A comparative study of intelligent methods for predicting ship fuel consumption using navigation monitoring data

Ship fuel consumption prediction plays a crucial role in navigation decision-making and the intelligent evaluation of energy efficiency, particularly for future Marine Autonomous Surface Ships (MASS). This study leverages an onboard measurement and data acquisition system installed on a 28,000 DWT bulk carrier operating on global routes. With the system, navigation-related data from 2010 to 2016 across different sea areas, loading conditions, and meteorological and sea states were collected and analyzed, including ship speed, course, sway, main engine speed, and environmental parameters. Using real-time inputs such as wave height, wave direction, speed, wind speed, pitch angle, main engine power, and main engine speed, a fuel consumption prediction model was developed based on the lightGBM algorithm. The performance of this model was compared with other machine learning approaches, including Support Vector Regression (SVR), Long Short-term Memory (LSTM), Gated Recurrent Unit (GRU), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGBootst). Results show that the proposed model achieves superior performance, with RMSE reduced by 7.26%, MAE reduced by at least 2.62%, R² increased by 0.23%, and runtime shortened by 73.76%. Furthermore, the navigation data was divided into four subsets based on actual loading conditions to further validate the generalization capability of the lightGBM model. The results indicate that the proposed LightGBM model provides an effective solution for predicting ship fuel consumption, striking a balance between accuracy and computation efficiency. This study also provides a valuable reference for selecting optimal fuel consumption prediction methods in comparable vessel types.