This study proposes a ship fuel consumption prediction approach aligned with the International Maritime Organization (IMO)’s carbon-neutrality goals. Voyage log data from 19 bulk carriers and 1 oil tanker are integrated, and TabM, a deep learning mo...
This study proposes a ship fuel consumption prediction approach aligned with the International Maritime Organization (IMO)’s carbon-neutrality goals. Voyage log data from 19 bulk carriers and 1 oil tanker are integrated, and TabM, a deep learning model specialized for tabular data, is applied to perform multi-ship integrated learning. In contrast to previous studies that rely on dynamic variables measured during operation, such as RPM and engine power, and are typically trained on single-ship datasets, the proposed model uses only 17 pre-departure planning variables (speed, draft, cargo amount, and meteorological–ocean forecasts), allowing application at the voyage planning stage. The TabM-based integrated model reduces the average mean absolute percentage error (MAPE) from 11.44% to 10.01% compared with single-ship models, corresponding to a 1.43 percentage point improvement that is statistically significant across ships (paired t-test, p < 0.05). In contrast, LightGBM and polynomial regression degrade under integrated learning. These results indicate that TabM can jointly capture ship-specific behavior and cross-ship regularities, supporting fuel-efficiency optimization and greenhouse gas reduction in maritime operations.