Thesis Topics

The purpose of this project is to new optimization algorithms for container vessel stowage planning. This problem is too large to solve optimally. Thus, the goal of the project is to develop fast and accurate approximation algorithms. A wide range of techniques may be applied to this end including concrete heuristics, stochastic local search, large neighborhood search, column generation, branch-and-bound hybrids, symbolic representations, integer programming and hierarchical decompositions. The project is a part of an industrial collaboration with A.P. Møller-Mærsk A/S.

Liner Shipping Network Optimization

Container vessels sail on cyclic routes after a fixed schedule. A network consists of a set of such cycles each with a sufficient number of vessels to maintain a weekly schedule. Network design is a complex strategic optimization task. The prupose of the project is to develop decision support tools for network design. A wide range of optimization techniques may be applied including concrete heuristics, stochastic local search, large neighborhood search, column generation, branch-and-bound hybrids, symbolic representations, integer programming and hierarchical decompositions. The project is a part of an industrial collaboration with A.P. Møller-Mærsk A/S.

Stowage Planning using Column Generation

The prupose of this project is to participate in ongoing efforts on using a column generation combined with constraint programming for stwaoge planning.

Booking forecasting using Machine Learning

Container shipments are booked in advance. Surprisingly, however, there is no fee for breaking a booking. For that reason, it is important for shipping companies to be able to predict whether a booking is likely to be broken. The aim of the thesis is to develop data mining techniques based on machine learning that given historical data about customer behavior can assess the likelihood that a new booking order is broken.

Vessel reassignment planning

Container vessels maintain closed loop schedules just like city buses. But unlike buses, vessels are always active and often have several specific features that fit with the regions they serve. For that reason, it is challenging to reassign a vessel or take it out of service for maintenance. The purpose of the thesis is to investigate to what extend automated planning techniques can be used to address these problems. In particular, the thesis investigates which planning domains are relevant for the problems and possibly which heuristics a planning system can use in order to solve the problems efficiently.

Stowage configurations

The stowage planning problem involves assigning a given set of containers to slots on the vessel. Sometimes, however, it is interesting to let the set of containers be free parameter and determine combinations of container types that are most profitable to stow on a vessel. Such tools can be used to characterize new vessel types and figure out how well they fit to a particular string. The objective of the thesis is to develop algorithms to solve this problem

http://www.itu.dk/~rmj/data/figures/books.gif

Incremental Evaluation

Local search (LS) is an optimization technique that explores a space of complete solutions by making many small changes. For that reason, LS relies on a fast recomputation of the objective function that exploits that only a few arguments change value in each iteration. This is done by evaluating the objective function incrementally. The purpose of this project is to implement a library in C++ that supports incremental update of general mathematical functions.

Stowage configuration

It is a very complex task to assign containers to slots on a vessel. There a many rules about how containers can be arranged (e.g., due to their length, height, width, and what cargo they carry). Often a large number of containers are stored in the same area and they may all interact with each other. The purpose of the thesis is to use configuration techniques based on Binary Decision Diagrams (BDDs) to build a tool that can help human coordinators to assign containers to slots in a single bay. It is expected that a simple visual prototype is developed as a part of the thesis.