A hybrid digital twin–machine learning framework for short-term offshore wind turbine energy generation prediction

Abstract

Accurate short-term forecasting of offshore wind turbine output is crucial for grid integration, operational planning, and real-time decision-making in renewable energy systems. Traditional data-driven models often struggle to generalize under changing offshore conditions, while purely physics-based models lack adaptability to environmental variations. To address these issues, this study proposes a hybrid Digital Twin–Machine Learning (DT–ML) approach for short-term offshore wind energy prediction, combining physics-based turbine modeling with data-driven learning techniques. The framework utilizes a digital twin of the wind turbine system alongside ensemble machine learning models, including Random Forest, Gradient Boosting, and Artificial Neural Networks, optimized through Bayesian hyperparameter tuning. Model interpretability is enhanced using SHapley Additive exPlanations (SHAP), which identify the most influential predictors affecting power output. The model was tested with actual offshore SCADA and meteorological data over a short-term forecasting horizon. Results show the model achieves better performance in terms of lower prediction errors, higher generalization across wind regimes compared to the baseline ML models. Especially, the proposed model achieves a 40% reduction in RMSE and improved R² coefficients, demonstrating its robustness and predictive accuracy. The SHAP analysis highlights wind speed, air density, and blade pitch angle as the most influential parameters, aligning with the aerodynamic principles of turbine operation. This methodology offers a scalable and interpretable solution for short-term offshore wind energy forecasting. It can be integrated into existing SCADA infrastructure, supporting operational decisions that enhance grid stability and promote smart energy management.