Physics Informed Deep Learning for Traffic State Estimation

• Published in: 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC) pp. 1 - 6
• Main Authors: Huang, Archie J., Agarwal, Shaurya
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20.09.2020
• Subjects: Biological neural networks; Deep learning; Physics; Physics Informed Machine Learning; Real-time systems; Roads; Sensor Placement; State estimation; Traffic State Estimation; Training
• DOI: 10.1109/ITSC45102.2020.9294236

The challenge of traffic state estimation (TSE) lies in the sparsity of observed traffic data and the sensor noise present in the data. This paper presents a new approach - physics informed deep learning (PIDL) method - to tackle this problem. PIDL equips a deep learning neural network with the strength of the physical law governing traffic flow to better estimate traffic conditions. A case study is conducted where the accuracy and convergence-time of the algorithm are tested for varying levels of scarcely observed traffic density data - both in Lagrangian and Eulerian frames. The estimation results are encouraging and demonstrate the capability of PIDL in making accurate and prompt estimation of traffic states.