PROJECT TITLE :

Physics-Informed Deep Learning for Traffic State Estimation

ABSTRACT:

The lack of observed traffic data, in addition to the sensor noise that is present in the data, is the source of the difficulty associated with traffic state estimation (TSE). In order to solve this issue, the authors of this paper present a novel strategy known as the physics informed deep learning (PIDL) method. The purpose of PIDL is to improve the accuracy of traffic condition estimates by endowing a deep learning neural network with the power of the physical law that governs traffic flow. An application study is carried out, in which the precision and convergence time of the algorithm are evaluated for a range of different levels of infrequently observed traffic density data, and this is done in both the Lagrangian and the Eulerian frame. The results of the estimation are encouraging, and they demonstrate the capability of PIDL in making accurate and prompt estimations of traffic states.