基于时空动态约束图反馈的交通流预测

doi:10.13229/j.cnki.jdxbgxb.20240640

摘要/Abstract

摘要：

针对现有交通流预测研究中对路网节点隐藏空间关联时变特性考虑不充分的问题，提出了一种基于时空动态约束图反馈的交通流预测模型。首先，通过门控循环单元（GRU）提取时序特征，在STC-GCL组件内，利用时空图生成器和时空融合约束矩阵生成表征当前时刻路网邻域关系的动态约束图，再利用多层图结构卷积操作实现空间特征提取。其次，利用多尺度门控卷积单元动态调整重要特征信息流，完成对关键特征的精细化筛选。最后，通过将STC-GCL嵌入GRU的方式，实现时空特征的一致性提取。试验在高速路网PeMSD4、PeMSD8、成都-滴滴公开数据集上进行测试，结果表明：与当前主流交通流时空预测方法FGI相比，本文模型的MAE在3个数据集上分别降低了2.69%、1.88%、0.92%。

关键词: 交通流预测, 时空性, 动态性, 图卷积神经网络

Abstract:

Aiming at the problem of insufficient consideration of the time-varying characteristics of hidden spatial associations of road network nodes in the existing traffic flow prediction studies， this paper proposes a traffic flow prediction model based on the feedback of spatio-temporal dynamic constraint graph. First， the temporal features are extracted by GRU， generates a dynamic constraint graph characterising the neighbourhood relationship of the road network at the current moment by using a spatio-temporal graph generator and a spatio-temporal fusion constraint matrix within the STC-GCL component， and then realises spatial feature extraction by using a multilayer graph structure convolution operation. Second， the multi-scale gated convolution unit is used to dynamically adjust the information flow of important features to complete the fine screening of key features. Finally， the consistent extraction of spatio-temporal features is achieved by embedding STC-GCL into GRU. The experiments are tested on the public datasets of high-speed road network PeMSD4， PeMSD8， and Chengdu-DDT， and the results show that compared with the current mainstream spatio-temporal prediction methods for traffic flow FGI， the MAE of the proposed model in this paper reduced by 2.69%， 1.88%， and 0.92% in the three datasets， respectively.

Key words: traffic flow prediction, spatio-temporal nature, dynamic nature, graph convolutional neural network

中图分类号:

U491.1

侯越,张鑫,武月. 基于时空动态约束图反馈的交通流预测[J]. 吉林大学学报(工学版), 2026, 56(1): 183-198.

Yue HOU,Xin ZHANG,Yue WU. Traffic flow prediction based on spatio-temporal dynamic constraint graph feedback[J]. Journal of Jilin University(Engineering and Technology Edition), 2026, 56(1): 183-198.

图/表 22

图1

图2

图3

图4

图5

图6

图7

图8

表1

表2

图9

图10

图11

图12

表3

图13

表4

图14

图15

图16

表5

图17

参考文献 27

[1]	Dimitrakopoulos G, Demestichas P. Intelligent transportation systems[J].IEEE Vehicular Technology Magazine, 2010, 5(1): 77-84.
[2]	Sun Y J, Zhang G H, Yin H H. Passenger flow prediction of subway transfer stations based on nonparametric regression model[J]. Discrete Dynamicsin Nature and Society, 2014, 2014: No. 397154.
[3]	Kumar S V, Vanajakshi L. Short-term traffic flow prediction using seasonal ARIMA model with limited input data[J]. European Transport Research Review, 2015, 7: 1-9.
[4]	Tong J C, Gu X, Zhang M, et al. Traffic flow prediction based on improved SVR for VANET[C]∥The 4th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE), Piscataway, USA, 2021: 402-405.
[5]	谷远利, 张源, 芮小平, 等. 基于免疫算法优化LSSVM的短时交通流预测[J]. 吉林大学学报: 工学版, 2019, 49(6): 1852-1857.
	Gu Yuan-li, Zhang Yuan, Rui Xiao-ping,et al.Short⁃term traffic flow prediction based on LSS-VMoptimized by immune algorithm[J]. Journal of Jilin University(Engineering and Technology Edition), 2019, 49(6): 1852-1857.
[6]	Zaremba W, Sutskever I, Vinyals O. Recurrent neural network regularization[J/OL].[2014-12-22]. .
[7]	Zhang T P, Ding W L, Xing M D. Attention-based dynamic graph convolutional recurrent neural network for traffic flow prediction in highway transportation[J/OL].[2023-12-12]. .
[8]	Graves A. Long short-term memory[J]. Supervised Sequence Labelling with Recurrent Neural Networks, 2012, 385: 37-45.
[9]	曹洁, 苏广, 张红, 等. CapsNet融合D-BiLSTM的区域复杂路网交通速度预测[J]. 吉林大学学报: 工学版,2024,54(9):2531-2539.
	Cao Jie, Su Guang, Zhang Hong, et al. Traffic speed prediction for complex regional road networks based on CapsNet and D-BiLSTM fusion[J]. Journal of Jilin University(Engineering and Technology Edition),2024,54(9):2531-2539 .
[10]	Fu R, Zhang Z, Li L. Using LSTM and GRU neural network methods for traffic flow prediction[C]∥The 31st Youth Academic Annual Conference of Chinese Association of Automation(YAC), Piscataway, USA, 2016: 324-328.
[11]	李桃迎, 王婷, 张羽琪. 考虑多特征的高速公路交通流预测模型[J]. 交通运输系统工程与信息, 2021, 21(3): 101-111.
	Li Tao-ying, Wang Ting, Zhang Yu-qi. Highway traffic flow prediction model with multi-features[J]. Journal of Transportation Systems Engineering and Information Technology, 2021, 21(3): 101-111.
[12]	Ma X L, Dai Z, He Z B, et al. Learning traffic as images: a deep convolutional neural network for large-scale transportation network speed prediction[J]. Sensors, 2017, 17(4): No.818.
[13]	Wu S F. Spatiotemporal dynamic forecasting and analysis of regional traffic flow in urban road networks using deep learning convolutional neural network[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(2): 1607-1615.
[14]	Bruna J, Zaremba W, Szlam A, et al. Spectral n-etworks and locally connected networks on graphs[J/OL].[2013-12-20]. .
[15]	Li Y B, Zhao W, Fan H L. A spatiotemporal graph neural network approach for traffic flow prediction[J]. Mathematics, 2022, 10(10):No. 1754.
[16]	Kipf T N, Welling M. Semi-supervised classification with graph convolutional networks[J/OL].[2016-12-20]. .
[17]	井佩光, 田雨豆, 汪少初, 等. 动态扩散图卷积交通流量预测算法[J]. 吉林大学学报: 工学版, 2024, 54(6): 1582-1592.
	Jing Pei-guang, Tian Yu-dou, Wang Shao-chu,et al.Traffic flow prediction algorithm based on dynamic diffusion graph convolution[J]. Journal of Jilin University(Engineering and Technology Edition), 2024, 54(6): 1582-1592.
[18]	Zhu C F, Yu C X, Huo J Y. Research on spatio-temporal network prediction model of parallel-series traffic flow based on transformer and GCAT[J]. Physica A: Statistical Mechanics and its Applications, 2023, 610: No.128414.
[19]	Zhao L, Song Y J, Zhang C, et al. T-GCN: A temporal graph convolutional network for traffic prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 21(9): 3848-3858.
[20]	Bai L, Yao L N, Li C, et al. Adaptive graph convolutional recurrent network for traffic forecasting[J]. Advances in Neural Information Processing Systems, 2020, 33: 17804-17815.
[21]	Liu A Y, Zhang Y Y. Spatial-temporal interactive dynamic graph convolution network for traffic forecasting[J/OL].[2022-12-20]. .
[22]	Ju X, Xie X S, Yu Y F, Spatial-temporal dynamic fusion graph convolutional network for traffic forecasting[C]//The 5th International Academic Exchange Conference on Science and Technology Innovation (IAECST), Piscataway, USA, 2023: 709-713.
[23]	Cao C, Bao Y, Shi Q, et al. Dynamic spatiotemporal correlation graph convolutional network for traffic speed prediction[J]. Symmetry, 2024, 16(3):No.308.
[24]	Liu M H, Zeng A L, Xu Z, et al. Time series is a special sequence: forecasting with sample convolution and interaction[J/OL].[2021-12-20]. .
[25]	Choi J, Choi H, Hwang J, et al. Graph neural controlled differential equations for traffic forecastting[C]∥Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36(6): 6367-6374.
[26]	Qian W Z, Zhang D L, Zhao Y, et al. Uncertainty quantification for traffic forecasting: A unified approach[C]∥IEEE 39th International Conference on Data Engineering(ICDE), Piscataway, USA, 2023: 992-1004.
[27]	Hou Y, Zhang D, Li D, et al. Regional traffic flow combination prediction model considering virtual space of the road network[J]. Physica A: Statistical Mechanics and its Applications, 2024, 637: No.129598.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

数据集	地区	节点	日期	长度
PeMSD4	旧金山湾区	307	2018年1~2月	16 992
PeMSD8	圣贝纳迪诺区	170	2016年7~8月	17 856
成都-滴滴	成都	524	2018年1~4月	17 280

模型	PeMSD4			PeMSD8			成都-滴滴
模型	MAE	RMSE	R²	MAE	RMSE	R²	MAE	RMSE	R²
CNN-LSTM	20.69	35.05	0.920 1	16.41	26.63	0.926 4	2.32	3.51	0.693 3
ASTGCN	20.44	32.81	0.919 3	16.49	25.36	0.928 5	2.34	3.49	0.692 0
DCRNN	22.28	34.29	0.916 6	16.74	25.63	0.928 8	2.78	4.09	0.581 8
SCINet	20.41	32.48	0.960 0	16.69	26.15	0.938 0	2.25	3.45	0.690 5
STG-NCDE	20.29	31.99	0.960 0	15.82	24.89	0.960 8	2.32	3.44	0.693 2
DeepSTUQ	19.15	31.36	0.933 9	15.04	23.97	0.938 8	2.29	3.44	0.693 3
FGI	19.31	31.68	0.928 2	14.93	23.64	0.939 6	2.18	3.31	0.714 2
STC-GCL-MGRU	18.79	30.28	0.962 8	14.65	23.69	0.963 4	2.16	3.27	0.731 3

模型	5 min			30 min			60 min
模型	MAE	RMSE	R²	MAE	RMSE	R²	MAE	RMSE	R²
SCINet	14.49	22.17	0.940 1	18.12	28.32	0.926 5	19.81	31.14	0.910 4
STG-NCDE	14.22	21.79	0.948 0	17.38	27.44	0.928 7	19.27	30.36	0.912 3
DeepSTUQ	13.54	20.99	0.948 6	16.12	25.46	0.930 0	18.41	28.91	0.917 0
STC-GCL-MGRU	13.26	20.62	0.960 3	15.75	15.14	0.950 3	18.11	28.71	0.941 3

模型	PeMSD8
模型	MAE	RMSE	R²
GRU-GCL	23.09	33.29	0.950 0
STC-GCL-ADD-IG	17.64	28.39	0.961 4
STC-GCL-ADD-ST	17.19	27.79	0.963 1
STC-GCL-MGRU	14.65	23.69	0.963 4

模型	PeMSD8
模型	MAE	RMSE	R²
STC-GCL-MGRU-IG	20.23	31.93	0.951 0
STC-GCL-MGRU-ST	18.41	29.36	0.958 7
STC-GCL-MGRU-GCU	17.19	27.79	0.963 1
STC-GCL-MGRU	14.65	23.69	0.963 4