A Meta-Learning Framework for Short-Term Wind Power Forecasting with SCADA and Weather Data

This is the Author Accepted Manuscript of the article published as: A Meta-Learning Framework for Short-Term Wind Power Forecasting with SCADA and Weather Data, DOI: 10.1109/IISA66859.2025.11311208 Forecasting wind power accurately remains a critical challenge in integrating renewable energy into modern power systems. In response to this challenge, this paper presents a hybrid ensemble learning framework for wind power point forecasting, combining multiple Machine Learning regressors through a meta-learning strategy. The approach leverages historical SCADA measurements, weather forecasted data and physically derived theoretical power curves to enrich the feature space. A diverse set of base models, including linear, tree-based, and gradient boosting regressors, are trained on a unified dataset, and their predictions are subsequently used to train a Multi-Layer Perceptron (MLP) meta-learner. This stacked ensemble methodology captures both the individual model strengths and their inter-dependencies, leading to improved generalization. Experimental results, based on a real-world dataset from a Vestas V52 wind turbine in Ireland, demonstrate that the meta-learner consistently outperforms the individual regressors in terms of accuracy, achieving R2 scores of 0.99. This proves that this study offers an effective solution for improving short-term wind power forecasting.