Adaptive spatial-temporal graph attention networks for traffic flow forecasting

Kong, Xiangyuan; Zhang, Jian; Wei, Xiang; Xing, Weiwei; Lü, Wei

doi:10.1007/s10489-021-02648-0

Cited by 41 publications

(19 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, they can only learn between adjacent regions and cannot capture long-range spatial dependencies. Spatial-Temporal Adaptive Fusion Graph Network (STFAGN) [30] combines fused convolutional layers with novel adaptive dependency matrices through end-to-end training to capture hidden spatio-temporal dependencies on the data and obtain hidden spatio-temporal dependencies through fusion operations in parallel. The above methods assume that spatial correlation exists only on corresponding or near nodes.…”

Section: Traffic Flow Predictionmentioning

confidence: 99%

STSGAN: Spatial-Temporal Global Semantic Graph Attention Convolution Networks for Urban Flow Prediction

Zhou

Qin

et al. 2022

IJGI

View full text Add to dashboard Cite

Accurate urban traffic flow prediction plays a vital role in Intelligent Transportation System (ITS). The complex long-term and long-range spatiotemporal correlations of traffic flow pose a significant challenge to the prediction task. Most current research methods focus only on spatial correlations in local areas, ignoring global geographic contextual information. It is challenging to capture spatial information from distant nodes using shallow graph neural networks (GNNs) to model long-range spatial correlations. To handle this problem, we design a novel spatiotemporal global semantic graph-attentive convolutional network model (STSGAN), which is a deep-level network to achieve the simultaneous modelling of spatiotemporal correlations. First, we propose a graph-attentive convolutional network (GACN) to extract the importance of different spatial features and learn the spatial correlation of local regions and the global spatial semantic information. The temporal causal convolution structure (TCN) is utilized to capture the causal relationships between long-short times, thus enabling an integrated consideration of local and overall spatiotemporal correlations. Several experiments are conducted on two real-world traffic flow datasets, and the results show that our approach outperforms several state-of-the-art baselines.

show abstract

Section: Traffic Flow Predictionmentioning

confidence: 99%

STSGAN: Spatial-Temporal Global Semantic Graph Attention Convolution Networks for Urban Flow Prediction

Zhou

Qin

et al. 2022

IJGI

View full text Add to dashboard Cite

show abstract

“…Subsequently, DL has prevailed in correlated time series forecasting because of its superior ability to model complex functions and exploit underlying features without tricky feature engineering [ 26 , 27 ]. The capacity empowers traffic prediction to serve a vital role in ITS (e.g., driving decisions/services [ 28 ]).…”

Section: Related Workmentioning

confidence: 99%

“…The former was collected in 2015 on four connected freeways (I-5, I-405, I-90 and SR-520) in the Greater Seattle Area; the latter was collected on the highways of Los Angeles County from 1 March 2012 to 30 June 2012. Table 1 summarizes the key statistics [ 6 , 10 ] of both datasets, and Figure 5 provides their area maps [ 7 , 27 ]. In this experiment, we chose only the first two months of traffic data for modeling, and the sub-datasets were split in chronological order with 70% serving as the training set and the remaining 30% as the testing set.…”

Section: Experiments Studymentioning

confidence: 99%

“…Similar to [ 2 , 5 , 27 ], we conducted a comprehensive ablation study on the Seattle-Loop and METR-LA datasets to investigate the effectiveness of key components that contribute to the improved results of the ST-CRMF model. We named the ST-CRMF without different components as follows: (1) ST-CRMF w/o GL: ST-CRMF without the graph Laplacian spatial regularizer; (2) ST-CRMF w/o GRN: ST-CRMF without the GRN temporal regularizer; (3) ST-CRMF w/o RL: ST-CRMF without the residual learning; and the baseline for the ablation study was the ST-CRMF model.…”

Section: Experiments Studymentioning

confidence: 99%

See 1 more Smart Citation

ST-CRMF: Compensated Residual Matrix Factorization with Spatial-Temporal Regularization for Graph-Based Time Series Forecasting

Pian

et al. 2022

Sensors

View full text Add to dashboard Cite

Despite the extensive efforts, accurate traffic time series forecasting remains challenging. By taking into account the non-linear nature of traffic in-depth, we propose a novel ST-CRMF model consisting of the Compensated Residual Matrix Factorization with Spatial-Temporal regularization for graph-based traffic time series forecasting. Our model inherits the benefits of MF and regularizer optimization and further carries out the compensatory modeling of the spatial-temporal correlations through a well-designed bi-directional residual structure. Of particular concern is that MF modeling and later residual learning share and synchronize iterative updates as equal training parameters, which considerably alleviates the error propagation problem that associates with rolling forecasting. Besides, most of the existing prediction models have neglected the difficult-to-avoid issue of missing traffic data; the ST-CRMF model can repair the possible missing value while fulfilling the forecasting tasks. After testing the effects of key parameters on model performance, the numerous experimental results confirm that our ST-CRMF model can efficiently capture the comprehensive spatial-temporal dependencies and significantly outperform those state-of-the-art models in the short-to-long terms (5-/15-/30-/60-min) traffic forecasting tasks on the open Seattle-Loop and METR-LA traffic datasets.

show abstract

“…Specifically, when traffic flow at a sensor is congested, the traffic flow around it will also be affected. Recently, the graph method is usually used to analyze the spatial correlation, such as Graph Convolutional Network (GCN) [10], Graph Attention Network (GAT) [11].…”

Section: Introductionmentioning

confidence: 99%

A Hybrid Model Integrating Local and Global Spatial Correlation for Traffic Prediction

et al. 2022

View full text Add to dashboard Cite

Accurate traffic prediction can effectively alleviate traffic congestion problems. The complex spatial correlation of traffic flow contributes to the challenging prediction problem. Most current prediction methods focus on learning local spatial correlation, ignoring the spatial correlation of long-distance traffic flow. In this paper, we combine the improved Graph Convolutional Network (GCN) with Gated Recurrent Unit (GRU) to propose a hybrid model integrating local and global spatial correlation (T-LGGCN) for traffic prediction. The model consists of two parts: global spatial-temporal component and local spatial-temporal component. For the global spatial-temporal component, we construct the global correlation matrix to improve the GCN for obtaining the global spatial correlation. And GRU is stacked to obtain the global spatial-temporal correlation. For the local spatial-temporal component, we utilize the strategy of combining Fully Connected Layer (FCL) and GCN to analyze the local spatial correlation. Similarly, GRU is used to perform the output of local spatial-temporal correlation. The output of the two components is finally summed, and the prediction results are generated with the dense network. Experiments were conducted using the highway datasets PEMS04 and PEMS08 from the Caltrans Performance Measurement System, and the results show that our model significantly outperforms state-of-the-art baselines.

show abstract

Adaptive spatial-temporal graph attention networks for traffic flow forecasting

Cited by 41 publications

References 42 publications

STSGAN: Spatial-Temporal Global Semantic Graph Attention Convolution Networks for Urban Flow Prediction

STSGAN: Spatial-Temporal Global Semantic Graph Attention Convolution Networks for Urban Flow Prediction

ST-CRMF: Compensated Residual Matrix Factorization with Spatial-Temporal Regularization for Graph-Based Time Series Forecasting

A Hybrid Model Integrating Local and Global Spatial Correlation for Traffic Prediction

Contact Info

Product

Resources

About