Multi-dimensional performance verification of ship fuel consumption prediction model under dynamic operating conditions

Given stringent decarbonization targets in the global shipping industry, validating the accuracy and applicability of fuel-consumption prediction models using actual navigation data is critical for informed energy optimization, cost control, and emission management. To address shortcomings of existing grey-box models in cross-condition verification and adaptability assessment, this study develops a systematic, multi-dimensional testing framework. The framework is for a classified weighted fusion strategy-based grey-box model (CWFM), encompassing both single and representative combined operating conditions. Model performance is comprehensively evaluated using root mean square error (RMSE), coefficient of determination (R2), coefficient of variation (CV), and generalization error (GE). A case study demonstrates that, under combined operating conditions, CWFM reduces average RMSE by 0.28–0.35 kg/h versus the optimal sub-model, improves R2 by 5.5 %–7.5 %, narrows confidence-interval widths by 43.6 %, maintains CV below 0.4 %, and validates the dynamic weight allocation strategy's effectiveness in enhancing prediction performance. Furthermore, cross-operating conditions GE is 83.1 % lower than that of traditional mechanistic models; the RMSE ratios under combined operating conditions (0.31–0.40) outperforming comparison models. The developed framework ensures that fuel consumption prediction models adequately meet the actual operational requirements of ships, providing for achieving the goals of green shipping.