MGTTP: A multi-graph transformer model for traffic flow forecasting via bidirectional spatio-temporal interaction

doi:10.1088/1674-1056/ae6635

Abstract Accurate traffic flow forecasting hinges on modeling coupled spatio-temporal dependencies rather than treating space and time in isolation. Many prior methods process spatial and temporal features separately—either in series or in parallel—and then fuse them with simple operators, which weakens their ability to capture intrinsic space-time interactions. We propose multi-graph transformer for traffic flow forecasting (MGTTP), a framework with an innovatively designed bidirectional spatio-temporal interaction mechanism: temporal signals guide multi-graph spatial fusion, while spatial context guides attention-based temporal aggregation. It addresses the limitations of static spatial fusion in existing multi-graph models and the serial spatio-temporal modeling paradigm in vanilla transformer baselines, achieving deep coupled modeling of spatio-temporal features. First, MGTTP builds three complementary graphs—adjacency, reachability, and similarity—and applies temporal feature-guided attention to dynamically fuse their multi-dimensional spatial representations. Subsequently, a transformer encoder captures long-term temporal dependencies, with spatial feature-guided attention to aggregate the time series. Finally, a gated fusion module realizes the ultimate fusion of spatio-temporal features for prediction. Extensive experiments on four public real-world traffic datasets demonstrate that MGTTP outperforms all compared mainstream baseline models across all evaluation metrics, with statistically significant performance gaps, validating the effectiveness of the proposed bidirectional spatio-temporal interaction mechanism.

Keywords: traffic flow forecasting graph neural networks spatio-temporal modeling attention mechanisms transformer

Received: 21 December 2025
Revised: 25 April 2026
Accepted manuscript online: 29 April 2026

PACS:	89.40.Bb	(Land transportation)
	05.60.Gg	(Quantum transport)
	05.45.Tp	(Time series analysis)
	07.05.Mh	(Neural networks, fuzzy logic, artificial intelligence)
	02.70.Hm	(Spectral methods)

Fund: This work was supported by the National Natural Science Foundation of China (Grant No. 62176217), the Sichuan Provincial Science and Technology Program, China (Grant No. 2018RZ0081), and the Fundamental Research Funds of China West Normal University (Grant No. 17E063).

Corresponding Authors: Jialin He
E-mail: hejialin32@126.com

Cite this article:

Xiaolong Fan(范小龙) and Jialin He(何嘉林) MGTTP: A multi-graph transformer model for traffic flow forecasting via bidirectional spatio-temporal interaction 2026 Chin. Phys. B 35 068902

[1] Liao Z F, Sun K, Liu W L, Yu Z W, Liu C G and Song Y C 2024 Chin. Phys. B 33 078901
[2] Hamed M M, Al-Masaeid H R and Said Z M B 1995 J. Transp. Eng. 121 249
[3] Yu B, Yin H and Zhu Z 2017 IJCAI 3634
[4] Yin X Y, Wu Z H, Wang S, Wang Y and Yu P S 2021 IEEE Trans. Intell. Transp. Syst. 23 4927
[5] Zhou H, Zhang S, Peng J, Zhang S, Li J, Xiong H and Zhang W 2021 Proc. AAAI Conf. Artif. Intell. 35 11106
[6] Jiang W and Luo J 2022 Expert Syst. Appl. 207 117921
[7] Li Y, Yu R, Shahabi C and Liu Y 2017 arXiv:1707.01926 [cs.LG]
[8] Song C, Lin Y, Guo S and Wan H 2020 Proc. AAAI Conf. Artif. Intell. 34 914
[9] Geng X, Li Y, Wang L, Zhang L, Yang Q, Ye J and Liu Y 2019 Proc. AAAI Conf. Artif. Intell. 33 3656
[10] Lv M, Hong Z, Chen L, Chen T, Zhu T and Ji Y 2020 IEEE Trans. Intell. Transp. Syst. 22 3337
[11] Sherstinsky A 2020 Phys. D 404 132306
[12] Liu C, Yang S, Xu Q, Wang Y, Zhao J and Wang J 2024 Proc. IEEE Int. Conf. Mobile Data Manag. 31
[13] Wu Z, Pan S, Long G, Jiang J and Zhang C 2019 arXiv:1906.00121 [cs.LG]
[14] Bai L, Yao L, Li C,Wang X andWang C 2020 Adv. Neural Inf. Process. Syst. 33 17804
[15] Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A N, Kaiser Ł and Polosukhin I 2017 Adv. Neural Inf. Process. Syst. 30 5998
[16] Xu M, Dai W, Liu C, Gao X, Lin W, Qi G J and Xiong H 2020 arXiv prep 2020 arXiv:2001.02908 [eess.SP]
[17] Estrada E and Hatano N 2008 Phys. Rev. E 77 036111
[18] Berndt D J and Clifford J 1994 Proc. 3rd Int. Conf. Knowl. Discov. Data Min. 359
[19] Salvador S and Chan P 2007 Intell. Data Anal. 11 561
[20] Lin T Y, Goyal P, Girshick R, He K and Dollár P 2017 Proc. IEEE Int. Conf. Comput. Vis. 2980
[21] Sutskever I, Vinyals O and Le Q V 2014 Adv. Neural Inf. Process. Syst. 27 3104
[22] Guo S, Lin Y, Feng N, Song C andWan H 2019 Proc. AAAI Conf. Artif. Intell. 33 922
[23] Fang Z, Long Q, Song G and Xie K 2021 Proc. ACM SIGKDD Conf. Knowl. Discov. Data Min. 364

[1]	Quantum toric code decoding method based on syndrome-preliminary error fusion module and ResNet architecture Nai-Hua Ji(纪乃华), Ping-Li Song(宋平俐), Wei Wang(王伟), Hui-Qian Sun(孙汇倩), and Hong-Yang Ma(马鸿洋). Chin. Phys. B, 2026, 35(6): 060303.
[2]	Graph neural networks unveil universal dynamics in directed percolation Ji-Hui Han(韩继辉), Cheng-Yi Zhang(张程义), Gao-Gao Dong(董高高), Yue-Feng Shi(石月凤), Long-Feng Zhao(赵龙峰), and Yi-Jiang Zou(邹以江). Chin. Phys. B, 2025, 34(8): 080702.
[3]	Graph distillation with network symmetry Feng Lin(林峰) and Jia-Lin He(何嘉林). Chin. Phys. B, 2025, 34(4): 040204.
[4]	Global receptive field transformer decoder method on quantum surface code data and syndrome error correction Ao-Qing Li(李熬庆), Ce-Wen Tian(田策文), Xiao-Xuan Xu(徐晓璇), Hong-Yang Ma(马鸿洋), and Jun-Qing Liang(梁俊卿). Chin. Phys. B, 2025, 34(3): 030306.
[5]	Combining deep reinforcement learning with heuristics to solve the traveling salesman problem Li Hong(洪莉), Yu Liu(刘宇), Mengqiao Xu(徐梦俏), and Wenhui Deng(邓文慧). Chin. Phys. B, 2025, 34(1): 018705.
[6]	Suppression of ferroresonance using passive memristor emulator S Poornima. Chin. Phys. B, 2021, 30(6): 068401.

No Suggested Reading articles found!

Viewed

Full text

Abstract

Cited

Metrics
Related Articles

MGTTP: A multi-graph transformer model for traffic flow forecasting via bidirectional spatio-temporal interaction

Cite this article:

Online attention