Freshwater Filling Optimization Based on Price Using XGBoost and Particle Swarm Optimization on Cargo Ship Voyage

Abstract

Purpose: Efficient freshwater management is critical in cargo ship operations, yet current practices often involve fixed refilling strategies that ignore price differences across ports and fail to predict actual consumption accurately. These inefficiencies lead to unnecessary operational costs. To address this, the study introduces a combined approach using XGBoost for predict freshwater usage and Particle Swarm Optimization (PSO) to minimize refilling costs through optimal port selection.

Methods: Freshwater demand was predicted using an XGBoost regression model trained on real operational data from 2024, which included historical voyage distances and freshwater consumption records from cargo ships. Based on these predictions, Particle Swarm Optimization (PSO) was applied to identify cost-efficient refilling locations along each ship’s route, minimizing total water procurement cost while satisfying operational constraints. The proposed framework was validated through simulated voyage scenarios to evaluate its impact on cost efficiency and planning effectiveness.

Result: The integration of XGBoost and PSO effectively optimized freshwater refilling strategies, achieving a relative prediction error of 9.48% in freshwater consumption prediction and cost savings from 9 to 40% from across 3 ships sample through strategic port selection based on consumption patterns and price variability.

Novelty: Unlike prior works focused on fuel or generic logistics optimization, aim of this study is to combine XGBoost and PSO for optimizing freshwater refilling on cargo ship voyages using actual operational data. The results demonstrate practical, scalable improvements in cost efficiency, making a novel contribution to maritime resource planning.