School of Professional Studies

Environmental Conscious Models for Wind Power Forecasting

Document Type

Conference Proceeding

Abstract

Accurate forecasting of wind power is important for improving the reliability and integration of renewable energy into the power grid. However, the variability of wind conditions presents challenges for achieving consistent results. In addition, training forecasting models consumes significant energy, making their environmental impact an important consideration. In this study, we evaluate a range of machine learning and deep learning models, including linear regression, XGBoost, LightGBM, multi-layer perceptron, and long short-term memory networks, to assess their ability to predict wind power output. To enhance model performance, we apply feature engineering techniques, such as lag terms, interaction effects, rolling statistics, and physics-informed residuals, to better capture temporal patterns in wind data. We also introduce two custom loss functions: an asymmetric mean squared error and a hybrid of mean squared error with physics-based constraints. Interestingly, when tested without feature engineering, these custom loss functions improve prediction accuracy by at least 5%. Results show that all models, both tree-based and neural network approaches, achieve predictive accuracy above 99%, while their energy consumption and associated carbon emissions vary considerably. LightGBM emerged as the most environmentally efficient model, achieving strong predictive performance with lower carbon emissions. Overall, this study demonstrates that accurate and environmentally sustainable wind power forecasting can be achieved simultaneously. © 2026 IEEE.

Publication Title

2026 International Conference on Modern Computing, Networking and Applications, MCNA 2026

Publication Date

5-2026

First Page

94

Last Page

100

ISBN

9798331548353

DOI

10.1109/MCNA69896.2026.11574186

Keywords

component, formatting, insert, style, styling

Cross Post Location

Student Publications

Link to Full Text

Share

COinS