ITU Research Talks

Bring your lunch and join fellow faculty across ITU for this months research lunchtime talk. January's host is the Data Science section.

Speaker: Jaike Van Twiller; Postdoc.
Title: Navigating Demand Uncertainty in Container Shipping: Deep Reinforcement Learning for Enabling Adaptive and Feasible Master Stowage Planning.

Abstract:

Uncertainty is a central challenge in container shipping, as deviations from the expected flows of cargo cause delays, extra costs and unnecessary emissions. In stowage plans, containers are placed in ship locations to maximize profit, while meeting capacity, stability and safety constraints. Although reinforcement learning (RL) is promising for such challenging tasks in planning under uncertainty, RL often struggles to satisfy strict constraints.

In this work, we develop a deep RL approach that enforces strict constraints by projecting actions into a valid region. The model integrates information about the problem instance, the evolving plan, and demand uncertainty to guide planning. In the experiments, our model outperforms leading baselines, adapts to varying uncertainty, and scales to larger problems. In doing so, our approach supports stowage planning under uncertainty, contributing to more reliable and sustainable container shipping operations. Read more