Developing A Physics-Informed Deep Learning Paradigm For Traffic State Estimation