Generalising Traffic Forecasting to Regions without Traffic Observations

2025-08-12

AI Summary

To address the poor generalization of traffic flow forecasting in sensor-free regions—caused by insufficient historical observational data—this paper proposes GenCast, a novel physics-informed deep learning framework. GenCast integrates Physics-Informed Neural Networks (PINNs) to embed traffic dynamics priors as soft constraints; introduces an external signal learning module to dynamically fuse heterogeneous contextual features (e.g., weather, time-of-day); and incorporates a spatial grouping module to adaptively suppress noisy local patterns and enhance cross-region transferability. By jointly modeling multi-source data, GenCast achieves significant improvements across multiple real-world datasets, reducing MAE and RMSE by an average of 18.3%. Notably, it demonstrates strong robustness and stable prediction performance in zero-shot regions—where no local training data is available. This work establishes a new, interpretable, and generalizable paradigm for traffic state inference in uninstrumented urban environments.

Technology Category

• Planning, Routing, and Scheduling: Model-Based Reasoning
• Search and Optimization: Learning to Search
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Traffic forecasting is essential for intelligent transportation systems. Accurate forecasting relies on continuous observations collected by traffic sensors. However, due to high deployment and maintenance costs, not all regions are equipped with such sensors. This paper aims to forecast for regions without traffic sensors, where the lack of historical traffic observations challenges the generalisability of existing models. We propose a model named GenCast, the core idea of which is to exploit external knowledge to compensate for the missing observations and to enhance generalisation. We integrate physics-informed neural networks into GenCast, enabling physical principles to regularise the learning process. We introduce an external signal learning module to explore correlations between traffic states and external signals such as weather conditions, further improving model generalisability. Additionally, we design a spatial grouping module to filter localised features that hinder model generalisability. Extensive experiments show that GenCast consistently reduces forecasting errors on multiple real-world datasets.

Problem

• Forecast traffic in regions lacking sensor data
• Compensate missing observations using external knowledge
• Improve model generalisability via physics-informed learning

Innovation

• Physics-informed neural networks regularize learning
• External signal module enhances traffic correlations
• Spatial grouping filters localised generalisation hindrances