Towards realistic digital twins: calibrating metropolitan network simulations across multiple cities

This talk explores the critical role of calibration in developing accurate and reliable digital twins of metropolitan transportation networks. We address the challenge of underdetermination in large-scale stochastic simulation calibration, where limited real-world data can hinder accurate counterfactual analysis. To overcome this, we introduce sample-efficient methodologies that leverage metamodels, machine learning, and diverse emerging data sources. Specially, we will discuss the use of abundant path travel times for demand calibration. The enhancement in simulation fidelity is demonstrated through case studies of multiple metropolitan areas in the US.