Dynamic fusion language model improves traffic flow predictions
A Dynamic Fusion Large Language Model for Traffic Flow Prediction
Machine Learning
Summary
Predicting traffic patterns helps reduce congestion and plan better roads. Traditional AI models either focus on single types of information or miss important details about how different places and times relate to each other. The authors designed a new model that combines multiple sources of data and uses a language-based AI system to better understand complex traffic patterns. Their model also includes special techniques to keep learning stable and effective. Tests showed this approach predicts traffic flow more accurately than previous methods.
What this means in practice
- •For traffic management centers: Improve real-time routing and congestion control by providing more accurate forecasts of traffic flows using a fused spatiotemporal language model.
- •For smart city planners: Plan infrastructure projects with enhanced predictions of traffic dynamics that integrate spatial and temporal dependencies better than traditional models.
Authors
Xue Qiu, Jianli Xiao
Abstract
Traffic flow prediction is a core supporting technology for intelligent transportation systems. It uses historical data to infer future traffic dynamics in specific areas, thereby helping to alleviate congestion and improve resource allocation efficiency. Traditional neural networks struggle to break through accuracy limits due to their reliance on singular feature modeling, while large language models (LLMs) suffer from insufficient capture of spatial topological information and mining spatiotemporal correlation. This study proposes a Dynamic Fusion Large Language Model (DF-LLM) for traffic flow prediction. The model incorporates three core components: spatiotemporal embedding module, spatiotemporal fusion module, and LLM backbone. The spatiotemporal embedding module enables synergistic representation of multi-scale spatiotemporal features. The spatiotemporal fusion module integrates spatial topology and dynamic dependencies via graph convolution. The LLM backbone adopts a differentiated parameter adaptation strategy to balance training efficiency and traffic data adaptability. Additionally, it introduces a context aggregation attention module to strengthens global dependencies. More importantly, the LLM backbone takes the residual connections to mitigate the gradient vanishing in deep networks. Experiments show that DF-LLM has achieved better performance by comparing the metrics on all the four datasets.