Abstract: Improving ship energy efficiency and reducing greenhouse gas emissions are major research priorities in the maritime industry. Accurate prediction of main engine power is fundamental to enhancing vessel energy efficiency. Using historical operational data collected from a Very Large Crude Carrier (VLCC), this study integrated and cleaned meteorological data to construct training and test datasets. Three models for main-engine power estimation are investigated and compared: a mechanistic model (SNNM), a non-mechanistic model based on Random Forest (RF), and a semi-mechanistic RF-based model. Simulation results indicate that while the mechanistic SNNM model can meet application requirements under specific engineering conditions, but R² coefficient is relatively low. In contrast, both the non-mechanistic model based on RF and the semi-mechanistic RF-based model demonstrated excellent predictive accuracy for both main engine shaft rotational speed and power, with R² values exceeding 0.98.