Content area
The maritime sector is currently undergoing a monumental energy transition. The international maritime organization has set a goal for reducing ship-borne anthropogenic greenhouse gas emissions by 50% by 2050 compared to the emission levels in 2008. To reach this goal, significant advancements must be made in the way the existing tonnage is operated, new energy efficient technologies must be adopted and old technologies need to be enhanced. In this dissertation, state-of-the-art methodologies for ship energy system unit commitment and speed profile optimization are studied and developed.
Efficient unit commitment of a ship's energy system is crucially important to minimize operational costs and greenhouse gas emissions. However, developing an effective commitment logic is difficult due to the complex nature of such systems. This leads to the primary research problem of this dissertation: "How to achieve real-time and optimal unit commitment control a ship's energy system of arbitrary complexity?".
Novel methods based on machine learning and mathematical optimization are proposed for this task. Both of these methods are developed and their performance tested based on real operational ship data. The method based on machine learning leveraged reinforcement learning to produce a unit commitment method that learned to control the energy system by itself. The benefit of such an approach is the dynamic nature of learning, where the controller can automatically evolve in accordance to a changing environment. In addition, the trial-and-error based learning phase can reveal surprisingly effective control sequences unattainable by other methods.
The method based on mathematical optimization consisted of two parts: prediction of future power demand via a Gaussian process, and a mixed-integer linear program for optimizing the unit commitment response of the energy system based on the demand prediction. The power demand was predicted according to the expected future speed profile of the ship, which was assumed to be known beforehand. The method proved out to be an effective and holistic approach for real-time ship energy system unit commitment.
Effective ship energy system unit commitment methodologies are based on a prediction of future power demand, as was the case in the mathematical optimization based method in this work. As the primary predictor of power demand is the speed of a ship, the quality of a power demand prediction is highly dependent on the accuracy of future speed profile estimation. Because of this, this dissertation also focused on research revolving around the optimization of a ship's speed profile for a predetermined route and schedule. A computationally efficient convex optimization model was proposed for ship speed profile optimization, which produced a continuous speed profile proposition as a result.