考虑动态空间关系的短时交通流预测方法

赵振兴; 曾伟; 唐晨嘉

doi:10.3963/j.jssn.1674-4861.2023.04.015

考虑动态空间关系的短时交通流预测方法

doi: 10.3963/j.jssn.1674-4861.2023.04.015

华中科技大学人工智能与自动化学院武汉 430074

基金项目:

国家自然科学基金项目 71871100

详细信息

作者简介:
赵振兴（1999—），硕士研究生. 研究方向：智能交通. E-mail：zx_zhao@hust.edu.cn

通讯作者:
曾伟（1968—），博士，副教授. 研究方向：系统工程. E-mail：zengwei@mail.hust.edu.cn

中图分类号: U491.14
计量
- 文章访问数: 360
- HTML全文浏览量: 190
- PDF下载量: 23
- 被引次数: 0
出版历程
- 收稿日期: 2023-02-10
- 网络出版日期: 2023-11-23

A Short-term Traffic Flow Forecasting Model Considering Dynamic Spatio-temporal Relationship

School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan 430074, China

摘要

摘要: 为有效提取交通流的时空特征，提升交通流的预测精度，研究了基于动态时空图卷积网络的短时交通流预测模型（DySTGCN）。DySTGCN不仅实现了对交通流时空维度的信息建模，而且考虑了时间维度信息对空间维度信息的影响，创新性提出了基于时间信息的空间拓扑结构——时变空间图（spatial topology graph，TSG），并设计出了1种能够高效、简便地计算时变空间图的深层网络结构。该结构通过编码、解码方式提取不同节点的交通流数据的相关性特征并实现降噪处理。时变空间图反映了交通网络的实时空间特征，基于交通网络中节点空间位置的稳定空间图（stable spatial graph，SG）反映了交通网络的稳定空间特征。TSG与SG在图卷积过程中共同指导交通流预测，更加准确地刻画了交通流的时空特性，以提高预测精度。为测试模型的预测效果，在2个权威公开数据集上进行实验，结果表明：DySTGCN学习到的时变空间图可以较为准确地反映出不同节点的交通流之间的相关性，在平均绝对误差、均方根误差，以及加权平均绝对百分比误差指标上，比其他时空图卷积网络模型如STGCN、ASTGCN等降低了近13.40%、10.98%、16.72%，充分验证了动态空间关系在短时交通流预测中的重要作用。此外，DySTGCN能够提取交通流的周期性特征，实现了对交通流的连续不间断预测。
- 交通运输系统工程 /
- 短时交通流预测 /
- 时变空间图 /
- 动态图卷积 /
- 时空融合 /
- 交通大数据
Abstract: A traffic flow prediction model based on dynamic spatio-temporal graph convolutional network (DySTGCN) is developed, to effectively extract the spatio-temporal features of traffic flows and improve the accuracy of traffic flow prediction. DySTGCN models not only Spatio-temporal information of traffic flow but also the influence of temporal information on spatial information. Meanwhile, a spatial structure based on temporal information, a time-varying spatial topology graph (TSG), is proposed and a deep neural network structure to efficiently calculate the TSG is designed. The structure extracts correlation features of traffic flow at different nodes and can reduce the noise through encoding and decoding. TSG reflects the real-time spatial feature of the traffic network, a stable spatial graph (SG) based on the spatial position of nodes reflects the stable spatial feature. The TSG and SG jointly guide the traffic flow prediction and depict the spatio-temporal feature more accurately to improve the prediction precision. To test the prediction effect of the model, experiments are carried out on two authoritative public data sets. The results show that TSG learned by DySTGCN can accurately reflect the correlation between traffic flows at different nodes and MAE, RMSE and WMAPE of DySTGCN are 13.40%, 10.98%, and 16.72% lower than other spatio-temporal graph convolutional network models such as STGCN, ASTGCN, verifying that dynamic spatial relation plays an important role in short-term traffic flow prediction fully. Besides, DySTGCN can extract periodic features of traffic flow and achieve continuous and uninterrupted prediction of traffic flow.
- engineering of communications and transportation system /
- short-term traffic flow prediction /
- time-varying spatial topology graph /
- dynamic graph convolution /
- spatio-temporal fusion /
- big traffic data

HTML全文

图 1 DySTGCN框架

Figure 1. Framework of DySTGCN

下载: 全尺寸图片幻灯片

图 2 TSG估计器框架

Figure 2. Framework of TG estimator

下载: 全尺寸图片幻灯片

图 3 参数选择测试结果

Figure 3. Test results of parameter selection

下载: 全尺寸图片幻灯片

图 4 不同模型对于预测时长对预测时长敏感程度

Figure 4. Sensitivity of prediction length for models

下载: 全尺寸图片幻灯片

图 5 在2个连续时段上生成的部分节点的时变空间图

Figure 5. TSG of partial nodes generated over two consecutive period

下载: 全尺寸图片幻灯片

图 6 不同位置色块对应的2个节点流量变化趋势

Figure 6. Traffic flow trends of two nodes at different location

下载: 全尺寸图片幻灯片

图 7 消融实验结果

Figure 7. The results of ablation study

下载: 全尺寸图片幻灯片

图 8 模型短期预测结果

Figure 8. Short-term prediction of model

下载: 全尺寸图片幻灯片

图 9 7 d的交通流量预测结果

Figure 9. Traffic flow prediction of 7 days

下载: 全尺寸图片幻灯片

表 1 数据集格式

Table 1. Format of dataset

数据集	网络节点总数/ 个	连边总数/条	可用总步长/ 个	数据单位
PeMSD4	307	340	16992	辆/5min
PeMSD8	170	277	17856	辆/5min

下载: 导出CSV

表 2 PeMSD4数据集实验结果

Table 2. Result of PeMSD4 dataset

模型	MAE	RMSE	WMAPE/%
HA	30.96±0.00	46.57±0.00	14.54±0.00
SVR	24.91±0.00	39.15±0.00	11.70±0.00
STGCN	22.43±0.61	34.38±0.73	11.06±0.21
ASTGCN	22.28±1.22	34.86±1.87	11.44±1.24
STSGCN	21.27±0.15	33.58±0.24	10.79±0.07
STFGNN	19.20±0.09	31.36±0.20	9.75±0.09
DySTGCN	18.43±0.11	29.86±0.21	8.94±0.09

下载: 导出CSV

表 3 PeMSD8数据集实验结果

Table 3. Result of PeMSD8 dataset

模型	MAE	RMSE	WMAPE/%
HA	24.51±0.00	37.13±0.00	10.42%±0.00
SVR	19.98±0.00	32.70±0.00	8.49%±0.00
STGCN	18.23±0.14	27.69±0.21	8.29%±0.11
ASTGCN	17.86±0.42	27.58±0.48	8.48%±0.92
STSGCN	17.65±0.09	27.08±0.20	8.16%±0.07
STFGNN	16.03±0.09	25.82±0.18	7.89%±0.06
DySTGCN	15.84±0.10	24.46±0.18	7.18%±0.07

下载: 导出CSV

参考文献(23)

[1]	申雷霄, 陆宇航, 郭建华. 卡尔曼滤波短时交通流预测普通国省道适应性研究[J]. 交通信息与安全, 2021, 39(5): 117-127. doi: 10.3963/j.jssn.1674-4861.2021.05.015 SHEN L X, LU Y H, GUO J H. Adaptability of Kalman filter for short-time traffic flow forecasting on national and provincial highways[J]. Journal of Transport Information and Safety, 2021, 39(5): 117-127(. in Chinese) doi: 10.3963/j.jssn.1674-4861.2021.05.015
[2]	何祖杰, 吴新烨, 刘中华. 基于改进灰狼算法优化支持向量机的短期交通流预测[J]. 厦门大学学报(自然科学版), 2022, 61(2): 288-297. https://www.cnki.com.cn/Article/CJFDTOTAL-XDZK202202019.htm HE Z J, WU X Y, LIU Z H. Optimized SVM model for short-term traffic flow prediction based on improved gray wolf optimizer[J]. Journal of Xiamen University(Natural Science Edition), 2022, 61(2): 288-297(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-XDZK202202019.htm
[3]	童林, 官铮. 改进鲸鱼优化支持向量机的交通流量模糊粒化预测[J]. 计算机应用, 2021, 41(10): 2919-2927. doi: 10.11772/j.issn.1001-9081.2020122048 TONG L, GUAN Z. Fuzzy granulation prediction of traffic flow based on improved whale optimization support vector machine[J]. Journal of Computer Applications, 2021, 41(10): 2919-2927(. in Chinese) doi: 10.11772/j.issn.1001-9081.2020122048
[4]	龚勃文, 林赐云, 李静, 等. 基于核自组织映射—前馈神经网络的交通流短时预测[J]. 吉林大学学报(工学版), 2011, 41(4): 938-943. https://www.cnki.com.cn/Article/CJFDTOTAL-JLGY201104008.htm GONG B W, LIN C Y, LI J, et al. Short-term traffic flow prediction based on KSOM-BP neural network[J]. Journal of Jilin University(Engineering and Technology Edition), 2011, 41(4): 938-943(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-JLGY201104008.htm
[5]	冯金巧, 杨兆升, 孙占全, 等. 基于小波分析的交通参数组合预测方法[J]. 吉林大学学报(工学版), 2010, 40(5): 1220-1224. https://www.cnki.com.cn/Article/CJFDTOTAL-JLGY201005011.htm FENG J Q, YANG Z S, SUN Z Q, et al. Combined method for traffic parameter prediction based on wavelet analysis[J]. Journal of Jilin University(Engineering and Technology Edition), 2010, 40(5): 1220-1224(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-JLGY201005011.htm
[6]	曹洁, 张敏, 张红等. 基于IFA优化RBF神经网络的短时交通流预测模型[J]. 兰州理工大学学报, 2022, 48(4): 99-104. https://www.cnki.com.cn/Article/CJFDTOTAL-GSGY202204015.htm CAO J, ZHANG M, ZHANG H, et al. Short-term traffic flow prediction model based on IFA optimized RBF neural network[J]. Journal of Lanzhou University of Technology, 2022, 48(4): 99-104(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-GSGY202204015.htm
[7]	赵庶旭, 崔方. 一种改进的深度置信网络在交通流预测中的应用[J]. 计算机应用研究, 2019, 36(3): 772-775, 785. https://www.cnki.com.cn/Article/CJFDTOTAL-JSYJ201903027.htm ZHAO S X, CUI F. Application of improved deep belief network in traffic flow forecasting[J]. Application Research of Computers, 2019, 36(3): 772-775, 785(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-JSYJ201903027.htm
[8]	陈宇, 王炜, 华雪东, 等. 基于递归框架的高速公路交通流量实时预测方法[J]. 交通信息与安全, 2023, 41(1): 124-131. doi: 10.3963/j.jssn.1674-4861.2023.01.013 CHEN Y, WANG W, HUA X D, et al. A recursive framework-based approach for real-time traffic flow forecasting for highways[J]. Journal of Transport Information and Safety, 2023, 41(1): 124-131(. in Chinese) doi: 10.3963/j.jssn.1674-4861.2023.01.013
[9]	张玺君, 陶冶, 张冠男, 等. 基于ACapsGRU的短时交通流预测研究[J]. 华中科技大学学报(自然科学版), 2022, 50(4): 51-56. https://www.cnki.com.cn/Article/CJFDTOTAL-HZLG202204009.htm ZHANG X J, TAO Y, ZHANG G N, et al. Research on short-term traffic flow forecast based on ACapsGRU[J]. Journal of Huazhong University of Science and Technology(Natural Science Edition), 2022, 50(4): 51-56(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-HZLG202204009.htm
[10]	MA X, DAI Z, HE Z, et al. Learning traffic as images: A deep convolutional neural network for large-scale transportation network speed prediction[J]. Sensors, 2017, 17(4): 818.
[11]	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. https://www.cnki.com.cn/Article/CJFDTOTAL-HNLG202011013.htm YUAN H, CHEN Z H. Short-term traffic flow prediction based on temporal convolutional networks[J]. Journal of South China University of Technology(Natural Science Edition), 2020, 48(11): 107-113, 122(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-HNLG202011013.htm
[12]	ZHAO L, SONG Y, 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.
[13]	BAI J, ZHU J, SONG Y, et al. A3T-GCN: attention temporal graph convolutional network for traffic forecasting[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 10(7): 485.
[14]	CHEN L, SHAO W, LYU M, et al. AARGNN: An attentive attributed recurrent graph neural network for traffic flow prediction considering multiple dynamic factors[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(10): 17201-17211.
[15]	YU B, YIN H, ZHU Z. Spatio-temporal graph convolutional networks: a deep learning framework for traffic forecasting[C]. The 27^th International Joint Conference on Artificial Intelligence, Freiburg, GER: IJCAI, 2018.
[16]	GUO S, LIN Y, FENG N, et al. Attention based spatial-temporal graph convolutional networks for traffic flow forecasting[C]. 33^th AAAI Conference on Artificial Intelligence, Palo Alto, USA: AAAI, 2019.
[17]	SONG C, LIN Y, GUO S, et al. Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting[C]. 34^th AAAI Conference On Artificial Intelligence, Palo Alto, USA: AAAI, 2020.
[18]	YE J, ZHAO J, YE K, et al. How to build a graph-based deep learning architecture in traffic domain: A survey[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(5): 3904-3924.
[19]	JIA T, YAN P. Predicting citywide road traffic flow using deep spatiotemporal neural networks[J]. IEEE Transactions on Intelligent Transportation Systems, 2021, 22(5): 3101-3111.
[20]	刘宜成, 李志鹏, 吕淳朴, 等. 基于动态时间调整的时空图卷积路网交通流量预测研究[J]. 交通运输系统工程与信息, 2022, 22(3): 147-157, 178. https://www.cnki.com.cn/Article/CJFDTOTAL-YSXT202203017.htm LIU Y C, LI Z P, LYU C P, et al. Network-wide traffic flow prediction research based on DTW algorithm spatial-temporal graph convolution[J]. Journal of Transportation Systems Engineering and Information Technology, 2022, 22(3): 147-157, 178(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-YSXT202203017.htm
[21]	甘萍, 农丽萍, 张文辉, 等. 一种用于交通预测的注意力时空图神经网络[J]. 西安电子科技大学学报, 2023, 50(1): 168-176. https://www.cnki.com.cn/Article/CJFDTOTAL-XDKD202301019.htm GAN P, NONG L P, ZHANG W H, et al. Attention spatial-temporal graph neural network for traffic prediction[J]. Journal of Xidian University, 2023, 50(1): 168-176(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-XDKD202301019.htm
[22]	杨兴锐, 赵寿为, 张如学, 等. 结合自注意力和残差的BiLSTM_CNN文本分类模型[J]. 计算机工程与应用, 2022, 58(3): 172-180. https://www.cnki.com.cn/Article/CJFDTOTAL-JSGG202203015.htm YANG X R, ZHAO S W, ZHANG R X, et al. BiLSTM_CNN classfication model based on self-attention and residual network[J]. Computer Engineering and Applications, 2022, 58(3): 172-180(. in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-JSGG202203015.htm
[23]	刘文星. 网络攻击频率混沌时间序列预测[D]. 长沙: 国防科学技术大学, 2008. LIU W X. Prediction of network attack frequency based on chaotic time series[D]. Changsha: National University of Defense Technology, 2008(. in Chinese)