Robust yard space allocation based on adversarial reinforcement learning considering vessel arrival time uncertainty

In an era marked by volatile global trade patterns and increasing climate-related disruptions, operational resilience has become a critical determinant of port competitiveness. While optimized container stacking and retrieval strategies are recognized as vital levers for enhancing port resilience, their effectiveness is severely compromised by pervasive uncertainties in vessel arrival times (VAT) - a challenge amplified by complex transshipment networks and cascading delay effects. This study addresses the robust container stacking and retrieving (R-CSR) problem under VAT uncertainty through an integrated optimization and learning framework. A min-max robust optimization model is developed that simultaneously optimizes storage allocation and retrieval sequencing to minimize worst-case container travel distances across all plausible VAT scenarios. Recognizing the computational complexity of solving this problem in dynamic environments, we propose a novel robust adversarial reinforcement learning (R-ARL) algorithm which features a protagonist-adversary architecture. Numerical experiments demonstrate that the R-CSR model and algorithm are effective for test cases of different scales.