Predicting Ship Power Using Machine Learning Methods

Abstract

One of the biggest challenges facing the shipping industry in the coming decades is the reduction of carbon emissions. A promising approach to this end is the use of the growing amount of data collected by vessels to optimize a voyage so as to minimize power consumption. The focus of this paper is on building and testing machine learning models that can accurately predict the shaft power of a vessel under different conditions. The models examined include pure empirical models, pure neural network models, and combinations of the two. Using data on two car carrying vessels for 8 years it was found that neural networks incorporating some physical intuition can achieve a mean absolute percentage error of less than 5%, and an R2 above 95%. This performance can be further improved by the addition of wave information, but it deteriorates when the data collection becomes less frequent.