Verbetering van de Kalibratie van Verkeersmicrosimulatiemodellen: de Invloed van de Doelfunctie en de Integratie van Rijstrookwisseldynamiek

This dissertation advances the methodological foundations of traffic microsimulation model calibration, addressing long-standing issues of robustness, noise sensitivity, and behavioral realism. While microsimulation is central to modern transport analysis, its predictive reliability critically depends on how well behavioral parameters are calibrated. Despite decades of research, several gaps remain unresolved: inconsistent understanding of parameter sensitivity, limited exploration of objective function design, and the underrepresentation of lane-changing dynamics. This thesis systematically tackles these gaps through a combination of sensitivity analysis, objective function development, and multi-faceted calibration experiments. The first part develops a comprehensive framework, based on the Morris screening method, to evaluate parameter sensitivity under uncertainty across free-flow and congested regimes. Unlike most prior studies, it explicitly accounts for stochastic simulation noise through multiple random-seed replications and statistical filtering, enabling more reliable estimation of parameter importance. Results show that parameter influence is context-dependent—varying across traffic states, locations, and performance indicators—highlighting that sensitivity outcomes cannot be generalized. Moreover, findings suggest that full-set calibration, where all relevant parameters are optimized simultaneously, can effectively identify influential parameters without requiring prior dimensionality reduction. The second part investigates the design of calibration objective functions and their impact on optimization behavior. Several dissimilarity metrics are compared, revealing that classical measures such as the root-mean-square relative error and Kolmogorov-Smirnov statistic are often sensitive to outliers and limited in representing distributional differences. In contrast, the Wasserstein distance, grounded in optimal transport theory, provides a more robust and informative comparison by considering the entire shape of empirical distributions. Calibration experiments demonstrate that using the Wasserstein distance leads to smoother convergence and more stable parameters. Furthermore, combining multiple key performance indicators—such as speed and headway—within a unified objective function enhances both behavioral representativeness and model consistency. The third part extends the calibration framework to include lateral driving behavior, a critical yet often neglected aspect of microsimulation. Lane-changing dynamics, particularly in weaving and merging areas, are captured using new objective function components based on lane-change probabilities and temporal distributions between detection stations and origin-destination pairs. This enables simultaneous calibration of car-following and lane-changing parameters, capturing their interdependencies. Results show that integrating lane-change facets with longitudinal dynamics yields substantially more realistic and consistent driver behavior, underscoring the need for a holistic calibration approach. All methodological developments are demonstrated on a 2.7-km motorway weaving section of the Antwerp ring road (R1) using the PTV VISSIM microsimulation platform. Through systematic sensitivity analysis and calibration of Wiedemann 99 and lane-changing models' parameters under various objective function designs, the results consistently show that incorporating multiple behavioral facets leads to improved model reliability and behavioral realism. In conclusion, this research provides a structured and behaviorally informed calibration methodology that unifies sensitivity analysis with advanced objective function design. The framework enables modelers to achieve more robust, consistent, and realistic simulations across diverse traffic regimes, marking a methodological shift from merely fitting numerical outputs toward reproducing the behavioral logic underlying real-world traffic phenomena.