Improving traffic and environment forecasts despite changing sensor networks
More Sensors Only One Field: Rethinking Continual Spatio-Temporal Forecasting
Machine Learning
Summary
Forecasting traffic or environmental conditions gets tricky when new sensors are added over time, because traditional methods tie learning to the fixed sensor set. The authors propose a new approach called STFO that treats the data as a shared evolving field, allowing the model to adapt when sensors appear or disappear. This method keeps track of spatial patterns on a fixed grid and adjusts to changes in conditions, improving forecasting accuracy. Their experiments on real-world traffic and air quality data show better predictions than existing methods.
What this means in practice
- •For traffic management teams: Produce accurate short-term traffic forecasts even as new sensors are installed in the network to optimize flow and reduce congestion.
- •For environmental monitoring agencies: Improve prediction of air pollution levels in urban areas despite sensor additions or relocations, aiding timely alerts and policy decisions.
Authors
Lewei Xie, Haoyu Zhang, Jiajun Zhou, Yulong Chen, Guanxing Chen, Yu-An Huang, Hau-San Wong, Yifan Zhang, Zhi-An Huang
Abstract
Continual spatio-temporal forecasting supports traffic management and environmental monitoring under evolving dynamics and expanding sensor networks. However, conventional graph-based continual learning methods tie forecasting representations to the current sensor layout, so sensor expansion can alter the representation of learned spatial relationships. Our key insight is that sensor expansion changes the evidence available about a process without necessarily changing the dynamics to be learned. We propose STFO (Spatio-Temporal Field Operator), which parameterizes forecasting knowledge as a shared field-evolution operator and handles changing sensor layouts through observation and query interfaces. Normalized coordinate-based aggregation lifts irregular sensor histories onto a fixed latent grid, enabling reuse of learned spatial maps across observation sets without sensor-specific parameters. To accommodate process drift, a spectral descriptor summarizes variation across spatial scales and conditions Fourier propagation and attention to adapt operator responses to the current spatial regime. Coordinate-based decoding queries the evolved field at sensor locations and combines spatial corrections with local-history predictions. Experiments on PEMS-Stream, CA-Stream, and AIR-Stream demonstrate state-of-the-art average forecasting performance. STFO-Large reduces average MAE over DOL by 8.4% on PEMS-Stream and 4.7% on CA-Stream. Our code is available at https://github.com/Xielewei/Spatio-Temporal-Field-Operator.