A review of hybrid physics-based machine learning approaches in traffic state estimation

Abstract Traffic state estimation (TSE) plays a significant role in traffic control and operations since it can provide accurate and high-resolution traffic estimations for locations without traffic states are measured with partially observed or flawed traffic data. Several comprehensive survey papers in recent years have summarized classical physics-based and pure data-driven approaches in TSE and found that both approaches have limitations on accurately modeling traffic states. Hence, a paradigm of hybrid physics-based machine learning (ML) method has been extensively developed to overcome this problem recently. However, there is not a clear understanding of the challenges specific and research gap of hybrid physics-based ML method in TSE. In this article, we provide a comprehensive survey of existing hybrid physics-based ML methods for TSE problem. This survey leads us to uncover inherent challenges and gaps in the current state of research. The results have profound implications for evaluating the applicability of hybrid physics-based ML TSE methods and identifying future research directions.