Adaptive ship stabilization: Control of ship stabilizing actuators using modified deep model reference adaptive control

Abstract

For common control methods, there are often an extensive modeling and tuning procedure needed to obtain a good control performance. This thesis investigates a way of omitting that procedure by utilizing a machine learning based control method. An adaptive method is presented of how to control the stabilizing actuators of a ship with limited or no information about the ship’s dynamics, and where the core of the adaptiveness lies in the combination of a feature vector, extracted from an artificial neural network (ANN), and a weight vector. The main objective was to reduce the most critical ship movement, rolling motions, using active fin stabilizers. The proposed adaptive controller showed promising results that increased the performance of the ordinary linear controller, while also being able to adapt when changes in the plant were introduced. The second objective was to see if the proposed controller was capable of controlling multiple kinds of actuators. The results indicate that this is the case, however, some additional modifications need to be made to achieve the same performance as in the first case when using only fins.