Enhanced metamodeling strategy for uncertainty quantification and reliability verification in heterogeneous connected and automated vehicle platoon control models

Vehicle platooning can improve traffic safety, efficiency, and fuel consumption by coordinating vehicles to travel closely while maintaining safe inter-vehicle distances. The performance of a platoon depends critically on the controller model, and uncertainties can degrade control effectiveness, increasing collision risk. This paper proposes a novel methodology to verify the reliability of heterogeneous vehicle platoons under parametric uncertainties, focusing on spacing errors. The approach combines a Multiple Input Single Output (MISO) Nonlinear Autoregressive Exogenous (NARX) model with a Kriging interpolator, linking uncertain parameters to the nonlinear dynamic response of each vehicle. Two strategies are presented: one prioritizes predictive accuracy for platoons of up to five vehicles, while the other balances accuracy and computational cost for larger platoons. The metamodel structure is optimized to select the most significant regressors, maintaining high predictive performance. Validation is performed using Monte Carlo simulations, and comparisons with classical KNARX and PC-NARX models demonstrate improved scalability, efficiency, and reliability. While the current study assumes ideal PLF topology, the framework can accommodate other topologies and information delays. Furthermore, the methodology can integrate sparse sensor data, enabling potential digital twin implementations, and future work may extend it to non-parametric uncertainties such as varying road conditions or traffic disturbances.