Abstract
It is important to reduce CO2CO2 emissions. Accurate power prediction of vessels allows evaluation of performance which is essential for this. Neural networks have been shown to accurately predict powering of vessels in weather, which is not straightforward using traditional methods. In previous applications only 6 intuitively relevant variables are used to predict powering with neural networks, including wave height, which is expensive and time consuming to collect. This study investigates the influence of the wave height variable on prediction, concluding that it increases accuracy of prediction by 0.5%. A high frequency dataset from a merchant vessel containing 100 variables is statistically analysed and a maximal subset of variables is derived. This subset is used for principal component analysis, focussing on redundancy. Investigation into the inclusion of non-intuitively relevant variables to increase prediction accuracy for power prediction with a neural network is performed.
