Surrogate-modeling for Pavement Maintenance Scheduling

Abstract Incorporating traffic dynamics into pavement maintenance scheduling improves decision quality by accounting for user delays, but it introduces significant computational challenges. Existing approaches rely on detailed macroscopic traffic flow models to evaluate the impact of maintenance activities on user. These models are typically embedded within population-based optimization algorithms to explore complex decision spaces. However, repeated evaluations during optimization result in high computational costs, making large-scale applications impractical. This paper investigates the use of surrogate modeling techniques—including statistical and machine learning models—to approximate traffic flow responses in place of direct simulation. These surrogate models are trained on data that reflects critical traffic dynamics to capture key traffic relationships and are integrated into the maintenance planning framework to support optimization. In addition to evaluating accuracy, computational efficiency, and the quality of resulting maintenance policies, the study also investigates the generalizability of the surrogate models across different network configurations. Results demonstrate that the proposed method significantly reduces computational burden while maintaining reliable and transferable performance in traffic-aware pavement maintenance planning. The paper concludes with practical guidelines for researchers and practitioners seeking to represent realistic user costs without relying on computationally intensive traffic evaluations.