用于交通流预测的时空异质化两阶段融合网络

doi:10.12141/j.issn.1000-565X.240480

摘要/Abstract

摘要：

针对现有交通流预测研究中存在的未能充分融合复杂时空相关性和时空异质性的问题，该文设计了一种基于栅格数据的交通流预测网络——时空异质化两阶段融合网络（Spatiotemporal Heterogeneous Two-Stage Fusion Neural Network，ST_HTFNN）。该网络使用分阶段、层次化的时空特征提取架构，采用静态和动态特征提取阶段串行的新模式，在静态特征提取阶段引入新颖的类曼巴线性注意力（Mamba-Like Linear Attention，MLLA）块作为静态异质化融合单元，实现空间上的相关性和异质性融合挖掘，在动态特征提取阶段设计了简单高效的动态异质化融合单元，通过膨胀卷积和门控机制的结合来自适应融合捕捉全局和局部的时空相关性和异质性。同时，针对细致到道路级的交通流特征，设计了道路特征增强模块来重建和增强道路信息，以解决深度卷积过程中道路特征平滑的问题。最后，设计了外部扰动特征融合模块来融合外部扰动特征对交通流预测结果的影响。在3个现实世界的交通数据集BikeNYC、TaxiCQ和TaxiBJ上进行的模型实验表明，ST_HTFNN模型展现出了超越现有基线模型的卓越性能，相应的预测精度平均绝对误差分别降低了6.13%、0.8%和7.01%。

关键词: 交通流预测, 栅格数据, 时空异质化, 膨胀卷积, 门控机制

Abstract:

In response to the existing traffic flow prediction studies that fail to fully integrate complex spatiotemporal correlations and heterogeneities, this paper designs a traffic flow prediction network based on grid data, namely the spatiotemporal heterogeneous two-stage fusion neural network marked as ST_HTFNN. This network employs a phased and hierarchical spatiotemporal feature extraction architecture, and adopts a new model where the static and dynamic feature extraction stages are serialized. In the static feature extraction stage, a novel Mamba-like linear attention (MLLA) block is introduced as a static heterogeneous fusion unit to achieve spatial correlation and heterogeneity fusion mining. In the dynamic feature extraction stage, a simple and efficient dynamic heterogeneous fusion unit is designed, and dilated convolution is combined with gating mechanisms to adaptively fuse and capture global and local spatiotemporal correlations and heterogeneities. Furthermore, to address the smoothing of road features during the deep convolution process for road-level traffic flow characteristics, a road feature enhancement module is designed to reconstruct and enhance road information. Finally, an external disturbance feature fusion module is designed to integrate the impact of external disturbance features on traffic flow prediction results. Experimental results on three real-world traffic datasets, namely BikeNYC, TaxiCQ and TaxiBJ, demonstrate that the ST_HTFNN model outperforms the existing benchmark methods, respectively with a decrease of 6.13%, 0.8% and 7.01% in the mean absolute error of prediction accuracy.

Key words: traffic flow prediction, grid data, spatiotemporal heterogeneity, dilated convolution, gating mechanism

中图分类号:

TP391

侯越, 尹杰, 张志豪, 卢可可. 用于交通流预测的时空异质化两阶段融合网络[J]. 华南理工大学学报(自然科学版), 2025, 53(5): 82-93.

HOU Yue, YIN Jie, ZHANG Zhihao, LU Keke. A Spatiotemporal Heterogeneous Two-Stage Fusion Network for Traffic Flow Prediction[J]. Journal of South China University of Technology(Natural Science Edition), 2025, 53(5): 82-93.

图/表 15

图1

图2

图3

图4

图5

表1

图6

表2

图7

图8

表3

图9

表4

图10

图11

参考文献 38

1	JILANI U， ASIF M， ZIA M Y I，et al ．A systematic review on urban road traffic congestion［J］．Wireless Personal Communications，2023，140 1-29．
2	ZHANG Y， LI X， WANG H ．Urbanization and traffic congestion：challenges and solutions for sustainable urban mobility［J］．Sustainable Cities and Society，2021，75：103098/1-14．
3	秦严严．交通流分析理论［M］．北京：人民交通出版社，2023：1-4．
4	CHEN B， CHEN Y， WU Y，et al ．The effects of autonomous vehicles on traffic efficiency and energy consumption［J］．Systems，2023，11（7）：347/1-23．
5	潘理虎，尹佳莉，张睿，等．面向交通流预测的全局‒局部时空感知模型［J/OL］．计算机工程，2024．Doi： 10.19678/j.issn.1000-3428.0069550．
	PAN Lihu， YIN Jiali， ZHANG Rui，et al ．Global-local spatio-temporal perception model for traffic flow prediction［J/OL］．Computer Engineering，2024．Doi： 10.19678/j.issn.1000-3428.0069550．
6	崔建勋，要甲，赵泊媛．基于深度学习的短期交通流预测方法综述［J］．交通运输工程学报，2024，24（2）：50-64．
	CUI Jian-xun， YAO Jia， ZHAO Bo-yuan ．Review on short-term traffic flow prediction methods based on deep learning［J］．Journal of Traffic and Transportation Engineering，2024，24（2）：50-64．
7	MEDINA-SALGADO B， SÁNCHEZ-DELACRUZ E， POZOS-PARRA P，et al ．Urban traffic flow prediction techniques：a review［J］．Sustainable Computing：Informatics and Systems，2022，35：100739/1-16．
8	WILLIAMS B M， HOEL L A ．Modeling and forecasting vehicular traffic flow as a seasonal ARIMA process：theoretical basis and empirical results［J］．Journal of Transportation Engineering，2003，129（6）：664-672．
9	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．
10	LUO X， LI D， ZHANG S ．Traffic flow prediction during the holidays based on DFT and SVR［J］．Journal of Sensors，2019，2019（1）：6461450/1-10．
11	CORTES C， VAPNIK V ．Support-vector networks［J］．Machine Learning，1995，20（3）：273-297．
12	BREIMAN L ．Random forests［J］．Machine Learning，2001，45：5-32．
13	LUO C， HUANG C， CAO J，et al ．Short-term traffic flow prediction based on least square support vector machine with hybrid optimization algorithm［J］．Neural Processing Letters，2019，50：2305-2322．
14	ZAREI N， GHAYOUR M A， HASHEMI S ．Road tra-ffic prediction using context-aware random forest based on volatility nature of traffic flows［C］∥Proceedings of the 5th Asian Conference on Intelligent Information and Database Systems．Kuala Lumpur：Springer Berlin Heidelberg，2013：196-205．
15	HINTON G E， SALAKHUTDINOV R R ．Reducing the dimensionality of data with neural networks［J］．Science，2006，313（5786）：504-507．
16	MEDSKER L R， JAIN L ．Recurrent neural networks：design and applications［M］．［S.l.］：CRC Press，1999．
17	HOCHREITER S， SCHMIDHUBER J ．Long short-term memory［J］．Neural Computation，1997，9（8）：1735-1780．
18	LIU Y， ZHENG H， FENG X，et al ．Short-term tra-ffic flow prediction with Conv-LSTM［C］∥Proceedings of 2017 9th International Conference on Wireless Communications and Signal Processing．Nanjing：IEEE，2017：1-6．
19	ZHANG J， ZHENG Y， QI D ．Deep spatio-temporal residual networks for citywide crowd flows prediction［C］∥Proceedings of the AAAI Conference on Artificial Intelligence．San Francisco：AAAI Press，2017：1655-1661．
20	HE K， ZHANG X， REN S，et al ．Deep residual learning for image recognition［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Re-cognition．Las Vegas：IEEE Computer Society，2016：770-778．
21	LIN Z， FENG J， LU Z，et al ．DeepSTN+：context-aware spatial-temporal neural network for crowd flow prediction in metropolis［C］∥Proceedings of the AAAI Conference on Artificial Intelligence．Hawaii：AAAI Press，2019：1020-1027．
22	LIN Z， LI M， ZHENG Z，et al ．Self-attention Conv-LSTM for spatiotemporal prediction［C］∥Proceedings of the AAAI Conference on Artificial Intelligence．New York：AAAI Press，2020：11531-11538．
23	GUO S， LIN Y， LI S，et al ．Deep spatial-temporal 3D convolutional neural networks for traffic data forecasting［J］．IEEE Transactions on Intelligent Transportation Systems，2019，20（10）：3913-3926．
24	HE R， ZHANG C， XIAO Y，et al ．Deep spatio-temporal 3D dilated dense neural network for traffic flow prediction［J］．Expert Systems with Applications，2024，237：121394．
25	GU A，DAO T ．Mamba：linear-time sequence modeling with selective state spaces［EB/OL］．（2023-12-01）［2024-09-20］．．
26	HAN D， WANG Z， XIA Z，et al ．Demystify Mamba in vision：a linear attention perspective［EB/OL］．（2024-05-26）［2024-09-20］．．
27	YU F ．Multi-scale context aggregation by dilated convolutions［EB/OL］．（2014-07-07）［2024-09-20］．．
28	CHOLLET F ．Xception：deep learning with depthwise separable convolutions［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition．Honolulu：IEEE Computer Society，2017：1251-1258．
29	LU Z， LI J， LIU H，et al ．Transformer for single-image super-resolution［C］∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition．New York：IEEE，2022：457-466．
30	VASWANI A ．Attention is all you need［J］．Advances in Neural Information Processing Systems，2017，30：5998-6008．
31	LIU Z， MAO H， WU C Y，et al ．A convnet for the 2020s［C］∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition．Honolulu：IEEE Computer Society，2022：11976-11986．
32	于政．基于深度学习的文本向量化研究与应用［D］．上海：华东师范大学，2016．
33	HE R， LIU Y， XIAO Y，et al ．Deep spatio-temporal 3D densenet with multiscale ConvLSTM-Resnet network for citywide traffic flow forecasting［J］．Knowledge-Based Systems，2022，250（C）：109054/1-17．
34	KINGMA D， BA J ．Adam：a method for stochastic optimization［J］．Computer Science，2014．Doi：10.48550/arXiv.1412.6980．
35	CHEN Y， ZOU X， LI K，et al ．Multiple local 3D CNNs for region-based prediction in smart cities［J］．Information Sciences，2021，542：476-491．
36	HE R， XIAO Y， LU X，et al ．ST-3DGMR：spatio-temporal 3D grouped multiscale ResNet network for region-based urban traffic flow prediction［J］．Information Sciences，2023，624：68-93．
37	WANG Z， BOVIK A C， SHEIKH H R，et al ．Image quality assessment：from error visibility to structural similarity［J］．IEEE transactions on Image Processing，2004，13（4）：600-612．
38	HORE A， ZIOU D ．Image quality metrics：PSNR vs. SSIM［C］∥Proceedings of 2010 20th International Conference on Pattern Recognition．Piscataway：IEEE，2010：2366-2369．

模型		时间特性	空间特性	时空特性	多尺度特性	异质性
传统	HA	√
传统	ARIMA	√
2D CNN	ST-ResNet	√	√		√
2D CNN	DeepSTN+	√	√		√
CNN+RNN	Conv-LSTM	√	√
CNN+RNN	SA-ConvLSTM	√	√
3D CNN	ST-3DNet	√	√	√	√	√
	LMST3D-ResNet	√	√	√	√
	ST-3DGMR	√	√	√	√
	ST-3DMDDN	√	√	√	√	√
SSM	Mamba	√

模型	MSE	MAE	SSIM	PSNR
HA	3.146 3	33.918 0	0.747 8	31.923 8
ARIMA	2.047 4	25.216 9	0.849 9	33.352 2
Conv-LSTM	0.303 7	9.963 0	0.966 9	37.964 1
ST-ResNet	0.248 2	9.366 3	0.967 8	38.386 4
DeepSTN+	0.237 4	9.269 6	0.969 9	38.637 6
ST-3DNet	0.235 3	9.181 6	0.970 9	38.501 1
SA-ConvLSTM	0.212 2	8.396 8	0.976 5	39.314 8
LMST3D-ResNet	0.228 0	8.744 8	0.971 7	38.949 6
ST-3DGMR	0.228 2	8.629 2	0.973 1	39.039 7
Mamba	0.268 5	9.989 9	0.965 9	37.830 0
ST-3DMDDN	0.214 4	8.556 9	0.976 5	39.215 4
ST_HTFNN	0.199 7	7.957 0	0.977 3	39.682 1

模型	MAE	每轮训练时间/s	参数量/10³
Conv-LSTM	9.963 0	5	334.658
ST-ResNet	9.366 3	4	713.348
DeepSTN+	9.269 6	9	33 969.560
ST-3DNet	9.181 6	4	681.786
SA-ConvLSTM	8.396 8	10	426.060
LMST3D-ResNet	8.744 8	17	805.540
ST-3DGMR	8.629 2	15	568.356
Mamba	9.989 9	2	153.088
ST-3DMDDN	8.556 9	25	492.083
ST_HTFNN-w/o-S	8.220 4	13	1 081.760
ST_HTFNN	7.957 0	19	1 087.750

模型	MSE	MAE	SSIM	PSNR
ST_HTFNN-w/o-S	0.206 7	8.220 4	0.976 3	39.388 9
ST_HTFNN-w/o-D	0.210 5	8.260 7	0.975 7	39.382 4
ST_HTFNN-w/o-E	0.203 3	8.025 9	0.977 1	39.590 7
ST_HTFNN-w/o-R	0.206 6	8.071 9	0.976 1	39.575 2
ST_HTFNN	0.199 7	7.957 0	0.977 3	39.682 1

[1]	周楚昊, 林培群, 闫明月. 基于自监督学习的交通数据补全算法[J]. 华南理工大学学报(自然科学版), 2023, 51(4): 101-114.
[2]	袁华陈泽濠. 基于时间卷积神经网络的短时交通流预测算法[J]. 华南理工大学学报(自然科学版), 2020, 48(11): 107-113,122.
[3]	邱敦国兰时勇杨红雨. 基于时空特性的短时交通流预测模型[J]. 华南理工大学学报（自然科学版）, 2014, 42(7): 49-54.