基于时空Transformer网络的隧道交通运行风险动态辨识方法

doi:10.13229/j.cnki.jdxbgxb.20230689

摘要/Abstract

摘要：

为了及时发现、评估与应对高速公路隧道交通风险隐患，确保隧道运行安全通畅，本文构建了基于时空Transformer网络的隧道运行风险状态动态辨识方法。以隧道交通流全域检测数据与关键断面集计数据为输入，通过空间CNN卷积与时序LSTM对隧道车流不同运行状态的时空分布特征进行无监督提取；利用大样本训练Transformer网络层参数，以捕获隧道交通运行状态在高维风险特征空间的分布与差异，实现隧道交通状态的风险划分与评估。采用真实隧道交通检测数据验证了本文方法的有效性，对隧道运行风险评估精度约为96%。

关键词: 交通运输工程, 隧道交通安全, 动态评估, 深度学习, 时序卷积网络

Abstract:

To promptly detect， evaluate， and address potential traffic risks in highway tunnels， ensuring the safe and efficient operation of tunnels， a dynamic estimation method for tunnel operational risk states was proposed based on spatial-temporal Transformer network. Tunnel traffic flow holographic detection and key cross-section aggregation information as inputs was utilized， the spatial convolution and temporal LSTM was employed by proposed model for unsupervised extraction of spatiotemporal distribution features of different tunnel traffic operational states. Through extensive sample training of Transformer network layer parameters，it aims to capture the distribution and variances of tunnel traffic states in a high-dimensional risk feature space. This facilitates the estimation of operational risk of tunnel traffic flow. The effectiveness of the proposed method is verified by using real tunnel traffic detection data， and the accuracy of tunnel operation risk estimation is about 96%.

Key words: traffic transport engineering, tunnel traffic safety, dynamic evaluation, deep learning, time series convolutional network

中图分类号:

U458

李振江,万利,周世睿,陶楚青,魏巍. 基于时空Transformer网络的隧道交通运行风险动态辨识方法[J]. 吉林大学学报(工学版), 2025, 55(4): 1336-1345.

Zhen-jiang LI,Li WAN,Shi-rui ZHOU,Chu-qing TAO,Wei WEI. Dynamic estimation of operational risk of tunnel traffic flow based on spatial-temporal Transformer network[J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(4): 1336-1345.

图/表 12

图1

图2

图3

图4

图5

表1

深度时序卷积网络其他相关参数设置"

相关参数	设定值
激活函数 $σ$	Relu
网络学习率	0.001
最大训练次数 $T t r a i n$	30 000
最小批训练尺寸	10
式（13）中分类数	3

表1

表2

图6

表3

图7

表4

表5

参考文献 26

[1]	周昱. 基于贝叶斯网的高速公路隧道交通事故预测及应急预案研究[D]. 西安:长安大学公路学院, 2018.
	Zhou Yu. Research on traffic accident prediction and contingency plan of expressway tunnel based on Bayesian network[D]. Xi'an: Highway School, Chang'an University, 2018.
[2]	牛文静. 高速公路隧道(群)交通流特征研究[D]. 西安: 长安大学公路学院, 2012.
	Niu Wen-jing. Research on the traffic flow characteristics of the freeway tunnel (group) [D]. Xi'an: Highway School, Chang'an University, 2012.
[3]	Tadaki S, Nishinari K, Kikuchi M, et al. Observation of congested two-lane traffic caused by a tunnel[J]. Journal of the Physical Society of Japan, 2002, 71(9): 2326-2334.
[4]	周林英, 韩静, 赵忠杰. 事件状态下高速公路隧道群的交通流特性分析[J]. 中国科技论文, 2017, 12(7): 839-844.
	Zhou Lin-ying, Han Jing, Zhao Zhong-jie. The traffic flow characteristics of highway tunnel groups based on cellular automata under the traffic incidents[J]. China Sciencepaper,2017, 12(7): 839-844.
[5]	倪娜. 山区高速公路隧道密集段交通特性及安全保障技术研究[D]. 西安: 长安大学公路学院, 2017.
	Ni Na. Research on traffic characteristics and security assurance technologies of tunnel intensive section in mountainous expressway[D]. Xi'an: Highway School, Chang'an University, 2017.
[6]	Tian L L, Jiang J C, Tian L. Safety analysis of traffic flow characteristics of highway tunnel based on artificial intelligence flow net algorithm[J]. Cluster Computing-The Journal of Networks Software Tools and Applications,2019, 22: 573-582.
[7]	李舒涵. 高速公路隧道交通状态判别研究[D]. 西安:长安大学公路学院, 2020.
	Li Shu-han. Research on traffic state identification of highway tunnel[D]. Xi'an: Highway School, Chang'an University, 2020.
[8]	张俊儒, 燕波, 龚彦峰. 隧道工程智能监测及信息管理系统的研究现状与展望[J]. 地下空间与工程学报, 2021, 17(2): 567-579.
	Zhang Jun-ru, Yan Bo, Gong Yan-feng. Research status and prospects of intelligent monitoring technology and information management system for tunnel engineering[J]. Chinese Journal of Underground Space and Engineering, 2021, 17(2): 567-579.
[9]	Lu Y, Liu S, Tian F, et al. Studies and prospects of technologies for monitoring highway tunnel fire[J]. IOP Conference Series Earth and Environmental Science, 2021, 634(1): No.012099.
[10]	Zhao Y, Zhang S, Ma C. Application of video detection technology in tunnel traffic safety monitoring system[J]. IPPTA: Quarterly Journal of Indian Pulp and Paper Technical Association, 2018, 30(8): 376-381.
[11]	Li S, Zhou Q, Wang K. Video-based tunnel luminance detection[J]. Automation in Construction, 2019, 107(11): No.102900.
[12]	贾磊, 李清勇, 俞浩敏. 基于监控视频的隧道交通冲突预测方法[J]. 北京交通大学学报, 2023, 47(3):61-69.
	Jia Lei, Li Qing-yong, Yu Hao-min. Method for predicting traffic conflicts in tunnels based on monitoring videos[J]. Journal of Beijing Jiaotong University, 2023, 47(3): 61-69.
[13]	Chen Z, Wen H. Modeling a car-following model with comprehensive safety field in freeway tunnels[J]. Journal of Transportation Engineering Part A-Systems, 2022, 148(7): No.04022040.
[14]	陈丰, 姜茗馨, 朱明, 等. 基于碰撞时间的大流量隧道口追尾隐患概率模型构建[J]. 北京交通大学学报, 2020, 44(6): 90-96.
	Chen Feng, Jiang Ming-xin, Zhu Ming, et al. A rear-end collision probability model at heavy-traffic tunnel entrance based on time-to-collision[J], Journal of Beijing Jiaotong University, 2020, 44(6): 90-96.
[15]	陈娟娟. 公路隧道交通安全状态特征选择与评估方法研究[D]. 福州:福州大学土木工程学院, 2014.
	Chen Juan-juan. Research on traffic safety status feature selection and evaluation methods of highway tunnel[D]. Fuzhou: School of Civil Engineering, Fuzhou University, 2014.
[16]	王玉婷. 特长隧道交通事故车辆排队长度演变分析与建模[D].重庆: 重庆交通大学交通运输学院, 2022.
	Wang Yu-ting. Analysis and modeling of vehicle queuing length evolution by traffic accident of extra-long tunnel [D].Chongqing: College of Traffic and Transportation, Chongqing Jiaotong University, 2022.
[17]	郑凯. 基于多源数据的城市桥隧结合段交通运行风险预测方法研究[D]. 重庆: 重庆交通大学交通运输学院, 2019.
	Zheng Kai. Research on traffic operation risk prediction method for urban bridge tunnel combined sections based on multisource data[D]. Chongqing: College of Traffic and Transportation, Chongqing Jiaotong University, 2019.
[18]	Tympakianaki A, Koutsopoulos H N, Jenelius E. Anatomy of tunnel congestion: causes and implications for tunnel traffic management[J]. Tunnelling and Underground Space Technology, 2019, 83:498-508.
[19]	Gu J X, Wang Z H, Kuen J, et al. Recent advances in convolutional neural networks[J]. Pattern Recognition, 2018, 77, 354-377.
[20]	Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
[21]	Tay Y, Dehghani M, Bahri D, et al. Efficient transformers: a survey[J]. ACM Computing Surveys, 2022, 55(6): 1-28.
[22]	李海涛. 基于深度学习的交通流运行风险评估方法研究[D]. 长春: 吉林大学交通学院, 2021.
	Li Hai-tao. Research on traffic flow operation risk evaluation based on deep learning[D]. Changchun: College of Transportation, Jilin University, 2021.
[23]	李晓璐, 于昕明, 杜崇. 基于权值优化的FCM-MSVM算法及其在高速公路状态判别中的应用[J]. 北京交通大学学报, 2018, 42(4): 72-78, 84.
	Li Xiao-lu, Yu Xin-ming, Du Chong. FCM-MSVM algorithm based on weight optimization and its application in state identification of freeway[J]. Journal of Beijing Jiaotong University, 2018, 42(4): 72-78, 84.
[24]	张博. 基于机器学习的交通状态短时预测方法研究[D]. 长春:吉林大学软件学院, 2018.
	Zhang Bo. Research on short-term prediction of traffic state based on machine learning[D]. Changchun:College of Software, Jilin University, 2018.
[25]	Liu Q, Lu J, Zhao K, et al. Multiple Naïve bayes classifiers ensemble for traffic incident detection[J]. Mathematical Problems in Engineering, 2014(1):No.383671.
[26]	刘擎超, 陆建, 陈淑燕. 基于随机森林的交通事件检测方法设计与分析[J]. 东南大学学报:英文版, 2014, 30(1): 88-95.
	Liu Qing-chao, Lu Jian, Chen Shu-yan. Design and analysis of traffic incident detection based on random forest[J]. Journal of Southeast University(English Edition), 2014, 30(1): 88-95.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

真实类别	评估结果
真实类别	正例	反例
正例	真正例TP	假反例FN
反例	假正例FP	真反例TN

模型	精确率	召回率	F₁分数	准确率/%
全域数据+时空Transformer	0.941 3	0.931 5	0.937 7	93.12
断面数据+时空Transformer	0.910 2	0.898 1	0.904 2	90.37
本文方法	0.960 3	0.963 2	0.964 1	96.18

模型	精确率	召回率	F₁分数	准确率/%
SVM	0.773 2	0.763 1	0.771 5	77.32
XGBoost	0.738 7	0.753 6	0.752 2	75.57
CNN+Transformer	0.879 8	0.858 1	0.868 8	87.02
LSTM+Transformer	0.895 2	0.905 3	0.892 4	90.24
本文方法（全域数据）	0.941 3	0.931 5	0.937 7	93.12

模型	精确率	召回率	F₁分数	准确率/%
贝叶斯分类	0.798 2	0.815 4	0.806 7	80.57
随机森林	0.803 7	0.821 4	0.815 4	81.37
XGBoost	0.849 8	0.868 1	0.858 8	86.02
LSTM+Softmax	0.836 1	0.849 8	0.833 7	84.24
本文方法（断面集计数据）	0.910 2	0.898 1	0.904 2	90.37