Accurate forecasting of citywide traffic flow has been playing critical role in a variety of spatial-temporal mining applications, such as intelligent traffic control and public risk assessment. While previous work has made significant efforts to learn traffic temporal dynamics and spatial dependencies, two key limitations exist in current models. First, only the neighboring spatial correlations among adjacent regions are considered in most existing methods, and the global interregion dependency is ignored. Additionally, these methods fail to encode the complex traffic transition regularities exhibited with time-dependent and multi-resolution in nature. To tackle these challenges, we develop a new traffic prediction framework–Spatial-Temporal Graph Diffusion Network (ST-GDN). In particular, ST-GDN is a hierarchically structured graph neural architecture which learns not only the local region-wise geographical dependencies, but also the spatial semantics from a global perspective. Furthermore, a multiscale attention network is developed to empower ST-GDN with the capability of capturing multi-level temporal dynamics. Experiments on several real-life traffic datasets demonstrate that ST-GDN outperforms different types of state-ofthe-art baselines. Source codes of implementations are available at https://github.com/jill001/ST-GDN. Introduction Accurate forecasting of traffic flow across different geographical regions in a city, have played a critical role in smart transformation systems, such as intelligent transportation (Wei et al. 2018; Huang et al. 2020) and public risk assessment (Gao et al. 2019; Huang et al. 2018). For example, in disaster control, by predicting future traffic volume, local governments and communities is able to design better transportation scheduling and mobility management strategies, to mitigate the tragedies caused by the crowd flow (Zhao et al. 2017). In general, the objective of traffic prediction is to forecast the traffic volume (e.g., inflow and outflow of each region), from past traffic observations (Diao et al. 2019). *Corresponding author: Chao Huang Copyright © 2021, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. With the advancement of deep learning techniques, many efforts have been devoted to developing traffic prediction methods with various neural network architecture for spatial-temporal pattern modeling. Inspired by the sequence learning paradigm, recent neural networks have been utilized to model temporal effects of traffic variations (Liu et al. 2016; Yu et al. 2017). To make use of spatial features, some research work propose to adopt convolutional neural network to model correlations between adjacent regions (Zhang, Zheng, and Qi 2017), along with using recurrent neural layers on the temporal dimension (Yao et al. 2018). Although both spatial and temporal correlations have been considered in existing methods, several key challenges have not been well addressed. In real-life scenarios, traffic flow pattern is often complex and multi-periodic (Zhang, Zheng, and Qi 2017; Deng et al. 2016), as different views with respect to time resolutions (e.g., hourly, daily, weekly) reflect the traffic dynamics from different temporal dimensions. The captured temporal patterns are often complementary with each other (Wu et al. 2018). Hence, learning accurate representations of traffic variation patterns requires the collaboration of multiple views with different time resolutions. While recurrent neural network-based approaches have achieved good performance on various spatial-temporal sequence prediction tasks, they can only be effective for short-term, smooth dynamics and can hardly make predictions over the high-order multi-dimensional time horizons. Most current forecasting approaches merely focus on modeling nearby geographical correlations (Yao et al. 2018; Zhang, Zheng, and Qi 2017), while ignoring the cross-region inter-dependencies under a global context. For example, two geographical areas with similar urban functions (e.g., shopping zone or transportation hub) can be correlated in terms of their traffic distribution, although they are not spatially adjacent or even far away from each other (Shen et al. 2018; Wang and Li 2017). Hence, the learned region-wise relational structures without the global-level traffic transition information, are insufficient to distill not only local geographical dependencies, but also global relations across regions, The Thirty-Fifth AAAI Conference on Artificial Intelligence (AAAI-21)
[1]
David S. Rosenblum,et al.
UrbanFM: Inferring Fine-Grained Urban Flows
,
2019,
KDD.
[2]
Youfang Lin,et al.
Spatial-Temporal Synchronous Graph Convolutional Networks: A New Framework for Spatial-Temporal Network Data Forecasting
,
2020,
AAAI.
[3]
Xianfeng Tang,et al.
Revisiting Spatial-Temporal Similarity: A Deep Learning Framework for Traffic Prediction
,
2018,
AAAI.
[4]
Jieping Ye,et al.
Deep Multi-View Spatial-Temporal Network for Taxi Demand Prediction
,
2018,
AAAI.
[5]
Yisong Yue,et al.
Long-term Forecasting using Tensor-Train RNNs
,
2017,
ArXiv.
[6]
Dipti Srinivasan,et al.
Computational intelligence-based congestion prediction for a dynamic urban street network
,
2009,
Neurocomputing.
[7]
Cyrus Shahabi,et al.
Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting
,
2017,
ICLR.
[8]
Chao Huang,et al.
Cross-Interaction Hierarchical Attention Networks for Urban Anomaly Prediction
,
2020,
IJCAI.
[9]
Tieniu Tan,et al.
Predicting the Next Location: A Recurrent Model with Spatial and Temporal Contexts
,
2016,
AAAI.
[10]
Junbo Zhang,et al.
Urban Traffic Prediction from Spatio-Temporal Data Using Deep Meta Learning
,
2019,
KDD.
[11]
Dafang Zhang,et al.
Dynamic Spatial-Temporal Graph Convolutional Neural Networks for Traffic Forecasting
,
2019,
AAAI.
[12]
Weimin Zheng,et al.
StepDeep: A Novel Spatial-temporal Mobility Event Prediction Framework based on Deep Neural Network
,
2018,
KDD.
[13]
Xiaoyang Wang,et al.
Traffic Flow Prediction via Spatial Temporal Graph Neural Network
,
2020,
WWW.
[14]
Nitesh V. Chawla,et al.
RESTFul: Resolution-Aware Forecasting of Behavioral Time Series Data
,
2018,
CIKM.
[15]
Yu Zheng,et al.
GeoMAN: Multi-level Attention Networks for Geo-sensory Time Series Prediction
,
2018,
IJCAI.
[16]
Zhenhui Li,et al.
IntelliLight: A Reinforcement Learning Approach for Intelligent Traffic Light Control
,
2018,
KDD.
[17]
Zhenhui Li,et al.
Region Representation Learning via Mobility Flow
,
2017,
CIKM.
[18]
Ugur Demiryurek,et al.
Deep Learning: A Generic Approach for Extreme Condition Traffic Forecasting
,
2017,
SDM.
[19]
Nitesh V. Chawla,et al.
MiST: A Multiview and Multimodal Spatial-Temporal Learning Framework for Citywide Abnormal Event Forecasting
,
2019,
WWW.
[20]
Yu Zheng,et al.
Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction
,
2016,
AAAI.
[21]
Ugur Demiryurek,et al.
Utilizing Real-World Transportation Data for Accurate Traffic Prediction
,
2012,
2012 IEEE 12th International Conference on Data Mining.
[22]
Yanfang Ye,et al.
Incomplete Label Multi-Task Deep Learning for Spatio-Temporal Event Subtype Forecasting
,
2019,
AAAI.
[23]
Jieping Ye,et al.
Feature Constrained Multi-Task Learning Models for Spatiotemporal Event Forecasting
,
2017,
IEEE Transactions on Knowledge and Data Engineering.
[24]
Chih-Jen Lin,et al.
LIBSVM: A library for support vector machines
,
2011,
TIST.
[25]
Nitesh V. Chawla,et al.
DeepCrime: Attentive Hierarchical Recurrent Networks for Crime Prediction
,
2018,
CIKM.
[26]
Xiuwen Yi,et al.
DNN-based prediction model for spatio-temporal data
,
2016,
SIGSPATIAL/GIS.
[27]
Ning Feng,et al.
Attention Based Spatial-Temporal Graph Convolutional Networks for Traffic Flow Forecasting
,
2019,
AAAI.
[28]
Jieping Ye,et al.
Spatiotemporal Multi-Graph Convolution Network for Ride-Hailing Demand Forecasting
,
2019,
AAAI.
[29]
Cheng Wang,et al.
GMAN: A Graph Multi-Attention Network for Traffic Prediction
,
2019,
AAAI.
[30]
Ugur Demiryurek,et al.
Latent Space Model for Road Networks to Predict Time-Varying Traffic
,
2016,
KDD.
[31]
Chao Huang,et al.
Spatial-Temporal Convolutional Graph Attention Networks for Citywide Traffic Flow Forecasting
,
2020,
CIKM.
[32]
Zhanxing Zhu,et al.
Spatio-temporal Graph Convolutional Neural Network: A Deep Learning Framework for Traffic Forecasting
,
2017,
IJCAI.