Solving Container Pre-Marshalling Problem using Monte-Carlo Tree Search and Deep Neural NetworkLiu, Jianwei (2023) Solving Container Pre-Marshalling Problem using Monte-Carlo Tree Search and Deep Neural Network. Master's Thesis, University of Pittsburgh. (Unpublished) This is the latest version of this item.
AbstractThe container pre-marshaling problem (CPMP) is a significant challenge in container terminal operations, aiming to optimize the relocation of containers to improve efficiency. Despite extensive research on exact and heuristic methods for the container relocation problem (CRP) and CPMP, reinforcement learning (RL) remains underexplored in the literature. This thesis proposes a Monte-Carlo Tree Search (MCTS) method combined with the Lowest Priority First Heuristic (LPFH) to solve the CPMP efficiently. The MCTS method incorporates the LPFH heuristic to achieve consistent simulation results and minimize the number of movements needed for container relocation. Our approach achieves near state-of-the-art results in several instances with acceptable inference speed. Additionally, this thesis introduces a machine learning-based method to estimate the number of relocations required for a given CPMP configuration. We train a deep learning model with both convolutional neural network (CNN) and multi-layer perceptron (MLP) architectures on self-generated data, identifying important features to achieve over 90% classification accuracy in small instances. The proposed approach has the potential to provide more efficient and effective solutions to the CPMP than traditional optimization methods. Overall, this thesis contributes to the CPMP literature by introducing novel methods for solving the problem and providing valuable insights into the potential of machine learning and RL for solving complex optimization problems. Share
Details
Available Versions of this Item
MetricsMonthly Views for the past 3 yearsPlum AnalyticsActions (login required)
|
![[feed]](/images/feed-icon-32x32.png){kind=icon}