| 제목 | MuZero 강화학습을 이용한 항만 컨테이너 재정돈 계획 |
|---|---|
| Title | Container Pre-Marshalling with MuZero Reinforcement Learning Algorithm |
| 저자 |
최원균 (한국항공대학교 항공교통물류학과) 이기주 (한국항공대학교 항공교통물류학과) 채준재* (한국항공대학교 항공교통물류학과) |
| Author |
Wongyun Choi(School of Air Transport, Transportation and Logistics, Korea Aerospace University) Keyju Lee(School of Air Transport, Transportation and Logistics, Korea Aerospace University) Junjae Chae†(School of Air Transport, Transportation and Logistics, Korea Aerospace University) |
| Bibliography | Journal of Logistics Science & Technology, 2(2),19~37, 2021, |
| DOI | |
| Key Words | Pre-Marshalling, Container Reshuffling, Reinforcement Learning, MuZero |
| Abstract | This study provides a new solution approach for container pre-marshalling problems using a reinforcement learning method, the MuZero Algorithm. We have developed a customized pre-marshalling environment for an agent to be trained. To facilitate the training, we have devised i) some action masking methods, and ii) a reward function where lower bound for the number of rehandling is considered. Action masking methods and heuristically found lower bound were proven to be helpful in the learning process. Experiments with different sizes (from the minimum number of 8 and the maximum number of 12 containers) were carried out, within a stacking space of six rows and six tiers. Experimental results show that the MuZero algorithm implemented with our strategy is capable of training to solve the pre-marshalling problems for some small sized problems. |
| PDF download | JM_2[2]-P19~37.pdf |