Load Prediction of Hydrogen Powered Ship: A Deep Prediction Framework Considering Operating Environment and Error Correction

The load prediction for the future period can serve as an important reference for ship navigation planning. However, the small displacement of hydrogen powered ships and the complex operating environment and conditions pose a serious challenge to load prediction. This study develops an advanced load forecasting framework for hydrogen energy ships by integrating navigation data with hydro-meteorological conditions, proposing a novel strategy that combines classified load prediction with unified error correction. The methodology begins with Spearman correlation analysis to identify critical features for both propulsion and maintenance loads. LightGBM models then generate rapid preliminary predictions for these load categories, employing continuous training to enhance computational efficiency while producing initial forecast sequences. The framework's key innovation is the Int-ConvGRU network, which implements error correction through a parallel architecture that simultaneously processes historical data and preliminary predictions while preserving precise temporal relationships. This unique design significantly improves prediction accuracy by effectively capturing complex temporal dependencies. Comprehensive validation using data from hydrogen-powered vessels demonstrates the framework's superior performance against multiple benchmark models, particularly in handling the distinct load characteristics of modern ships.