A machine learning method to evaluate head sea induced weather impact on ship fuel consumption

A ship's fuel consumption is significantly affected due to ship motions caused by waves and wind when sailing under ocean weather conditions. An essential step to develop certain energy efficiency measures is to understand, model and estimate how much extra fuel consumption is caused by encountering weather conditions, and from which components of a ship's energy system that extra consumption is attributed to. In this study, experimental tests of added resistance in waves during the past decades in open literature are collected and a Gaussian process regression (GPR) model is developed to describe a generic ship's added resistance in head waves. The proposed GPR model achieves better prediction accuracy than semi-empirical formulas (white box) and gives more rational transfer function of added wave resistance coefficient than those produced by the artificial neural networks (ANN), especially in the short-wave regime. The proposed GPR model is integrated into a grey box prediction framework for ship fuel consumption using several years of performance monitoring data collected onboard a chemical tanker. The prediction results indicate an improvement in model performance when moving from the white box to the grey box model, with R2 increasing by 38 % and Root Mean Square Error (RMSE) decreasing by 65 %. Finally, the investigation of weather impact on the ship's extra fuel cost is demonstrated by the proposed model.