SCI

[배혜림] Synchronous Deep Reinforcement Learning for Optimized Storage Assignment of Ship Blocks

영문초록: This paper presents a deep reinforcement learning (RL) framework for optimizing block storage allocation in
shipbuilding yards. During the shipbuilding process, vessels are constructed in block units to maximize the operational
efficiency. Following assembly, these blocks must be stored in limited yard spaces, creating a complex variant of the
binary packing problem. This storage allocation problem is further complicated by operational constraints, including
the barge capacity and transportation time restrictions. Moreover, poor storage decisions can lead to redundant block
movements, which can adversely affect downstream processes and increase operational costs. To resolve this problem, a
policy-gradient-based synchronous RL model was developed and an embedding layer was introduced to improve its
inference performance. The proposed model was validated using both synthetic data and real-world stockyard data.
The synthetic dataset was used to prove the proposed model's efficacy and evaluate the hyper parameter value. The
real-world data with operational constraints derived from actual field conditions is used for systematic comparisons
with mathematical optimization. The comparison demonstrated that the proposed RL method effectively generated
solutions that satisfied practical operational constraints, offering a viable alternative to traditional mathematical
methods.