Dynamic traffic assignment is an increasingly practically useful tool for traffic analysis and evaluation. How to understand the data needs and further calibrate dynamic demand and supply elements for large-scale traffic simulators is a challenging issue from both theoretical and practical aspects. In current practice, the time-dependent demand matrix is typically generated by combing the peak-period demand from regional planning model with a temporal departure time profile. Before any scenario testing, it is important and necessary to use observed data, such as probe point-to-end travel time, sensor flow and speed data, for constructing time-dependent simulation representations consistent with real-world conditions.
In this study, we focus on a number of important model calibration issues. First, how to identify critical bottlenecks from limited sensor data, and how to jointly calibrate demand and capacity elements of key traffic bottlenecks with multiple sources of observation data (such as link speed or flow counts). We are also particularly interested in developing guidelines for adjusting temporal and spatial capacity for various types of freeway and arterial bottlenecks, such as merges, diverges and intersections with complex geometry features. In addition, we will discuss how to select and refine different traffic flow models, for example, point queue, spatial queue and kinematic wave models, in order to better represent route choice behavior in a heavily congested network. Based on a number of real-world test cases, we will discuss our system calibration and validation experiences for using open-source mesoscopic DTA packages in several cities in the United States and other international countries.