A Physics-Informed Deep Learning Paradigm for Traffic State and Fundamental Diagram Estimation

Traffic state estimation (TSE) bifurcates into two categories, model-driven and data-driven (e.g., machine learning, ML), while each suffers from either deficient physics or small data. To mitigate these limitations, recent studies introduced a hybrid paradigm, physics-informed deep learning (PIDL), which contains both components. This paper contributes an improved version, called with fundamental diagram learner (PIDL+FDL), integrates ML terms the component to learn functional form of (FD), i.e., mapping traffic density flow velocity. The proposed PIDL+FDL has advantages performing TSE model parameter identification, FD simultaneously. We demonstrate use solve popular first-order second-order models reconstruct relation as well parameters that are outside terms. then evaluate PIDL+FDL-based using Next Generation SIMulation (NGSIM) dataset. experimental results show superiority in accuracy data efficiency over advanced baseline methods, additionally, capacity properly unknown underlying relation.