Explainable Stacked Ensemble Learning for Accurate Wind Power Forecasting Using SCADA Time-Series Data

Sawant, Sanket Balasaheb (2025) Explainable Stacked Ensemble Learning for Accurate Wind Power Forecasting Using SCADA Time-Series Data. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (326kB) | Preview

Abstract

Wind power forecasting is the process of predicting electricity produced by wind turbines based on environmental and operating parameters which are usually gathered by SCADA (Supervisory Control and Data Acquisition) systems. The common issues of non-linear dependence, local anomalies and the absence of interpretability are faced by traditional forecasting methods like the physical model that is based on a numerical weather prediction and stand-alone statistical modes, such as ARIMA or linear regressors. The proposed work is an attempt at overcoming these shortcomings given the encouraging results of the study to propose a stacked ensemble learning model through which four effective regressors namely Support Vector Regressor (SVR), Random Forest, AdaBoost, and XGBoost and are combined using a meta-learner, which improves the accuracy and robustness of the proposed model. Time-series feature engineering approaches such as the rolling of windows are exploited in this model. To verify the efficiency of the model 5-fold cross-validation strategy was used and the R 2 score of this approach was astonishing 0.9292, mean squared error (MSE) was 0.0009, and root mean squared error (RMSE) was 0.0293 and this was the best model using any baseline model. Also, Local Interpretable Model-agnostic Explanations (LIME) was used to enable transparency and to identify important SCADA features impacting predictions. The results can support the claim that the stacked ensemble is scalable and reliable as a possible tool to make real-timely energy management in smart grid settings both interpretable and have high predictive accuracy.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Tomer, Vikas UNSPECIFIED
Uncontrolled Keywords:	Wind Power Forecasting; SCADA Data; Stacked Ensemble Learning; Time-Series Prediction; Explainable AI (LIME)
Subjects:	G Geography. Anthropology. Recreation > GE Environmental Sciences Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning H Social Sciences > HC Economic History and Conditions > Natural resources
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	03 Jul 2026 10:39
Last Modified:	03 Jul 2026 10:39
URI:	https://norma.ncirl.ie/id/eprint/9459

Actions (login required)

View Item