Parametric machine learning integrated approach for assessing environmental and engine variables on fuel consumption and carbon intensity

OpenAlex English

The study aims to identify the most optimal machine learning (ML) algorithm for predicting fuel consumption (FC) based on noon report (NR) data and to explore the impact of environmental and operational engine variables on the FC through a parametric study. The M5 Rules, Artificial Neural Networks, and Random Forests algorithms have been compared in this context. This study's innovative aspect lies in parametric analysis within the best-performing algorithm to explore how variations in selected control parameters influence FC and the Carbon Intensity Indicator rating. The NR data has been gathered from a tanker ship’s noon reports over a year. After feature selection for the parametric study, the adjusted data comprising the identified variables have been used to run the chosen model. The results showed that the M5 Rules algorithm is the most appropriate for the specific data, and the Beaufort scale/slip and scavenge pressure have the highest effects on the FC. The Beaufort scale/slip varies the FC annually between a 1341.27 t (−23.48%) reduction and a 2088.05 t (36.55%) increase. Similarly, the changes in scavenge pressure impact the FC from a decrease of 859.99 t (−15.05%) or increment up to 733.72 t (12.84%).