Lim, Linda (University of California, Berkeley), Moura, Scott (University of California, Berkeley), Delle Monache, Maria Laura (University of California, Berkeley)

From Links to Networks: A Data-Driven, Physics-Informed Koopman Framework for Traffic Networks

Scheduled for presentation during the Regular Session "S30b-Intelligent Modeling and Prediction of Traffic Dynamics" (TH-EA-T30), Thursday, November 20, 2025, 13:30−13:50, Gold Coast

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Modern transportation systems are high-dimensional, non linear, and subject to unpredictable disturbances, making real-time forecasting in large freeway networks especially challenging. This paper presents a data-driven framework that integrates Koopman Mode Decomposition (KMD) with physics-based preprocessing using the Cell Transmission Model (CTM) to enable interpretable network forecasting. We apply this framework to high-resolution traffic data from the San Jose metropolitan area, specifically focusing on three primary corridors: the Downtown Loop, Mid Corridors, and Outer Corridors. CTM enforces flow conservation and capacity limits across exchange zones, consistent with Daganzo's theory, enabling realistic congestion propagation. These physically consistent traffic states allow KMD to extract key spatiotemporal modes reflecting daily traffic rhythms and bottlenecks.The method achieves mean absolute error (MAE) below 2 mph and over 90% accuracy during peak hours, yielding reliable short-term forecasts under congested conditions. Validated using Mobiliti data, the framework is generalizable to other simulators or real-world sources. Overall, this approach supports physics-informed forecasting for multi-segment urban traffic networks and integration with connected and automated vehicle data streams.