This paper proposed a novel spatio-temporal graph convolutional network (STGCN) enhanced with a multi-head attention (MHA) mechanism and adaptive adjacency matrices. The MHA mechanism dynamically weighted spatio-temporal features and external environmental factors, while the adaptive adjacency matrices adjusted the connection weights of road networks to improve the model's ability to capture spatial dependencies. Experimental results demonstrated that the model outperformed state-of-the-art (SOTA) models on the London road network dataset, achieving significant improvements across multiple evaluation metrics.
JIANG Tianhao
,
WANG Rui
. Traffic accident prediction based on multi-head attention spatio-temporal graph convolutional network[J]. Journal of Shanghai University, 2025
, 31(4)
: 678
-690
.
DOI: 10.12066/j.issn.1007-2861.2676
[1] Pal C, Hirayama S, Narahari S, et al. An insight of World Health Organization (WHO) accident database by cluster analysis with self-organizing map (SOM) [J]. Tra-c Injury Prevention, 2018, 19: 1-9.
[2] Tamerius, Z X, Mantilla R, Greenfield-Huitt T. Precipitation efiects on motor vehicle crashes vary by space, time, and environmental conditions [J]. Weather, Climate, and Society,
[4] Barba L, Rodriguez N, Montt C. Smoothing strategies combined with ARIMA and neural networks to improve the forecasting of tra-c accidents [J]. The Scientiflc World Journal, 2014(8): 14222674.
[5] Gupta A K K, Ghosh S, Gupta R K. A Bayesian network based framework for real-time crash prediction on the basic freeway segments of urban expressways [J]. Accident Analysis & Prevention, 2012, 45: 373-381.
[6] Lin L, Wang Q, Sadek A W. A novel variable selection method based on frequent pattern tree for real-time tra-c accident risk prediction [J]. Transportation Research Part C: Emerging Technologies, 2015, 55: 444-459.
[7] Ren H, Song Y, Wang J, et al. A deep learning approach to the citywide tra-c accident risk prediction [C]// Proceedings of International Conference on Intelligent Transportation Systems. 2018: 3346-3351.
[8] Zhang J, Zheng Y, Qi D. Deep spatio-temporal residual networks for citywide crowd flows prediction [C]// Proceedings of 31st AAAI Conference on Artiflcial Intelligence. 2017: 1655- 1661.
[9] Chen C, Fan X, Zheng C, et al. SDCAE: stack denoising convolutional autoencoder model for accident risk prediction via tra-c big data [C]// 6th International Conference on Cloud, Big Data and Web Services. 2018: 328-333.
[10] 王竟成, 王雨薇, 高煜光. 基于图卷积神经网络的交通预测研究综述[J]. 北京工业大学学报, 2022, 48(1): 1-12.
[11] 朱凯利, 姜军, 梁朝阳. 基于图神经网络的城市交通流量预测研究[J]. 中国计算机学会通讯, 2021, 17(11): 71-79.
[12] Rodgers J L, Nicewander W A. Thirteen ways to look at the correlation coe-cient [J]. The American Statistician, 1988, 42(1): 59-66.
[13] Kingma D P, Ba J. Adam: a method for stochastic optimization [C]// International Conference on Learning Representations. 2014: 6980.
[14] Tibshirani R. Regression shrinkage and selection via the lasso [J]. Journal of the Royal Statistical Society Series B: Statistical Methodology, 1996, 58(1): 267-288.
[15] Cortes C, Vapnik V. Support-vector networks [J]. Machine Learning, 1995, 20(3): 273-297.
[16] Quinlan J R. Induction of decision trees [J]. Machine Learning, 1986(1): 81-106.
[17] Chen Q, Song X, Yamada H, et al. Learning deep representation from big and heterogeneous data for tra-c accident inference [C]// Proceedings of AAAI Conference on Artiflcial Intelligence. 2016: 1-7. 2016, 8(4): 399-407.
[3] Chang L Y, Chen W C. Data mining of tree-based models to analyze freeway accident frequency [J]. Journal of Safety Research, 2005, 36(4): 365-375.
