Physics Regularized Machine Learning for Smart Mobility
View recorded webinar on YouTube
In recent years, artificial intelligence (AI) such as machine learning (ML) techniques have been widely adopted to develop smart mobility systems. Particularly, as the foundation of smart traffic control systems, many ML-based models have been implemented to model traffic flows. However, despite those ML models showing great potentials in capturing the data uncertainties, they are often criticized by transportation researchers as their performance highly depends on data quality and their results are hard to interpret. To fill such gaps, the main objective of this research is to develop a new fundamental theory, named as physics regularized machine learning (PRML), to use physics knowledge from classical traffic flow models to regularize the training process of ML. The research develops PRML models to deal with both small and large datasets in transportation problems. More specifically, the theoretical frameworks of three sequential models, including PRML, physics regularized streaming learning (PRSL), and physics regularized multi-resolution learning (PRMRL), are delivered. The developed PRML models could have broader applications in transportation such as connected automated vehicle trajectory control. The research is groundbreaking and is expected to bring new insight into ML theory development and applications in the next generation of smart mobility systems.
Xianfeng Yang, Ph.D., is an assistant professor (transportation engineering) in the Department of Civil & Environmental Engineering at the University of Utah. Dr. Yang’s current research areas include machine learning for transportation modeling, traffic operations with connected automated vehicles, traffic safety, transportation equity, transportation planning, etc. He is the recipient of the prestigious NSF CAREER award in 2021. He has published over 100 peer-reviewed research articles in journals and conferences. He is currently the editorial board member of Transportation Research, Part C, the Associate Editor of ASCE Journal of Urban Planning and Development and IEEE OJ-Intelligent Transportation Systems, and the Handling Editor of TRB Transportation Research Record. He is also the Chair of INFORMS JST ITS committee and the Secretary of the ASCE Artificial Intelligence committee. He is the appointed member of two TRB standing committees. His research has been funded by multiple agencies such as the National Science Foundation, U.S. Department of Transportation, Department of Energy, Federal Highway Administration, and Utah Department of Transportation.