Abstract:Accurate traffic prediction is a challenging task in intelligent transportation systems because of the complex spatio-temporal dependencies in transportation networks. Many existing works utilize sophisticated temporal modeling approaches to incorporate with graph convolution networks (GCNs) for capturing short-term and long-term spatio-temporal dependencies. However, these separated modules with complicated designs could restrict effectiveness and efficiency of spatio-temporal representation learning. Furthermore, most previous works adopt the fixed graph construction methods to characterize the global spatio-temporal relations, which limits the learning capability of the model for different time periods and even different data scenarios. To overcome these limitations, we propose an automated dilated spatio-temporal synchronous graph network, named Auto-DSTSGN for traffic prediction. Specifically, we design an automated dilated spatio-temporal synchronous graph (Auto-DSTSG) module to capture the short-term and long-term spatio-temporal correlations by stacking deeper layers with dilation factors in an increasing order. Further, we propose a graph structure search approach to automatically construct the spatio-temporal synchronous graph that can adapt to different data scenarios. Extensive experiments on four real-world datasets demonstrate that our model can achieve about 10% improvements compared with the state-of-art methods. Source codes are available at <a class="link-external link-https" href="https://github.com/jinguangyin/Auto-DSTSGN" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

This paper attempts to solve two main problems in traffic prediction: 1. **Capturing complex long - and short - term spatio - temporal dependencies**: - Existing methods usually use graph convolutional networks (GCNs) and some time - modeling components (such as RNN, LSTM, GRU, etc.) to capture spatial and temporal correlations respectively. However, the design of these separate modules is complex and it is difficult to effectively capture more complex spatio - temporal dependencies. - For example, although RNN and its variants (such as LSTM and GRU) can alleviate the problem of vanishing or exploding gradients to a certain extent, their computational efficiency is low. The self - attention mechanism (such as Transformer) can capture long - term dependencies, but it is still time - consuming in the training and inference stages. TCN strikes a balance between performance and efficiency, but it is difficult to capture long - term dependencies due to the fixed convolution kernel size. 2. **Flexible spatio - temporal graph construction**: - Most existing works adopt fixed methods to construct spatio - temporal graphs, which limits the model's adaptability to different time periods and different datasets. Even on different datasets, the graph construction method cannot be adjusted. - Although learnable masks and adaptive graphs have been introduced to overcome the limitations of fixed spatio - temporal graphs, they are still global modeling, unable to depict spatio - temporal relationships for different time periods, and are weakly associated with the characteristics of the data itself, resulting in poor interpretability. To solve these problems, the authors propose a new framework named **Auto - DSTSGN (Automatic Dilated Spatio - Temporal Synchronous Graph Network)**. Specifically: - **Dilated spatio - temporal synchronous graph framework**: By stacking deeper layers with dilation factors, it can flexibly capture short - and long - term spatio - temporal complex dependencies. The increase in the dilation factor enlarges the receptive field, enabling it to effectively capture long - term dependencies while maintaining a low computational burden and GPU occupancy rate. - **Graph structure search operation**: Based on the DARTS framework, a graph structure search operation is proposed to automatically construct diverse spatio - temporal synchronous graphs (STSG) according to different input data to meet the needs of different time periods and datasets. This is the first attempt to use automated machine learning methods to search for adjacency matrices rather than neural architectures. - **Experimental verification**: Extensive experiments were carried out on four real - world datasets, and the results show that this model can achieve an improvement of about 4.9% - 10.3% compared with the existing state - of - the - art methods. Through these innovations, Auto - DSTSGN can more effectively capture complex spatio - temporal dependencies in traffic prediction tasks and improve the adaptability and flexibility of the model.

Automated Dilated Spatio-Temporal Synchronous Graph Modeling for Traffic Prediction

A Spatial-Temporal Gated Hypergraph Convolution Network for Traffic Prediction

Automated Spatio-Temporal Synchronous Modeling with Multiple Graphs for Traffic Prediction

Dynamic multiple-graph spatial-temporal synchronous aggregation framework for traffic prediction in intelligent transportation systems

ADGCN: an Asynchronous Dilation Graph Convolutional Network for Traffic Flow Prediction

Adaptive spatial-temporal graph attention networks for traffic flow forecasting

STG4Traffic: A Survey and Benchmark of Spatial-Temporal Graph Neural Networks for Traffic Prediction

Spatial–Temporal Dynamic Graph Convolutional Network With Interactive Learning for Traffic Forecasting

Dual Dynamic Spatial-Temporal Graph Convolution Network for Traffic Prediction

A Self-Adaptive Spatio-Temporal Graph Convolutional Network for Traffic Prediction

STSD: Modeling Spatial Temporal Staticity and Dynamicity in Traffic Forecasting

Spatial–Temporal-Correlation-Constrained Dynamic Graph Convolutional Network for Traffic Flow Forecasting

Spatio-Temporal Graph Neural Network for Traffic Prediction Based on Adaptive Neighborhood Selection

Attention based spatiotemporal graph attention networks for traffic flow forecasting

Spatial–Temporal Complex Graph Convolution Network for Traffic Flow Prediction

Dynamic multi-scale spatial-temporal graph convolutional network for traffic flow prediction

A dynamical spatial-temporal graph neural network for traffic demand prediction

Dynamic spatial‐temporal network for traffic forecasting based on joint latent space representation

Dynamic Multi-Granularity Spatial-Temporal Graph Attention Network for Traffic Forecasting

Make More Connections: Urban Traffic Flow Forecasting with Spatiotemporal Adaptive Gated Graph Convolution Network

Dynamic Spatiotemporal Straight-Flow Network for Efficient Learning and Accurate Forecasting in Traffic