Automatic Calibration of Mesoscopic Traffic Simulation Using Vehicle Trajectory Data
Ran Sun, Zihao Wang, Xingmin Wang, Henry X. Liu
TL;DR
This work tackles the challenge of calibrating large-scale mesoscopic traffic simulations using vehicle trajectory data. It introduces an automatic, trajectory-based framework that jointly calibrates OD demand (including route choices) and supply parameters (capacity and driving behavior) by combining a network-flow quadratic program for OD/path/link estimation with a stochastic-approximation SPSA-based simulator calibration, aided by demand and path clustering. The Birmingham, MI case study demonstrates the method's ability to achieve full network throughput with substantially lower travel-time error than baselines, validating the approach's efficiency and accuracy. The framework offers a scalable, data-driven path to automate calibration for operational planning and ITS deployment, even when trajectory data penetrations are limited.
Abstract
Traffic simulation models have long been popular in modern traffic planning and operation applications. Efficient calibration of simulation models is usually a crucial step in a simulation study. However, traditional calibration procedures are often resource-intensive and time-consuming, limiting the broader adoption of simulation models. In this study, a vehicle trajectory-based automatic calibration framework for mesoscopic traffic simulation is proposed. The framework incorporates behavior models from both the demand and the supply sides of a traffic network. An optimization-based network flow estimation model is designed for demand and route choice calibration. Dimensionality reduction techniques are incorporated to define the zoning system and the path choice set. A stochastic approximation model is established for capacity and driving behavior parameter calibration. The applicability and performance of the calibration framework are demonstrated through a case study for the City of Birmingham network in Michigan.