Global logistics systems and supply chains face disruptions to operations including demand fluctuations, natural and human-caused disasters, pandemics, market prices, technology innovations, regulations, etc. Maritime container ports are susceptible to the cascading effects of these disruptions. This paper explores the use of the MuZero reinforcement learning algorithm to manage the container stacking problem, and thus increase resilience of maritime container ports to disruptions. Where previous work focused on touches per container, this paper emphasizes distance traveled per container. This approach improves accuracy for energy usage and adds insight to dynamics of container stacking. The results provide port managers with essential understanding of container handling operations, including heuristics for improving efficiency and evaluating the performance gains from changing container storage. The paper has relevance for a variety of engineering systems seeking to improve enterprise resilience to disruptions.

Global logistics systems are in an unprecedented crisis from the pandemic, workforce disruptions, supply shortages, and demand surges. Shortages of goods and services, surges of demand, and an evolving workforce call for methods that address system resilience. Maritime ports in particular are vulnerable to these disruptions. The container stacking process is especially vulnerable, as it is a bottleneck in container management. This paper presents a simulation and reinforcement learning methodology for managing container stacking blocks. The container stacking problem is known to be difficult or impossible to optimize, with most ports using black box heuristic models. A simulation and reinforcement learning approach addresses these challenges. Simulation is an effective tool for and testing different inputs, parameter changes, noise, and changes to systems due to disruptive scenarios. Reinforcement learning is especially helpful for contexts in which traditional optimization is cost and computationally prohibitive, or where data is difficult to collect and analyze. This paper applies reinforcement learning with a mathematical simulation of the container stacking problem, achieving similar results to maritime ports. The results can be used to assess performance under disruptive scenarios, as well as to test new container storage configurations.