Path flow reconstruction under sparse automatic vehicle identification coverage and limited probe vehicle penetration

Reconstructing path flows for transport planning is challenging due to dimensional discrepancies between unknown estimates and traffic observations, limitations of models in capturing real-world behavior, and data acquisition difficulties. We introduce a framework to address these challenges in environments with sparse automatic vehicle identification (AVI) coverage and limited probe vehicle penetration. This framework integrates deep spatio-temporal residual networks (ST-ResNet) with a path flow estimator (PFE) based on the stochastic user equilibrium assumption. ST-ResNet uses historical probe vehicle data to learn driving behaviors and provides initial grid flow ratios, while the PFE refines these estimates using AVI data. Our method was tested under various AVI and probe vehicle penetration rates, demonstrating superior performance in reconstructing path flows compared to other approaches.