Towards a Deep Reinforcement Learning Model of Master Bay Stowage Planning

Major liner shipping companies aim to solve the stowage planning problem by optimally allocating containers to vessel locations during a multi-port voyage. Due to a large variety of combinatorial aspects, a scalable algorithm to solve a representative problem is yet to be found. This paper will show that deep reinforcement learning can optimize a non-trivial master bay planning problem. Our experiments show that proximal policy optimization efficiently finds reasonable solutions, serving as preliminary evidence of the potential value of deep reinforcement learning in stowage planning. In future work, we will extend our architecture to address a full-featured master bay planning problem.