Building Transportation Foundation Model via Generative Graph Transformer

TL;DR

This paper introduces the Transportation Foundation Model (TFM), a novel data-driven approach using generative graph transformers to improve traffic prediction by capturing complex transportation system interactions.

Contribution

The paper presents TFM, a new generative graph transformer-based model that integrates traffic simulation principles for enhanced urban traffic prediction.

Findings

Accurately predicts urban traffic outcomes

Addresses structural complexity of transportation systems

Utilizes big data effectively

Abstract

Efficient traffic management is crucial for maintaining urban mobility, especially in densely populated areas where congestion, accidents, and delays can lead to frustrating and expensive commutes. However, existing prediction methods face challenges in terms of optimizing a single objective and understanding the complex composition of the transportation system. Moreover, they lack the ability to understand the macroscopic system and cannot efficiently utilize big data. In this paper, we propose a novel approach, Transportation Foundation Model (TFM), which integrates the principles of traffic simulation into traffic prediction. TFM uses graph structures and dynamic graph generation algorithms to capture the participatory behavior and interaction of transportation system actors. This data-driven and model-free simulation method addresses the challenges faced by traditional systems in terms of structural complexity and model accuracy and provides a foundation for solving complex transportation problems with real data. The proposed approach shows promising results in accurately predicting traffic outcomes in an urban transportation setting.