A path-aware Bayesian optimization (PABO) framework for cost-efficient sampling site exploration

초록

This study proposes a path-aware Bayesian optimization (PABO) framework for cost-efficient optimization in applications where both sampling and movement costs are significant. Conventional Bayesian optimization (BO) methods primarily focus on reducing the number of function evaluations, while recent cost-aware and distanceaware approaches consider either sampling cost or movement cost separately. However, these approaches do not explicitly address the trade-off between sampling effort and travel effort within a unified framework. To overcome this limitation, a unified cost formulation is proposed to jointly account for sampling and movement costs. Based on this formulation, the proposed framework integrates an adaptive directional sampling region (ADSR) and a path-aware acquisition function. The ADSR promotes spatially coherent exploration by adaptively restricting the search space according to geometric distance and model uncertainty, while the acquisition function balances expected improvement and operational cost to improve overall exploration efficiency. The effectiveness of the proposed framework is demonstrated through benchmark problems, sensitivity analyses, and parametric studies under various cost configurations, which consistently show that PABO reduces total cost compared with conventional BO methods. Furthermore, the applicability of the proposed framework is demonstrated using geological datasets in the context of marine geological exploration problems. In these realistic exploration scenarios, PABO achieves statistically significant improvements in total cost, number of added samples, and total travel distance. The results demonstrate that jointly considering sampling cost, movement cost, and path consistency within a unified framework can improve cost efficiency in practical spatial optimization problems.

키워드