehensive evaluation. The case study shows that the proposed model outperforms other models in learning the temporal/spatiotemporal features from traffic data with a high prediction accuracy， which can be further applied to making short-term prediction for other traffic parameters.





PDF全文下载地址:

点我下载PDF