Capturing traffic state variation process: An analytical modeling approach

Precise and dependable identification of traffic states is crucial for optimizing traffic system, which forms the foundation for mitigating congestion and enhancing the overall efficiency and stability of traffic operations. Existing research has mainly adopted methods such as signal processing methods and traffic fundamental diagrams, but each has its own shortcomings. Therefore, this study introduces a Bayesian online changepoint detection method, which can dynamically detect changepoints in traffic flow observation sequences to explain the progression of traffic state variation including traffic flow breakdown. This method is more flexible and adaptable compared to traditional methods. We use this method for empirical analysis. Moreover, this study proposes an adaptive multi-state traffic fundamental diagram model to identify changes in traffic states based on a modified s-shaped three-parameter (S3) fundamental diagram. Our proposed traffic state identification approach is highly interpretable, and can be used to capture traffic state features consisted of at most five different states with four density reference values. We use the method for theoretical analysis. Furthermore, this study applies the above two methods to a high-resolution vehicle trajectory dataset, achieving a comprehensive analysis of the process of variations in traffic state. The findings indicate a strong alignment in the traffic states detected by both techniques, thereby validating the enhanced efficacy of our methodologies for the recognition and analysis of traffic flow dynamics. By comparing the findings of field data analysis and theoretical analysis, a deeper understanding of the traffic state dynamics is achieved, which encompasses the transition from a free-flow condition to a congested one, along with the features of various traffic states such as the stable, metastable, and unstable phases.