Vessels experience additional resistance by waves during navigation, which becomes a factor that increases energy consumption and exhaust gas emissions. Proper estimation and understanding of this additional resistance is an important task in the marine industry. In this study, we propose a machine-learning model that predicts added resistance in arbitrary wave headings using basic ship parameters. First, extensive model experimental data on added resistance for different ship types and sizes of ships were acquired. To build a proper machine learning model, algorithms such as extreme gradient boosting (XGB), random forest (RF), artificial neural network (ANN), k-nearest neighbor (ANN), gaussian process regression (GPR), and support vector regression (SVR) were considered. Through nested cross-validation, the evaluation and hyperparameter tuning of algorithms were performed together. As a result, SVR was selected among the candidate models due to high accuracy with robustness to the outliers. In the validation with test data of head waves and all wave headings, the R2 scores of the selected model were 0.6738–0.7584 and 0.6744–0.7449, respectively, which was better than estimation methods for added resistance in head waves such as STAWAVE-2 and Cepowski (2020), and similar accuracy to those applicable in arbitrary wave headings. Even estimation of added resistance in irregular waves of sea states, the relative deviation with the semi-empirical methods for arbitrary waves was not large, on average 10%.

This content is only available via PDF.
You do not currently have access to this content.