Data augmentation-based approach to enhance the accuracy, generalization, and reliability of ship fuel consumption prediction

Artikel i vetenskaplig tidskrift, 2025

An accurate, stable, and reliable ship fuel consumption prediction model is of great significance for supporting energy conservation and emission reduction. However, most current studies mainly focus on the improvement of the model itself, often neglecting the problems existing in the dataset, such as uneven distribution and limited data, which affect the model's prediction performance. To solve these problems, this study proposes four data augmentation strategies and combines them with three prediction models to evaluate the improvement effect of data augmentation on prediction. At the same time, an interval prediction model is constructed further to analyze the reliability and uncertainty of the model. Taking the actual operation data of an LPG carrier as a case, the results show that the data augmentation methods significantly improve the distribution characteristics of the original dataset and enhance the prediction performance of the model. Compared with the original dataset, the models with data augmentation perform better in terms of MAPE and R². Among them, the GMM data augmentation method combined with LSTM achieves the most tremendous, with MAPE reduced by 22.43 % and R² increased by 18.60 %. In addition, data augmentation effectively narrows the CWC, verifying its practical value in ship energy consumption modeling.