Vol 77 No 4 77411

Abstract

Turbocharger exhaust outlet temperature is a key indicator for assessing thermal load and operational safety in marine diesel engines. Accurate prediction remains challenging under transient operating conditions because the temperature signal is governed by strong thermal inertia, nonlinear coupling, and step-response behavior. To address these challenges, this study develops a data-driven prediction framework based on LightGBM. The framework introduces a step-sensitive weighting strategy to strengthen model learning in transient regions and uses Bayesian optimization for automatic hyperparameter tuning. A dynamic consistency correction module and a residual compensation module are further integrated to improve temporal smoothness and prediction accuracy. The method is validated using real shipboard data collected from the vessel “Xin Hong Zhuan”. On the test set, the proposed model achieves a mean absolute error (MAE) of 6.41 °C, a root mean square error (RMSE) of 7.38 °C, and a coefficient of determination (R²) of 0.990. Compared with the best-performing baseline model, the proposed approach reduces RMSE by approximately 33 %, indicating clear advantages over traditional regression methods and representative deep learning baselines. These results suggest that the proposed framework provides an accurate and robust approach for temperature prediction in complex marine engineering systems.