Developing a Container Ship Loading-Planning Program Using Reinforcement Learning

This study presents an optimized container-stowage plan using reinforcement learning to tackle the complex logistical challenges in maritime shipping. Traditional stowage-planning methods often rely on manual processes that account for factors like container weight, unloading order, and balance, which results in significant time and resource consumption. To address these inefficiencies, we developed a two-phase stowage plan: Phase 1 involves bay selection using a Proximal Policy Optimization (PPO) algorithm, while Phase 2 focuses on row and tier placement. The proposed model was evaluated against traditional methods, demonstrating that the PPO algorithm provides more efficient loading plans with faster convergence compared to Deep Q-Learning (DQN). Additionally, the model successfully minimized rehandling and maintained an even distribution of weight across the vessel, ensuring operational safety and stability. This approach shows great potential for enhancing stowage efficiency and can be applied to real-world shipping scenarios, improving productivity. Future work will aim to incorporate additional factors, such as container size, type, and cargo fragility, to further improve the robustness and adaptability of the stowage-planning system. By integrating these additional considerations, the system will become even more capable of handling the complexities of modern maritime logistics.