Sustainable mobility transitions require a comprehensive understanding of transport dynamics. To this end, Activity-Based Models (ABMs) provide a foundational micro-simulation framework capable of representing complex, person-level activity-travel patterns. With thousands of parameters and a lack of closed-form structure, ABMs present significant calibration and scalability challenges, including high-dimensional complexity, conflicting objectives, and prohibitive computational costs. While current Bayesian Optimization (BO) methods address these issues through dimension reduction, they often sacrifice accuracy by relying on restrictive sparsity assumptions. Furthermore, existing approaches frequently under-utilize transportation domain knowledge and modular structure, limiting their robustness and scalability. This study introduces a novel dimensionality reduction scheme for BO that is further augmented by Large Language Model (LLM) assistance to prioritize influential variables based on their functional roles. The optimization is further refined by an entropy-based acquisition function designed to mitigate output saturation from extreme inputs. By exploiting the inherent modularity of ABMs, the framework implements a sequential calibration workflow tailored specifically for multi-modal transportation models. Experiments show that the proposed method achieves a lower evaluation cost and higher calibration accuracy than state-of-the-art alternatives, while improving computational scalability for modular large-scale ABMs. The calibrated ABM outputs provide a rich baseline for policy evaluation, transport operations, and downstream sustainability applications.
Le et al. (Thu,) studied this question.