基于图小波卷积神经网络的时空图挖掘模型

doi:10.19678/j.issn.1000-3428.0065313

摘要/Abstract

摘要：

针对传统时空图网络模型对时空序列数据空间结构刻画和时空特性挖掘不充分的问题，提出一种基于图小波神经网络的时空图挖掘模型(ST-GWNN)。基于图小波神经网络通过学习节点特征的局部化表达来捕捉时空序列数据中的空间拓扑结构，时间门控卷积层通过门控线性单元所堆叠的因果卷积来提取时间特征信息，并将多个时间步的空间图相融合来学习时间和空间2个维度关联特征的能力，以更好地捕获时空序列中复杂的时空相关性信息。在公共交通数据集PEMS-BAY上的实验结果表明，ST-GWNN模型能够获得较好的预测效果，当预测时长为15 min时，在MAE、RMSE、MAPE 3个评价指标上相较于基准模型取得最小值，且较基准模型最优值分别降低了2.31%、6.96%、5.84%；当预测时长为30 min和60 min时，较基准模型最优的MAPE、RMSE值分别降低了4.9%、3.51%和6.05%、6.68%，可适用于图网络属性的时空关系预测任务。

关键词: 时空图, 图神经网络, 时空序列数据, 图小波网络, 因果卷积

Abstract:

To address the insufficiency of traditional spatio-temporal graph network models in describing the spatial structure and mining of the spatio-temporal characteristics of spatio-temporal series data, this study proposes a Spatio-Temporal graph mining model based on a Graph Wavelet convolutional Neural Network(ST-GWNN). The model adopts a graph wavelet neural network to capture the spatial topology in spatio-temporal sequence data by learning the localized representation of node features. The temporally gated convolutional layer extracts temporal feature information by stacking causal convolutions of gated linear units. In addition, the model fuses the spatial graphs of multiple time steps and can simultaneously learn the correlation features of the two dimensions of time and space, thus better enabling complex spatio-temporal correlation information to be captured in the spatio-temporal sequence. Experimental results on the public transport dataset PMS-Bay show that the ST-GWNN model has better prediction effects. When the prediction time is 15 min, the ST-GWNN model can obtain the minimum values of Mean Absolute Error(MAE), Root Mean Square Error(RMSE), and Mean Absolute Percentage Error(MAPE), which are 2.31%, 6.96%, and 5.84% less, respectively, than the optimal values of the benchmark model. When the prediction time is 30 and 60 min, the MAPE and RMSE decrease by 4.9% and 3.51% and by 6.05% and 6.68%, respectively, as compared with those of the benchmark model, which can be applied to the spatio-temporal relationship prediction task of graph network attributes.

Key words: spatio-temporal graph, graph neural network, spatio-temporal sequence data, graph wavelet network, causal convolution

赵世豪, 毛国君, 熊保平, 黄山, 林江宏. 基于图小波卷积神经网络的时空图挖掘模型[J]. 计算机工程, 2023, 49(7): 85-93.

Shihao ZHAO, Guojun MAO, Baoping XIONG, Shan HUANG, Jianghong LIN. Spatio-temporal Graph Mining Model Based on Graph Wavelet Convolutional Neural Network[J]. Computer Engineering, 2023, 49(7): 85-93.

https://www.ecice06.com/CN/Y2023/V49/I7/85

图/表 10

图1 时空图预测模型演化过程

Fig.1 Evolution process of the spatio-temporal graph prediction model

图2 ST-GWNN模型架构

Fig.2 Architecture of ST-GWNN model

图3 ST-GWNN模型的时间卷积层结构与因果卷积计算过程

Fig.3 Temporal convolution layer structure and causal convolution calculation process of ST-GWNN model

图4 图小波卷积操作的结构

Fig.4 Structure of the graph wavelet convolution operation

图5 PEMS-BAY传感器网络与加权邻接矩阵的可视化

Fig.5 Visualization of PEMS-BAY sensor network and weighted adjacency matrix

表1 在PEMS-BAY数据集上的不同模型性能对比分析

Table 1 Comparative analysis of the performance of different models in the PEMS-BAY dataset

模型	预测时长为15 min			预测时长为30 min			预测时长为60 min
模型	MAE	MAPE/%	RMSE	MAE	MAPE/%	RMSE	MAE	MAPE/%	RMSE
HA	2.88	6.80	5.59	2.88	6.80	5.59	2.88	6.80	5.59
ARIMA	1.62	3.50	3.30	2.33	5.40	4.76	3.38	8.30	6.50
VAR	1.74	3.60	3.16	2.32	5.00	4.25	2.93	6.50	5.44
SVR	1.85	3.80	3.59	2.48	5.50	5.18	3.28	8.00	7.08
FNN	2.20	5.19	4.42	2.30	5.43	4.63	2.46	5.89	4.98
FC-LSTM	2.05	4.80	4.19	2.20	5.20	4.55	2.37	5.70	4.96
DCRNN	1.38	2.90	2.95	1.74	3.90	3.97	2.07	4.90	4.74
STGCN	1.37	2.95	2.98	1.85	4.22	4.34	2.49	5.81	5.70
ASTGCN	1.52	3.22	3.13	2.01	4.48	4.27	2.61	6.00	5.42
Graph WaveNet	1.30	2.73	2.74	1.66	3.67	3.70	1.95	4.63	4.52
STSGCN	1.44	3.04	3.01	1.83	4.17	4.18	2.26	5.40	5.21
MTGNN	1.32	2.77	2.79	1.65	3.69	3.74	1.94	4.53	4.49
FC-GAGA	1.36	2.87	2.86	1.68	3.71	3.80	1.97	4.67	4.52
ST-GWNN	1.27	2.54	2.58	1.68	3.49	3.57	2.00	4.35	4.19

图6 ST-GWNN模型与基准模型的差值箱线图

Fig.6 Box line plot of the difference between the ST-GWNN model and the benchmark model

图7 不同预测时长下随尺度因子s变化的ST-GWNN模型性能分析

Fig.7 Performance analysis of ST-GWNN model varying with scale factor s under different prediction durations

图8 不同尺度因子s下模型训练损失函数的变化曲线

Fig.8 Change curves of the model training loss function under different scale factors s

图9 不同预测时长下Spectral CNN与ST-GWNN模型的性能对比

Fig.9 Performance comparison of Spectral CNN and ST-GWNN model with different prediction durations

参考文献 28

1	YU B, LEE Y J, SOHN K. Forecasting road traffic speeds by considering area-wide spatio-temporal dependencies based on a graph convolutional neural network. Emerging Technologies, 2020, 114, 189- 204. doi: 10.1016/j.trc.2020.02.013
2	刘博, 王明烁, 李永, 等. 深度学习在时空序列预测中的应用综述. 北京工业大学学报, 2021, 47 (8): 925- 941. URL
	LIU B, WANG M S, LI Y, et al. Deep learning for spatio-temporal sequence forecasting: a survey. Journal of Beijing University of Technology, 2021, 47 (8): 925- 941. URL
3	WU Z, PAN S, LONG G, et al. Graph wavenet for deep spatial-temporal graph modeling[EB/OL]. [2022-06-10]. https://arxiv.org/abs/1906.00121.
4	HAN S, KIM J. Video scene change detection using convolution neural network[C]//Proceedings of International Conference on Information Technology. New York, USA: ACM Press, 2017: 116-119.
5	MA T S, KUANG P, TIAN W H. An improved recurrent neural networks for 3d object reconstruction. Applied Intelligence, 2020, 50 (3): 905- 923. doi: 10.1007/s10489-019-01523-3
6	PRADHYUMNA P, SHREYA G P, MOHANA. Graph neural network in image and video understanding using deep learning for computer vision applications[C]//Proceedings of the 2nd International Conference on Electronics and Sustainable Communication Systems. Washington D. C., USA: IEEE Press, 2021: 1183-1189.
7	ROY A, ROY K K, ALI A A, et al. Unified spatio-temporal modeling for traffic forecasting using graph neural network[C]//Proceedings of International Joint Conference on Neural Networks. Washington D. C., USA: IEEE Press, 2021: 1-8.
8	徐冰冰, 岑科廷, 黄俊杰, 等. 图卷积神经网络综述. 计算机学报, 2020, 43 (5): 755- 780. URL
	XU B B, CEN K T, HUANG J J, et al. A survey on graph convolutional neural network. Chinese Journal of Computers, 2020, 43 (5): 755- 780. URL
9	WU Z H, PAN S R, CHEN F W, et al. A comprehensive survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32 (1): 4- 24. doi: 10.1109/TNNLS.2020.2978386
10	LI Y G, YU R, SHAHABI C, et al. Diffusion convolutional recurrent neural network: data-driven traffic forecasting[EB/OL]. [2022-06-10]. https://arxiv.org/abs/1707.01926.
11	CHEN C, LI K L, TEO S G, et al. Gated residual recurrent graph neural networks for traffic prediction. Artificial Intelligence, 2019, 33 (1): 485- 492. URL
12	HAMMOND D K, VANDERGHEYNST P, GRIBONVAL R. Wavelets on graphs via spectral graph theory. Applied and Computational Harmonic Analysis, 2011, 30 (2): 129- 150. doi: 10.1016/j.acha.2010.04.005
13	XU B, SHEN H, CAO Q, et al. Graph wavelet neural network[EB/OL]. [2022-06-10]. https://arxiv.org/abs/1904.07785.
14	SEO Y, DEFFERRARD M, VANDERGHEYNST P, et al. Structured sequence modeling with graph convolutional recurrent networks. Berlin, Germany: Springer, 2018: 362- 373.
15	KADAMBARI S K, PRABHAKAR CHEPURI S. Fast graph convolutional recurrent neural networks[C]//Proceedings of the 53rd Conference on Signals, Systems, and Computers. Washington D. C., USA: IEEE Press, 2020: 467-471.
16	NAGY A M, SIMON V. Survey on traffic prediction in smart cities. Pervasive and Mobile Computing, 2018, 50, 148- 163. doi: 10.1016/j.pmcj.2018.07.004
17	ZHAO L, SONG Y J, ZHANG C, et al. T-GCN: a temporal graph convolutional network for traffic prediction. IEEE Transactions on Intelligent Transportation Systems, 2020, 21 (9): 3848- 3858. doi: 10.1109/TITS.2019.2935152
18	YU B, YIN H T, ZHU Z X. Spatio-temporal graph convolutional networks: a deep learning framework for traffic forecasting[EB/OL]. [2022-06-10]. https://arxiv.org/abs/1709.04875.
19	KIPF T N, WELLING M. Semi-supervised classification with graph convolutional networks[EB/OL]. [2022-06-10]. https://arxiv.org/abs/1609.02907.
20	黎维, 陶蔚, 周星宇, 等. 时空序列预测方法综述. 计算机应用研究, 2020, 37 (10): 2881- 2888. doi: 10.19734/j.issn.1001-3695.2019.05.0184
	LI W, TAO W, ZHOU X Y, et al. Survey of spatio-temporal sequence prediction methods. Application Research of Computers, 2020, 37 (10): 2881- 2888. doi: 10.19734/j.issn.1001-3695.2019.05.0184
21	DAUPHIN Y N, FAN A, AULI M, et al. Language modeling with gated convolutional networks[EB/OL]. [2022-06-10]. https://arxiv.org/abs/1612.08083.
22	BAI S J, KOLTER J Z, KOLTUN V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling[EB/OL]. [2022-06-10]. https://arxiv.org/abs/1803.01271.
23	CHEN X, WANG J S, XIE K Q. TrafficStream: a streaming traffic flow forecasting framework based on graph neural networks and continual learning[EB/OL]. [2022-06-10]. https://arxiv.org/abs/2106.06273.
24	SUTSKEVER I, VINYALS O, LE Q V. Sequence to sequence learning with neural networks[EB/OL]. [2022-06-10]. https://arxiv.org/abs/1409.3215.
25	GUO S N, LIN Y F, FENG N, et al. Attention based spatial-temporal graph convolutional networks for traffic flow forecasting. Artificial Intelligence, 2019, 33 (1): 922- 929. URL
26	SONG C, LIN Y F, GUO S N, et al. Spatial-temporal synchronous graph convolutional networks: a new framework for spatial-temporal network data forecasting. Artificial Intelligence, 2020, 34 (1): 914- 921. URL
27	WU Z H, PAN S R, LONG G D, et al. Connecting the dots: multivariate time series forecasting with graph neural networks[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York, USA: ACM Press, 2020: 753-763.
28	ORESHKIN B N, AMINI A, COYLE L, et al. FC-GAGA: fully connected gated graph architecture for spatio-temporal traffic forecasting. Artificial Intelligence, 2021, 35 (10): 9233- 9241. URL

[1]	何杏宇, 周易歆, 罗东旭, 杨桂松. 基于图神经网络和多主体评价的教学资源推荐[J]. 计算机工程, 2024, 50(7): 13-22.
[2]	张正康, 杨丹, 聂铁铮, 寇月. 基于图结构聚类的自监督学习疾病诊断方法[J]. 计算机工程, 2024, 50(7): 360-371.
[3]	孙文洁, 李宗民, 孙浩淼. 基于图神经网络的多智能体强化学习值函数分解方法[J]. 计算机工程, 2024, 50(5): 62-70.
[4]	游奔, 李晓红, 姚锦, 冯绍杰. 基于多粒度图与注意力机制的半监督短文本分类[J]. 计算机工程, 2024, 50(5): 83-90.
[5]	张宝鑫, 杨丹, 聂铁铮, 寇月. 基于自监督的多视角图协同过滤推荐方法[J]. 计算机工程, 2024, 50(5): 100-110.
[6]	吴正洋, 张广涛, 黄立, 汤庸. 基于异质图嵌入和会话交互的课程推荐模型[J]. 计算机工程, 2024, 50(4): 95-103.
[7]	王华珍, 许泽, 孙悦, 丘斌, 陈坚, 邱强斌. 基于事件演化图的多标记事件预测模型[J]. 计算机工程, 2024, 50(4): 132-140.
[8]	王俊, 赖会霞, 万玥, 张仕. 基于角度的图神经网络高维数据异常检测方法[J]. 计算机工程, 2024, 50(3): 156-165.
[9]	王柏涵, 姜晓燕, 范柳伊. 基于深度监督隐空间构建的语义分割改进方法[J]. 计算机工程, 2024, 50(3): 191-199.
[10]	李浩阳, 贺小伟, 王宾, 吴昊, 尤琪. 基于改进Informer的云计算资源负载预测[J]. 计算机工程, 2024, 50(2): 43-50.
[11]	杜海军, 余粟. 基于时空图注意力网络的服务机器人动态避障[J]. 计算机工程, 2024, 50(2): 105-112.
[12]	李伟健, 胡慧君. 基于潜在特征增强网络的视频描述生成方法[J]. 计算机工程, 2024, 50(2): 266-272.
[13]	吴志强, 解庆, 李琳, 刘永坚. 基于多模态融合的图神经网络推荐算法[J]. 计算机工程, 2024, 50(1): 91-100.
[14]	刘昀抒, 申彦明, 齐恒, 尹宝才. 基于层次结构图的多跳知识图谱问答模型[J]. 计算机工程, 2024, 50(1): 101-109.
[15]	顾嘉静, 杨丹, 聂铁铮, 寇月. 基于多视图融合跨层对比学习的推荐算法[J]. 计算机工程, 2024, 50(1): 120-128.

选择文件类型/文献管理软件名称

选择包含的内容