Predicting Fuel Consumptions and Exhaust Gas Emissions for LNG Carriers via Machine Learning with Hyperparameter Optimization Available to Purchase

The prediction of marine fuel consumption and ship exhaust gas emissions are indispensable to evaluating ship sustainable performance under current shipping fuel standards. Big data with evolved machine learning techniques have been proved to be an effective way to contain uncertainties for ship activities. This work collects the latest global LNG carrier fleet with 435 data points and attempts to predict the marine fuel consumptions and ship-resulted global warming potential (GWP) gas emissions, including CO₂, CH₄, N₂O, and black carbon aerosols. Gaussian process regression and ensemble machine learning approaches, to achieve this goal, are employed to infer the relationship between predictors (i.e., dimensional parameters, machinery parameters, and tonnage) and response variables (fuel consumptions and GWP exhaust gas emissions), providing exceptional insight into ship sustainable solutions. To improve the prediction accuracy, the hyperparameter optimization analysis via random search and Bayesian optimization is adopted to find the optimal machine learning model. The appealing results are in line with the validation data, illustrating high effectiveness and robustness of the proposed machine learning models. The procedure established in this study presents a novel approach for accelerating the research and development of sustainable shipping fuels under normal ship activities.