Artificial Intelligence for Automatic Container Stowage Planning Optimisation

Container ships visit many ports along a route, and transport large numbers of containers of varying dimensions and contents. The unloading of containers destined for each port, and the loading of containers destined for subsequent ports, is arranged to follow pre-determined stowage arrangements and to make efficient use of cranes. These arrangements are termed stowage plans, and the problem of determining the best arrangement of containers within a container ship, on leaving a given port, is called the container-ship stowage problem.
Most published work on stowage planning has focussed on semi-automation and tools used to assist stowage planners. However, full automation of the production of solutions to the stowage problem has received some attention, with a recent growth of research in the area. Approaches to full automation have used techniques as varied as the comparison of plausible plans by simulations of voyages, meta-heuristic search, and (most commonly) the application of linear programming
techniques to a mathematical model of the cargo-space. However, many published approaches have relied on simplifications of the problem domain which have rendered them of no commercial value.
This paper explains and compares a range of approaches taken by the authors to the solution of the container-ship stowage problem, which focuses on the reliable production of valid, sub-optimal
solutions. These include the strong decomposition of the problem into different conceptual levels of planning to which branch & bound, Tabu search techniques and Genetic algorithms have been
applied. A particular focus of the paper is the relationships between the methods of solution and the corresponding models of cargo and stowage spaces, and the consequences that these models have for the accuracy and usefulness of the solutions produced.