Resource leveling for LNG carrier production with deep reinforcement learning

Abstract

We address the resource leveling problem in the Liquefied Natural Gas (LNG) carrier production environment. In the shipbuilding industry, workload balancing is important as manpower fluctuations cause additional costs such as hiring, firing, and outsourcing costs. In this paper, we propose a two-step leveling approach for workload balancing that can be applied within a reasonable time. The proposed method first reschedules the given schedule with tank unit movements using reinforcement learning and then performs an iterative greedy algorithm. We apply the proposed two-step leveling in real-world data to minimize the fluctuation of the workload in the LNG carrier production environment. Experimental results on real data showed that our algorithm performed better than other algorithms and was able to lower the variance of the daily workload by more than 30% compared to the initial variance.