Many research efforts on reinforcement learning (RL) is dedicated to solving challenging combinatorial optimization problems, including the container relocation problem (CRP) arising from maritime industry. In CRP, large and heavy containers are stacked on top of each other, those above the target container should be relocated elsewhere before retrieving it. This problem, which minimizes the number of relocations in the retrieval process, is NP-hard, while existing RL studies do not provide satisfactory results yet. We study the structure of CRP problem and design a new heuristic RL framework. On a set of commonly used examples, our RL demonstrates a performance almost equivalent to the most famous results.
Paper
Full text
A New Heuristic Reinforcement Learning for Container Relocation Problem
Semantic Scholar · Computer Science · 2021
Abstract
Many research efforts on reinforcement learning (RL) is dedicated to solving challenging combinatorial optimization problems, including the container relocation problem (CRP) arising from maritime industry. In CRP, large and heavy containers are stacked on top of each other, those above the target container should be relocated elsewhere before retrieving it. This problem, which minimizes the number of relocations in the retrieval process, is NP-hard, while existing RL studies do not provide satisfactory results yet. We study the structure of CRP problem and design a new heuristic RL framework. On a set of commonly used examples, our RL demonstrates a performance almost equivalent to the most famous results.