Machine learning-based forecasting of onboard thermal, electrical, and fuel load for maritime decision support

Improving onboard energy efficiency directly affects operational costs, regulatory compliance, and environmental performance in maritime operations. To address these challenges, a data-driven framework is proposed to forecast thermal and electrical load under varying operational and environmental conditions, while also enabling fuel consumption analysis. High-frequency operational data collected from a large cruise ship were used to develop and optimise machine learning models. Data preprocessing included the treatment of missing values and the removal of outliers to ensure robustness and reliability. A correlation-based analysis was then employed to identify the most relevant input features. Fuel consumption predictions achieved a maximum deviation of 2.7% from measured values, demonstrating strong predictive accuracy. Model interpretability was enhanced through SHAP value analysis, providing insights into the influence of key variables. The best-performing models were deployed within an interactive Streamlit-based dashboard, supporting both real-time and batch predictions of load and fuel consumption. The resulting tool offers an intuitive interface and actionable insights for ship operators, facilitating informed decision-making and promoting energy-efficient maritime operations. The transferability of the proposed framework is demonstrated under different environmental and operational conditions, showing that reliable predictions can be achieved through limited domain adaptation.