Graph-based Multi-ODE Neural Networks for Spatio-Temporal Traffic Forecasting

Zibo Liu,Parshin Shojaee,Chandan K Reddy

2023-06-01

Abstract:There is a recent surge in the development of spatio-temporal forecasting models in the transportation domain. Long-range traffic forecasting, however, remains a challenging task due to the intricate and extensive spatio-temporal correlations observed in traffic networks. Current works primarily rely on road networks with graph structures and learn representations using graph neural networks (GNNs), but this approach suffers from over-smoothing problem in deep architectures. To tackle this problem, recent methods introduced the combination of GNNs with residual connections or neural ordinary differential equations (ODE). However, current graph ODE models face two key limitations in feature extraction: (1) they lean towards global temporal patterns, overlooking local patterns that are important for unexpected events; and (2) they lack dynamic semantic edges in their architectural design. In this paper, we propose a novel architecture called Graph-based Multi-ODE Neural Networks (GRAM-ODE) which is designed with multiple connective ODE-GNN modules to learn better representations by capturing different views of complex local and global dynamic spatio-temporal dependencies. We also add some techniques like shared weights and divergence constraints into the intermediate layers of distinct ODE-GNN modules to further improve their communication towards the forecasting task. Our extensive set of experiments conducted on six real-world datasets demonstrate the superior performance of GRAM-ODE compared with state-of-the-art baselines as well as the contribution of different components to the overall performance. The code is available at <a class="link-external link-https" href="https://github.com/zbliu98/GRAM-ODE" rel="external noopener nofollow">this https URL</a>

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

The paper attempts to address the issue of long-range spatiotemporal correlations in traffic prediction. Specifically, existing traffic prediction models have limitations in handling complex spatiotemporal relationships, particularly in capturing local temporal patterns (such as sudden traffic events) and dynamic node-edge associations. These issues lead to inaccuracies in prediction results and insufficient sensitivity to sudden events.

To overcome these limitations, the authors propose a new framework—Graph-based Multi-ODE Neural Networks (GRAM-ODE). This model introduces multiple connected ODE-GNN modules to learn better representations, thereby capturing complex local and global dynamic spatiotemporal dependencies from different perspectives. Additionally, GRAM-ODE incorporates techniques such as shared weights and divergence constraints to further enhance communication between different ODE-GNN modules, thereby improving the performance of the prediction task.

In summary, the main contributions of the paper include:
1. Developing a new ODE module to capture local temporal patterns, which is particularly important for handling sudden events.
2. Introducing a new ODE module to consider the dynamic correlations of traffic nodes and the dynamic semantic correlations of edges.
3. Constructing effective methods to connect different ODE modules, including shared weights between node-based and edge-based ODE modules, and adaptive similarity constraints on the outputs of local and global temporal ODE modules.
4. Designing a new aggregation module that enhances the model's ability to select and fuse features from different ODE modules and different types of traffic graph flows through a multi-head attention mechanism.

Through these innovations, GRAM-ODE aims to provide a more robust solution to address the challenges of complex spatiotemporal correlations in traffic prediction.