Traffic Dynamics & Modeling: Others

Traffic State Estimation

Our research advances traffic state and travel-time estimation across freeway and urban networks by combining physics-based stochastic modeling with data-driven learning:

• In connected urban environments, we modeled lane-level travel-time distributions by jointly considering link travel time and movement-specific signal delays, showing that kernel density estimation captures travel-time reliability more comprehensively than conventional parametric distributions^[1].
• For recurrently congested freeways with sparse or malfunctioning detectors, we integrated self-organizing maps with support vector regression to identify representative sensors and accurately estimate real-time route travel times from limited data^[2].
• We extended Newell's three-detector method to infer probabilistic traffic states at unobserved freeway locations from upstream and downstream counts while accounting for day-to-day demand variation, sensor errors, and uncertainty in fundamental-diagram parameters^[3].

Publications