On the morning of September 8, Li Nan’s research group held the 10th session of its series of invited academic lectures online. The activity was hosted by Dr. LI Nan, and Dr. YANG Xianfeng from the School of Civil and Environmental Engineering at the University of Utah was invited to present an academic report on Physics Regularized Machine Learning for Smart Mobility Systems. More than ten teachers and students from Li Nan’s research group and the institute of transportation of Tsinghua’s department of civil engineering attended this seminar online.
In recent years, the application of data-driven models and machine learning models in the field of civil engineering has received a large number of scholarly attention. However, the effects of these models are highly dependent on the quality of training data and it is difficult to explain the results, so there are certain limitations in the practical application. In his research, Dr. Yang introduced the constraints of the traditional physics models into the original machine learning model and constructed a Physics Regularized Machine Learning Model to enable it to better adapt to data noise and data fluctuation.
Picture 1 Presentation of Report Content (I)
Dr. YANG focuses on Traffic State Estimation, a classical traffic problem, and uses the physical knowledge in the classical traffic flow prediction model to regulate the training process of machine learning, and establishes a physical regularized machine learning model. Specifically, the model consists of three modules: physics regularized machine learning model, physics regularized stream learning model and physics regularized multi-resolution learning model. Through case analysis, Dr. Yang compared the estimated effects of machine learning model, traditional physics model and physics regularized machine learning model on road traffic state, and found that physics regularized machine learning had a better performance in dealing with data noise and data fluctuation.
Picture 2 Presentation of Report Content (II)
Dr. YANG also presented the theoretical and practical value of his research. Firstly, this study advances the knowledge on the development of machine learning models and their application in smart mobility systems and other fields. Secondly, the developed models can help other researchers to better calibration their own models, and provide guidance for the improvement of the traditional physics models. In addition, the developed models have a very wide range of applications in the field of transportation. For example, trajectory control of automatic driving vehicles can be realized by the physics regularized machine learning method presented in this study. Dr. YANG’s report was followed by a lively discussion session where students put forward their own opinions and questions on such issues as the advantages of physics regularized machine learning, the selection basis of the traditional physics model and data expansion methods for data analysis.
Picture 3 Presentation of Report Content (III)
At the end, Dr. YANG shared with all the students her experiences, thoughts and feelings in the process of studying for a doctorate and engaging in teaching, based on his years of learning and scientific research experience. He mentioned that there should be different priorities at different stages of the PhD and career development. In the process of scientific research, the most important thing was to cultivate the ability of independent research, so doctoral students should pay attention to improving their ability to find problems and put forward ideas. The newly enrolled doctoral students should actively seek advice from those experienced people and learn to manage themselves so as to avoid detours. After possessing the independent research ability, they should learn to lead their research teams, coordinate the work of the members in the research group, communicate with others and do more thinking. After officially becoming a young scholar, they should learn to allocate their time reasonably and find the focus of their career development. Dr. YANG’s sharing was very pertinent and sincere, benefiting the students here a lot. After the exchange and discussion with Dr. YANG, students not only got inspired to do research, but also had a deeper understanding of their future career development.
