Learning Operational Goals for Propulsion System Using Reinforcement Learning

Abstract: This degree project, conducted at ABB, aims to analyze and solve different situations that a crew on board a vessel might face by controlling its propulsion system. The propulsion system is viewed as static, transition-deterministic, as well as stochastic when measuring data. This system is then used to formulate a decision problem using a finite Markov Decision Process, which is attemped to be tackled using Q-learning, Speedy Q-learning and Double Q-learning for three different objectives that are relevant to the system's behaviour and performance. The objectives policies found from experiments are clearly working as intended and from the looks of it they seem to be near optimal, knowing that there is a proof of convergence for Q-learning based algorithms. The convergence rate for the three different algorithms are then compared to a solution that is seen as optimal, to see how fast they converge and try to determine the time needed to solve problems similar to the ones stated in this thesis.
 

Examiner: Prof. Cristian Rojas