An Artificial Neural Network-Based Method for the Container Relocation Problem

• Main Authors: Wen-Fu Yang, 楊文富
• Other Authors: Yat-Wah Wan
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/rvknha

碩士 === 國立東華大學 === 運籌管理研究所 === 107 === Container transportation have become more important in modern world because goods in containers are more valuable than other means of maritime transportation. Thus, people hope to effectively transport containers. Container terminals help transship containers between vessels and trucks. The Container Relocation Problem (CRP) is an issue related to the improvement of container terminals. There have been methods proposed for the CRP. Here, we first choose two different heuristics, Look-ahead N and Min-Max for the CRP, and apply Artificial Neural Network (ANN) to imitate how these two heuristics reshuffle containers; then by learning from the best of the two heuristics we check whether the performance of ANN can surpass them. We do experiments on two types of bay size: 4-row, 3-column, and 7-container bay size (small bay) and 4-row, 6-row, and 18-container bay size (large bay). Besides following the logic of two heuristics to generate datasets, we form a new type of datasets by combining best data instances of two heuristics. We train many ANNs for Min-Max, Look-ahead N and Best-of-Two to set their parameter values. Then we use the trained parameters from different ANNs to reshuffle containers and compare the results with the original reshuffle results of heuristics. ANN perfectly imitates the two heuristics and surpasses them in combined datasets that we generate for small bay size. For large bay size, ANN is unable to imitate nor surpass the two heuristics but the results are very close to them. In the end, we do further analysis on two methods to reduce computational time of training ANNs.