Data-driven calibration of a two-dimensional traffic flow model: Dimension reduction and optimization strategies
Abstract
We study the data-driven calibration of a two-dimensional first-order macroscopic traffic flow model, in which the movement along and across the road is described by a three-parameter flux model. The calibration is based on three real-world datasets, ranging from microscopic drone trajectories on German highways to aggregated urban sensor measurements. We recall the well-posedness of the underlying scalar conservation law and derive structural properties of the flux model. In particular, one parameter enters linearly and can be eliminated analytically, which reduces the non-linear least-squares problem from three to two dimensions. We solve the full and reduced calibration problem with two local optimization methods, Nelder-Mead and L-BFGS-B, and three global population-based methods Differential Evolution, Consensus-Based Optimization, and its deterministic variant Quasi-Monte Carlo consensus based optimization. In addition to accuracy and computational cost, we examine how sensitive the objective is to the parameters and how robust the algorithms are with respect to randomness and initialization. The reduced formulation is considerably cheaper and less sensitive to the choice of optimizer. The local methods are the most efficient when suitably initialized, whereas the global methods are more robust.
Related
- Dimension reduction for characterizing sexual dimorphism in biomechanics of the temporomandibular joint
- A Path-Dependent Agent-Based Microsimulation of Crime and Violence Reduction Policy Portfolios in Bolivia: Survey Calibration, Adaptive Emulator Ensembles, Global Sensitivity, Tempered MCMC, and Multi-Objective Policy Analysis
- Interpretable intrinsic dimension estimation through componentwise calibration of distance and angle
- Robust conditional dimension reduction for dissimilarity data
- Algebraic Dimension and Reduction of LVMB manifolds
BibTeX
@misc{gottlich2026datadriven,
title = {{Data-driven calibration of a two-dimensional traffic flow model: Dimension reduction and optimization strategies}},
author = {Simone Göttlich and Jacob Heieck},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.29557},
eprint = {2610.29557},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.29557},
}
RIS
TY - UNPB
TI - Data-driven calibration of a two-dimensional traffic flow model: Dimension reduction and optimization strategies
AU - Simone Göttlich
AU - Jacob Heieck
PY - 2026
DA - 2026/10/21
PB - arXiv
DO - 10.48550/arxiv.2610.29557
UR - https://arxiv.org/abs/2610.29557
AB - We study the data-driven calibration of a two-dimensional first-order macroscopic traffic flow model, in which the movement along and across the road is described by a three-parameter flux model. The calibration is based on three real-world datasets, ranging from microscopic drone trajectories on German highways to aggregated urban sensor measurements. We recall the well-posedness of the underlying scalar conservation law and derive structural properties of the flux model. In particular, one parameter enters linearly and can be eliminated analytically, which reduces the non-linear least-squares problem from three to two dimensions. We solve the full and reduced calibration problem with two local optimization methods, Nelder-Mead and L-BFGS-B, and three global population-based methods Differential Evolution, Consensus-Based Optimization, and its deterministic variant Quasi-Monte Carlo consensus based optimization. In addition to accuracy and computational cost, we examine how sensitive the objective is to the parameters and how robust the algorithms are with respect to randomness and initialization. The reduced formulation is considerably cheaper and less sensitive to the choice of optimizer. The local methods are the most efficient when suitably initialized, whereas the global methods are more robust.
ER -