Ship energy consumption prediction: Multi-model fusion methods and multi-dimensional performance evaluation

Artikel i vetenskaplig tidskrift, 2025

Improving energy consumption prediction performance is of significant importance in ship energy efficiency optimisation. However, existing research primarily relies on single-method modelling, with evaluation metrics typically limited to accuracy, which restricts the practical application of ship energy consumption models. Therefore, this study innovatively introduces multi-model fusion methods, including Stacking, Blending, and Voting, to establish three ship energy consumption fusion models. Additionally, to comprehensively evaluate model prediction performance, three distinct evaluation metrics are proposed: accuracy, extrapolation capability, and physical inconsistency. Using a large container ship as the research subject, validation is conducted based on real operational data. The results show that, compared to eight mainstream single models, the three fusion models perform better in terms of accuracy, extrapolation capability, and physical inconsistency. Furthermore, the comprehensive performance indicator of the Voting-based fusion model is improved by 44.4580%–84.4882%, demonstrating that multi-model fusion helps enhance ship energy consumption prediction performance. The research findings improve the reliability of ship energy consumption model predictions, further increasing their practical application value.