Optimization of berth allocation with learning working practices

Abstract

In the dynamic landscape of global trade, maritime transportation stands out as the primary mode, facilitating more than 80% of international trade. The efficiency of maritime trade hinges on the optimization of port operations, particularly the al- location of berths. This study explores the complex challenge of the Berth Alloca- tion Problem (BAP) by proposing an innovative methodology that combines machine learning, specifically XGBoost, with optimization modeling. Traditional approaches focused solely on minimizing the weighted sum of vessel waiting times, overlooking the preferences of port authorities. In contrast, our model not only addresses the pri- mary goal of minimizing delays but also integrates the unique preferences and work- ing practices of port authorities. Through rigorous experimentation on twenty-two real datasets, we assess the model’s performance, benchmarking it against heuristic warm start optimization and optimization-centric models. The results demonstrate the model’s ability to simultaneously accommodate port authority preferences and opti- mize vessel waiting times, overcoming the limitations of conventional methods. In the future, this research lays the groundwork for investigating the model’s flexibility to address various requirements at ports and opens up the potential for more advanced studies by potentially incorporating crane assignment considerations.