Influencia del Control de Hinchamiento en Reglas de Reubicación Diseñadas Automáticamente mediante Programación...

The container relocation problem (CRP) is a critical optimisation problem in maritime port operations, in which efficient container handling is essential for maximising terminal throughput. Relocation rules (RRs) are a widely adopted solution approach for the CRP, particularly in online and dynamic environments, as they enable fast, rule-based decision-making. However, the manual design of effective relocation rules is both time-consuming and highly dependent on problem-specific characteristics. To overcome this limitation, genetic programming (GP), a bio-inspired optimisation technique grounded in the principles of natural evolution, has been employed to automatically generate RRs. By emulating evolutionary processes such as selection, recombination, and mutation, GP can explore large heuristic search spaces and often produces rules that outperform manually designed alternatives. Despite these advantages and their inherently white-box nature, GP-generated relocation rules frequently exhibit excessive complexity, which hinders their interpretability and limits insight into the underlying decision logic. Motivated by the biomimetic observation that evolutionary systems tend to favour compact and efficient structures, this study investigates two mechanisms for controlling rule complexity, parsimony pressure, and solution pruning, and it analyses their effects on both the quality and size of relocation rules evolved by GP. The results demonstrate that substantial reductions in rule size can be achieved with only minor degradation in performance, measured as the number of relocated containers, highlighting a favourable trade-off between heuristic simplicity and solution quality. This enables the derivation of simpler and more interpretable heuristics while maintaining competitive performance, which is particularly valuable in operational settings where human planners must understand, trust, and potentially adjust automated decision rules.