A stacking ensemble learning approach for accurate and interpretable prediction of ship energy consumption

The accuracy and interpretability of ship energy consumption prediction results are important for ship energy efficiency optimization. In order to improve the accuracy of ship energy consumption prediction and enhance the model interpretability, this paper proposes a ship energy consumption prediction method based on Stacking and SHAP. Firstly, based on Stacking theory, multiple heterogeneous and complementary base models were selected using residual correlation analysis methods to construct a fusion model. And then, to address the “black box” characteristics of the fusion model, SHAP is used to analyze the base model and energy consumption impact characteristics of the fusion model in terms of their interpretability. A large container ship is used as the research object to verify the effectiveness and interpretability of the proposed method. The experimental results show that, in terms of accuracy, compared with the best single model (RF), the mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) of the Stacking fusion model are reduced by 4.1%, 16.1%, and 8.3%, respectively, and the R² is improved by 1.5%. Meanwhile, in terms of interpretability, SHAP reveals that Random Forest (RF), k-Nearest Neighbor (KNN), and Gradient Boosting (GB) models play a dominant role in the fusion model, with a total contribution value of about 67%. In addition, sailing speed, mean draft, and trim are the main factors affecting the energy consumption of a ship, and the contribution value of each influential feature can be quantitatively measured. The proposed method ensures the prediction accuracy while enhancing the model interpretability, which can provide more reliable and transparent decision support for ship energy efficiency management.