Abstract:Accurate traffic forecasting is crucial for effective urban planning and transportation management, enabling efficient resource allocation and enhanced travel experiences. However, existing models often face limitations in generalization, struggling with zero-shot prediction on unseen regions and cities, as well as diminished long-term accuracy. This is primarily due to the inherent challenges in handling the spatial and temporal heterogeneity of traffic data, coupled with the significant distribution shift across time and space. In this work, we aim to unlock new possibilities for building versatile, resilient and adaptive spatio-temporal foundation models for traffic prediction. To achieve this goal, we introduce a novel foundation model, named OpenCity, that can effectively capture and normalize the underlying spatio-temporal patterns from diverse data characteristics, facilitating zero-shot generalization across diverse urban environments. OpenCity integrates the Transformer architecture with graph neural networks to model the complex spatio-temporal dependencies in traffic data. By pre-training OpenCity on large-scale, heterogeneous traffic datasets, we enable the model to learn rich, generalizable representations that can be seamlessly applied to a wide range of traffic forecasting scenarios. Experimental results demonstrate that OpenCity exhibits exceptional zero-shot predictive performance. Moreover, OpenCity showcases promising scaling laws, suggesting the potential for developing a truly one-for-all traffic prediction solution that can adapt to new urban contexts with minimal overhead. We made our proposed OpenCity model open-source and it is available at the following link: <a class="link-external link-https" href="https://github.com/HKUDS/OpenCity" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the limitations of existing traffic prediction models in terms of generalization capability, particularly in zero-shot prediction and long-term prediction. Specifically, the paper focuses on the following two main issues: 1. **Cross-Regional Spatial Generalization**: - Existing traffic prediction models often perform poorly when applied to unseen regions or cities. This is because traffic patterns and dynamics vary significantly across different geographical areas, and current models typically learn only from data in specific regions, making it difficult to generalize effectively to new traffic environments. - This issue is crucial in practical applications because it is unrealistic to deploy sensor networks comprehensively to collect data for an entire city. Therefore, it is necessary to develop models that can utilize partial data to make effective predictions in unseen regions. 2. **Long-Term Temporal Generalization**: - Current traffic prediction models excel in short-term predictions (e.g., predicting traffic conditions for the next hour) but have significant limitations in long-term predictions (e.g., traffic conditions for the next few days or weeks). This is mainly because these models perform poorly in handling distribution changes over long time ranges. - Long-term traffic prediction is critical for urban planning and traffic management tasks, such as infrastructure planning, public transportation scheduling, and event coordination. However, existing models often fall short in providing reliable long-term predictions. To address these issues, the paper proposes a novel foundational model—OpenCity, which integrates the Transformer architecture and graph neural networks to effectively capture and normalize complex spatiotemporal dependencies in traffic data. Through pre-training on large-scale, heterogeneous traffic datasets, OpenCity can learn rich, transferable representations, enabling seamless application across various traffic prediction scenarios. Experimental results show that OpenCity excels in zero-shot prediction and long-term prediction, demonstrating good scalability and promising to become a truly unified traffic prediction solution.

OpenCity: Open Spatio-Temporal Foundation Models for Traffic Prediction

Short-Term Speed Forecasting of Large-Scale Urban Road Network Based on Transformer

UrbanDiT: A Foundation Model for Open-World Urban Spatio-Temporal Learning

One Size Fits All: A Unified Traffic Predictor for Capturing the Essential Spatial–Temporal Dependency

ESCORT: Fine-Grained Urban Crime Risk Inference Leveraging Heterogeneous Open Data

Cross-city Few-Shot Traffic Forecasting via Traffic Pattern Bank

Bidirectional Spatial-Temporal Adaptive Transformer for Urban Traffic Flow Forecasting

A Novel Spatio-Temporal Model for City-Scale Traffic Speed Prediction

FusionTransNet for Smart Urban Mobility: Spatiotemporal Traffic Forecasting Through Multimodal Network Integration

Spatiotemporal Context Awareness for Urban Traffic Modeling and Prediction: Sparse Representation Based Variable Selection.

An Efficient Traffic Prediction Model Using Deep Spatial-Temporal Network

LibCity: A Unified Library Towards Efficient and Comprehensive Urban Spatial-Temporal Prediction

A Foundation Model for Unified Urban Spatio-Temporal Flow Prediction

An Effective Dynamic Spatio-temporal Framework with Multi-Source Information for Traffic Prediction

Boosting Urban Traffic Speed Prediction via Integrating Implicit Spatial Correlations

Deep spatio-temporal neural network based on interactive attention for traffic flow prediction

Multi-scale Traffic Pattern Bank for Cross-city Few-shot Traffic Forecasting

Urban Traffic Flow Congestion Prediction Based on a Data-Driven Model

MultiSPANS: A Multi-range Spatial-Temporal Transformer Network for Traffic Forecast via Structural Entropy Optimization

TrafficPredict: Trajectory Prediction for Heterogeneous Traffic-Agents

Cross Space and Time: A Spatio-Temporal Unitized Model for Traffic Flow Forecasting