基于图时空模式学习网络的路网实时交通事件自动检测方法

doi:10.13229/j.cnki.jdxbgxb.20230875

摘要/Abstract

摘要：

为提升路网交通异常事件的检测精度并降低误报率，提出了一种基于图时空模式学习网络（GSTPL）的路网实时交通事件自动检测方法。将路网交通事件检测问题抽象为图结构异常检测任务；设计了交通时空融合图表达方法，筛选具有强时空依赖性与模式规律性的图节点信息作为网络输入；引入图时空卷积与图嵌入层对时空模式特征进行提取，构造多组件输入与融合预测结构对不同时间维的交通模式规律进行融合；设计了异常状态评估方法，通过对模型预测误差分布的学习，结合当前检测数据给出最终的异常事件判定结果。采用2个真实交通路网数据进行算法验证，实验结果表明，提出的GSTPL交通事件检测方法具有较高的检测精度、较低的误报率与更短的平均检测时间；在可接受误检率为5%与10%时，对异常交通事件的检测率可分别达到91%与96%以上。

关键词: 交通运输系统工程, 事件自动检测, 图神经网络, 模式学习, 异常评估

Abstract:

In order to improve the detection accuracy of road network traffic incidents and reduce the false alarm rate， a real-time automatic detection method of road network traffic incidents based on Graph Spatial-Temporal Pattern Learning Network （GSTPL） is proposed. Firstly， the traffic incident detection problem in the road network is abstracted into a graph structure anomaly detection task； a traffic spatial-temporal fusion graph representation method is designed to filter the road network graph node with strong spatial-temporal dependence and same pattern regularity as the input. Then， the graph spatial-temporal convolution and graph embedding layer are introduced to extract the spatial-temporal pattern features， and the multi-component input and fusion prediction structure are constructed to fuse traffic pattern rules in different time dimensions， and realize stable forecasts of graph node parameters. An abnormal state evaluation method is designed， and the final incident detection result is given by learning of the prediction error distribution and combining with the current detection data. Two real road networks datasets were used for validation experiments， and the proposed algorithm was compared with several typical traffic incident detection algorithms. The comparison results show that the proposed GSTPL has higher detection accuracy， lower false alarm rate and shorter average detection time. When the acceptable false positive rate is 5% and 10%， the detection rate of traffic incidents can reach more than 91% and 96% respectively.

Key words: engineering of communications and transportation system, automatic incident detection, graph neural network, pattern learning, abnormal evaluation

中图分类号:

U458

柴树山,周志强,李海涛,徐炅旸. 基于图时空模式学习网络的路网实时交通事件自动检测方法[J]. 吉林大学学报(工学版), 2025, 55(7): 2145-2161.

Shu-shan CHAI,Zhi-qiang ZHOU,Hai-tao LI,Jiong-yang XU. Real-time road network traffic anomaly incident detection based on graph spatial-temporal pattern learning network[J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(7): 2145-2161.

图/表 13

图1

图2

图3

图4

图5

图6

表1

GSTPL网络设置相关参数"

模型参数	设置值或参数类型
最大训练次数	20 000
多组件预测训练阶段学习率	0.001
组件权重联调阶段学习率	0.000 1
网络优化器	Adam
长范围滑动窗口 $L w$	3 000
短范围滑动窗口 $L s$	6
式（11）与式（12）中高斯核带宽 $h$	0.2
式（16）中调整系数 $γ$	0.8
异常状态可能性判定阈值 $ε$	0.3

表1

图7

图8

表2

表3

表4

表5

参考文献 32

[1]	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 detection based on random forest[J]. Journal of Southeast University (English Edition), 2014, 30(1): 88-95.
[2]	Evans J, Waterson B, Hamilton A. Evolution and future of urban road incident detection algorithms[J]. Journal of Transportation Engineering, Part A: Systems, 2020, 146(6): No.03120001.
[3]	Payne H J, Tignor S C. Freeway incident-detection algorithms based on decision trees with states[J]. Transportation Research Record, 1978, 682: 30-37.
[4]	Chakraborty P, Hegde C, Sharma A. Data-driven parallelizable traffic incident detection using spatio-temporally denoised robust thresholds[J]. Transportation Research Part C: Emerging Technologies, 2019, 105: 81-99.
[5]	程小洋. 交通事件检测算法的阈值自适应调整与优化[D]. 南京: 东南大学交通学院, 2021.
	Cheng Xiao-yang. Threshold adaptive adjustment and optimization of traffic incident detection algorithm[D]. Nanjing: School of Transportation, Southeast University, 2021.
[6]	Liu Q C, LU J, Chen S Y, et al. Multiple naive bayes classifiers ensemble for traffic incident detection[J]. Mathematical Problems in Engineering, 2014, 2014: No.383671.
[7]	Srinivasan D, Jin X, Cheu R L. Evaluation of adaptive neural network models for freeway incident detection[J]. IEEE Transactions on Intelligent Transportation Systems, 2004, 5(1): 1-11.
[8]	Liu G W, Jin H L, Li J Z, et al. A Bayesian deep learning method for freeway incident detection with uncertainty quantification[J]. Accident Analysis & Prevention, 2022, 176: No.106796.
[9]	商强, 林赐云, 杨兆升, 等. 基于变量选择和核极限学习机的交通事件检测[J]. 浙江大学学报: 工学版, 2017, 51(7): 1339-1346.
	Shang Qiang, Lin Ci-yun, Yang Zhao-sheng, et al. Traffic incident detection based on variable selection and kernel extreme learning machine[J]. Journal of Zhejiang University (Engineering Science), 2017, 51(7): 1339-1346.
[10]	Dogru N, Subasi A. Traffic accident detection using random forest classifier[C]∥The 15th Learning and Technology Conference: L&T, Jeddah, Saudi Arabia, 2018: 40-45.
[11]	Wang L L, Ngan H Y T, Yung N H C. Automatic incident classification for large-scale traffic data by adaptive boosting SVM[J]. Information Sciences, 2018, 467: 59-73.
[12]	Xiao J L. SVM and KNN ensemble learning for traffic incident detection[J]. Physica A: Statistical Mechanics and Its Applications, 2019, 517: 29-35.
[13]	Wang R, Work D B, Sowers R. Multiple model particle filter for traffic estimation and incident detection[J]. IEEE Transactions on Intelligent Transportation Systems, 2016, 17(12): 3461-3470.
[14]	尹春娥, 陈宽民, 万继志. 基于小波方程的高速公路交通事故自动检测方法[J]. 中国公路学报, 2014, 27(12): 106-112.
	Yin Chun-er, Chen Kuan-min, Wan Ji-zhi. Automatic detection method for expressway traffic accidents based on wavelet equation[J]. China Journal of Highway and Transport, 2014, 27(12): 106-112.
[15]	Adeli H, Samant A. An adaptive conjugate gradient neural network–wavelet model for traffic incident detection[J]. Computer‐Aided Civil and Infrastructure Engineering, 2000, 15(4): 251-260.
[16]	Lu J, Chen S Y, Wang W, et al. A hybrid model of partial least squares and neural network for traffic incident detection[J]. Expert Systems with Applications, 2012, 39(5): 4775-4784.
[17]	Tang S M, Gao H J. Traffic-incident detection-algorithm based on nonparametric regression[J]. IEEE Transactions on Intelligent Transportation Systems, 2005, 6(1): 38-42.
[18]	Jiang W W, Luo J Y. Graph neural network for traffic forecasting: A survey[J]. Expert Systems with Applications, 2022, 207: 117921.
[19]	冯宁, 郭晟楠, 宋超, 等. 面向交通流量预测的多组件时空图卷积网络[J]. 软件学报, 2019, 30(3): 759-769.
	Feng Ning, Guo Sheng-nan, Song Chao, et al. Multi-component spatial-temporal graph convolution networks for traffic flow forecasting[J]. Journal of Software, 2019, 30(3): 759-769.
[20]	Cui Z Y, Henrickson K, Ke R M, et al. Traffic graph convolutional recurrent neural network: A deep learning framework for network-scale traffic learning and forecasting[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 21(11): 4883-4894.
[21]	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.
[22]	Deng L Y, Lian D F, Huang Z Y, et al. Graph convolutional adversarial networks for spatiotemporal anomaly detection[J]. IEEE Transactions on Neural Networks and Learning System, 2022, 33(6): 2416-2428.
[23]	Zhang H Y, Zhao S Y, Liu R H, et al. Automatic traffic anomaly detection on the road network with spatial-temporal graph neural network representation learning[J]. Wireless Communications & Mobile Computing, 2022: No.4222827.
[24]	Keogh E, Ratanamaharana C A. Exact indexing of dynamic time warping[J]. Knowledge and Information Systems, 2005, 7(3): 358-386.
[25]	Ge L, Li S Y, Wang Y Q, et al. Global spatial-temporal graph convolutional network for urban traffic speed prediction[J]. Applied Sciences, 2020, 10(4): 1509.
[26]	Qiu J, Dong Y, Ma H, et al. Network embedding as matrix factorization: unifying deepwalk, line, PTE, and node2vec[C]∥The 11th ACM International Conference on Web Search and Data Mining. New York: ACM, 2018: 459-467.
[27]	李海涛, 李志慧, 王鑫, 等. 基于时间卷积自编码网络的实时交通事件自动检测方法[J]. 中国公路学报, 2022, 35(6): 265-276.
	Li Hai-tao, Li Zhi-hui, Wang Xin, et al. Real-time automatic method of detecting traffic incidents based on temporal convolutional autoencoder network[J]. China Journal of Highway and Transport, 2022, 35(6): 265-276.
[28]	Wang J J, Kumbasar T. Parameter optimization of interval Type-2 fuzzy neural networks based on PSO and BBBC methods[J]. IEEE/CAA Journal of Automatica Sinica, 2019, 6(1): 247-257.
[29]	Balke K, Conrad L D, Christopher E M. Using probe-measured travel times to detect major freeway incidents in Houston, Texas[J]. Transportation Research Record, 1996, 1554(1): 213-220.
[30]	Zhao Z, Chen W H, Wu X M, et al. LSTM network: a deep learning approach for short‐term traffic forecast[J]. IET Intelligent Transport Systems, 2017, 11(2): 68-75.
[31]	Kong X, Zhang J, Wei X, et al. Adaptive spatial-temporal graph attention networks for traffic flow forecasting[J]. Applied Intelligence, 2021, 52: 4300-4316.
[32]	Cao Y, Liu D T, Yin Q Z, et al. MSASGCN: Multi-head self-attention spatiotemporal graph convolutional network for traffic flow forecasting[J]. Journal of Advanced Transportation, 2022: 1-15.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

评价指标	美国西雅图数据集
模型	SND	MAD	多元时空动态阈值	贝叶斯分类	支持向量机	随机森林	卡尔曼滤波	STGAN	LSTM+ 异常评估	ASTGAT+ 异常评估	MSASGCN+异常评估	TCA+ 异常评估	GSTPL
检测率（误检率 5%）/%	78.21	80.52	82.79	86.92	84.68	79.25	89.26	89.61	91.34	90.78	91.62	93.15	94.52
检测率（误检率 10%）/%	84.75	87.32	88.43	89.72	86.92	86.25	92.56	94.59	95.32	94.62	95.78	98.61	98.71
误检率（检测率 90%）/%	28.76	25.93	24.81	18.85	20.15	27.74	8.51	7.06	4.40	6.17	34.832	3.52	3.12
误检率（检测率 95%）/%	45.92	43.64	40.25	33.21	39.08	44.52	12.23	10.58	9.12	9.73	9.26	6.72	6.09
平均检测时间/min	3.62	3.58	3.51	8.59	5.36	5.82	4.05	4.52	3.69	3.82	3.72	3.88	3.92
AUC	0.904 3	0.912 4	0.918 2	0.947 6	0.942 5	0.923 5	0.954	0.966 3	0.978 1	0.973 4	0.979 2	0.983 2	0.987 6
评价指标	国内某市数据集
模型	SND	MAD	多元时空动态阈值	贝叶斯分类	支持向量机	随机森林	卡尔曼滤波	STGAN	LSTM+ 异常评估	ASTGAT+ 异常评估	MSASGCN+异常评估	TCA+ 异常评估	GSTPL
检测率（误检率5%）/%	75.63	78.52	79.53	81.57	80.26	77.65	86.12	87.52	89.62	87.96	88.12	89.92	91.43
检测率（误检率10%）/%	81.45	82.46	84.43	87.46	85.17	83.14	89.35	91.92	94.68	93.42	93.62	95.03	96.54
误检率（检测率90%）/%	20.56	19.76	17.52	15.92	17.35	15.93	11.35	8.92	6.15	7.47	68.14	5.03	4.22
误检率（检测率95%）/%	41.62	37.92	38.72	31.54	34.17	29.25	17.62	12.59	10.25	11.38	10.75	9.83	8.23
平均检测时间/min	6.38	5.78	5.12	8.92	7.78	7.52	6.38	6.74	5.53	5.73	5.61	5.42	5.24
AUC	0.853 6	0.861 7	0.8732	0.902 5	0.886 4	0.879 6	0.931 4	0.943 2	0.949 2	0.943 7	0.945 8	0.946 2	0.970 7

评价指标	美国西雅图数据集			国内某市数据集
评价指标	空间图+GSTPL	时间图+GSTPL	时空融合图+GSTPL	空间图+GSTPL	时间图+GSTPL	时空融合图+GSTPL
检测率（误检率5%）/%	92.16	93.72	94.52	89.51	90.17	91.43
检测率（误检率10%）/%	96.43	98.32	98.71	93.54	95.26	96.54
误检率（检测率90%）/%	4.92	4.31	3.12	6.91	4.92	4.22
误检率（检测率95%）/%	8.52	7.16	6.09	10.31	9.73	8.23
平均检测时间/min	3.82	3.96	3.92	5.53	5.42	5.24
AUC	0.965 4	0.980 2	0.987 6	0.941	0.953 6	0.970 7

评价指标	美国西雅图数据集						国内某市数据集
评价指标	近期输入	日周期输入	周周期输入	近期+日周期输入	近期+周周期输入	本文方法	近期输入	日周期输入	周周期输入	近期+日周期输入	近期+周周期输入	本文方法
检测率（误检率5%）/%	92.56	91.28	89.18	93.47	92.35	94.52	89.63	87.12	86.51	90.15	89.94	91.43
检测率（误检率10%）/%	95.76	95.32	94.42	96.54	95.72	98.71	94.18	93.27	92.34	95.13	94.36	96.54
误检率（检测率90%）/%	4.95	4.92	6.62	4.38	4.76	3.12	6.94	8.82	9.32	4.92	6.35	4.22
误检率（检测率95%）/%	9.42	9.87	10.16	8.61	9.63	6.09	10.31	11.25	12.63	9.28	10.18	8.23
平均检测时间/min	3.61	4.37	4.53	4.01	3.84	3.92	5.04	5.73	5.76	5.62	5.13	5.24
AUC	0.961 3	0.951 6	0.943 2	0.976 1	0.963 2	0.987 6	0.950 3	0.948 1	0.942 6	0.952 5	0.962 3	0.970 7

评价指标	美国西雅图数据集			国内某市数据集
评价指标	GSTPL+ 固定阈值	GSTPL+ 动态阈值	本文方法	GSTPL+ 固定阈值	GSTPL+ 动态阈值	本文方法
检测率（误检率5%）/%	84.76	90.31	94.52	85.72	88.49	91.43
检测率（误检率10%）/%	92.15	94.67	98.71	89.67	93.84	96.54
误检率（检测率90%）/%	8.92	4.98	3.12	10.43	8.92	4.22
误检率（检测率95%）/%	14.67	10.37	6.09	13.62	11.36	8.23
平均检测时间/min	3.79	3.83	3.92	5.16 min	5.49 min	5.24 min
AUC	0.951 1	0.965 7	0.987 6	0.941 3	0.962 4	0.970 7