Big data techniques for ship performance study
Paper in proceedings, 2018
In this paper a new way to predict the propulsion power through the use of Big Data techniques is presented in order to improve the current way of evaluating the performance of a ship to lower emissions and for greener future operations. Big Data is a promising technological solution that can potentially improve the techniques used to evaluate ship performance by producing value from data. For Big Data techniques is intended the implementation of Machine Learning models for data analysis. For the study, real data collected from a LCTC M/V is used and, in particular, the data concerning the performance of the hull. The features used to predict the propulsion are: speed over ground, speed through water, wind intensity and direction, course, heading, rudder angle, roll, pitch, forward and aft draft. This data works as the input for the Machine Learning for the prediction of the propulsive power. The Machine Learning models used are the XGBoost, short for eXtreme Gradient Boosting of the gradient boosted trees, and the Multi-layer perceptron, of the Neural Network. The models are taken from the Scikit-learn Python library (Pedregosa et al., 2011). The data is divided in voyages, so that predictions of part of the voyages are made. The results are assessed with the R2 (coefficient of determination) and Mean Absolute Error, Machine Learning metrics, giving an accuracy around 10% depending on the voyage.