Markovian Approach to tanker Voyages

Specialeforsvar ved Trine Frydendahl Nørgaard

Titel: Markovian Approach to tanker Voyages

Abstract: A major challenge of tanker shipping companies is to optimize voyages in the long run. Freight rates and the availability of cargoes to be delivered differ across the globe. The problem is subject to a wide variety of research, but previous papers have not yet reached an agreement of an optimal approach to the problem. This thesis investigates if tanker voyages can be modelled as a Markov Decision Process. The thesis is made in
collaboration with the Danish shipping company Maersk Tankers, and addresses the challenge from their point of view. The first step of this thesis was to derive necessary knowledge of vessels' historic travel patterns. This was done by developing a method that estimates port stops from Automated Identification System (AIS) data. AIS is installed on all tanker vessels and transmits essential data on the ships' position and status. However, AIS data does not explicitly report when a vessel is in port. By developing a method to estimate port stops from the available AIS data fields, it was possible to base this thesis on voyage data from more than a thousand vessels over a time period from 2014 to 2016. A chi square test was made to investigate the Markov Property of tanker vessels, which showed no reason to deny tanker voyages this property. A Markov Decision Process was developed, which showed encouraging results. It was found that long run total expected profit of following the derived decision process was significantly higher than the total expected profit associated with simply planning voyages based on maximizing the immediate profit of each voyage. These results however, are based on simulations of a simplified setup, and the real world effectiveness of the model is left for investigation in
future work. This thesis highlights the possible advantages Maersk Tankers can
obtain by considering voyages as a Markov Decision Process

Vejleder: Rolf Poulsen
Censor: Niklas Kohl