Vehicle Trajectory-Based Calibration Procedure for Microsimulation

Abstract

In practice, traffic engineers and researchers calibrate microscopic driver behavior in microsimulation models using macroscopic inputs (e.g., aggregated traffic throughput) instead of microscopic inputs (e.g., inter-vehicle spacing and acceleration). This mismatch has naturally led to concerns that these models have not accurately captured the microscopic driver behaviors, despite the apparent goodness-of-fit in the macroscopic performance measures. There is renewed interest in trajectory-based calibration for microsimulation models given the recent improvements in data collection and data processing technologies. Toward this end, this project collected aerial vehicle trajectories using cameras attached to drones at three real-world sites (producing 800 ft trajectories) and a helicopter at one site (producing 1.2 mi trajectories). This study used these datasets to develop a new microsimulation model trajectory-based calibration method and calibrated four urban freeway models: I-270, I-15, I-75, and I-95. Next, this study calibrated models of the same four sites using macroscopic data (e.g., average segment speed and throughout). Finally, this study compared the two methods’ calibration accuracy to understand the benefits of the trajectory-based calibration method. The experimental results provided evidence that analysts should not trust traditional calibration methods to produce models with realistic vehicle trajectories. However, explicit integration of trajectories into the calibration process (i.e., hybrid calibration methods) can significantly improve the modeled trajectories’ realism. Calibrated model results were most impressive at I-75, which is the only site that collected data via helicopter, yielding significantly longer trajectories.