Fuel Consumption Prediction Models Based on Machine Learning and Mathematical Methods

An accurate fuel consumption prediction model is the basis for ship navigation status analysis, energy conservation, and emission reduction. In this study, we develop a black-box based on machine learning white-box mathematical methods to predict rates. We also apply Kwon formula as data preprocessing cleaning method that can eliminate generated during acceleration deceleration process. The test regression are employed evaluate accuracy of models. Furthermore, use predicted correlation between rates speed under simulated conditions performance validation. discuss applying data-cleaning in model. results demonstrate feasible support broad dense distribution noise collected from real ships. improved error 4% R2 0.9977 0.9922 XGBoost RF models, respectively. After method, value reach 0.9954, which provide decision operation shipping companies.