Trajectory Investigation for Enhanced Calibration of Microsimulation Models

FHWA-HRT-21-071

Trajectory Investigation for Enhanced Calibration of Microsimulation Models

David K. Hale (0000-0001-5486-9367), Xiaopeng Li (0000-0002-5264-3775), Amir Ghiasi (0000-0002-0986-9840), Dongfang Zhao (0000-0002-5424-9915), Farnoush Khalighi (0000-0003-3353-5194), Murat Aycin (0000-0002-5798-3381), Rachel James (0000-0001-9138-510X)

Leidos, Inc.
11251 Roger Bacon Drive
Reston, VA 20190

Office of Operations Research and Development
Federal Highway Administration
6300 Georgetown Pike
McLean, VA 22101

The Federal Task Manager was Rachel James (HRDO-20; ORCID: 0000-0001-9138-510X).

Traffic engineers and researchers calibrate microsimulation models using macroscopic inputs—such as aggregated traffic throughput—instead of microscopic inputs, such as intervehicle spacing and acceleration. This has led to concerns that these models have been capturing the microscopic driver behaviors inaccurately, despite the macroscopic performance measures' apparent goodness of fit. Given the recent improvements to data collection and data processing technologies, particularly concerning drone or unmanned aerial vehicle technologies and cost reductions, there is renewed interest in trajectory-based calibration for microsimulation models. Researchers behind this project developed a new methodology for trajectory-based calibration, and they tested this methodology against traditional calibration at real-world urban freeway locations: I–270 in Maryland; I–15 in California; I–75 in Florida; and I–95 in Virginia. The results provided evidence that traditional calibration indeed cannot be trusted to produce realistic vehicle trajectories. Moreover, explicit integration of trajectories into the calibration process can remedy this. Calibrated model results were most impressive at I–75, which is the only site where trajectories were collected by a helicopter (instead of by drones), producing 1.2-mi-long trajectories. This report and accompanying software scripts provide instructions and lessons learned for collecting, cleaning, post-processing, correcting, and validating trajectory data. The report and scripts also provide instructions and lessons learned for trajectory-based calibration and validation. Future applications of the proposed methodology may involve studying the importance of car-following versus lane-changing, calibrating separate driver models for different congestion regimes, and calibrating the trajectories of connected and automated vehicles.

Vehicle trajectory, NGSIM, microsimulation, calibration, traffic simulation

Unclassified