Optimization subproblem importance analysis based on machine learning prediction in a three stage-export container scheduling

A traditional way to optimize a complex optimization problem is to divide the problem into several subproblems and solve each subproblem separately or another problem that consists of some subproblems. Despite many attempts to define and solve various optimization problems in the container terminal logistics field, how to measure the importance of each subproblem is often ignored in many studies and remains a difficult issue. Most studies directly propose methods to solve a specific subproblem after stating the importance of the specific subproblem, with or without simply considering the effect of other subproblems as input. The advancement of machine learning techniques allows a new paradigm for understanding the importance of such optimization subproblems. In this study, a scheduling problem for export containers in a terminal is considered. The case considers scheduling subproblems on subsequent processing stages on yard cranes, internal trucks, and quay cranes. With the input of each stage’s processing time information (mean and standard deviation values) and the selected scheduling rule for each stage, the makespan of all containers’ processing is predicted. The numerical experiments show that the scheduling rules for quay cranes and internal trucks have the most significant impact on system performance. These finding challenges conventional approaches by revealing that not all subproblems contribute equally to system optimization. The proposed machine learning framework enables terminal operators to adapt their optimization focus to address high-impact areas, reducing computational complexity while providing a data-driven methodology for understanding interdependencies between operational components.