Predicting the energy output of hybrid PV–wind renewable energy system using feature selection technique for smart grids

Energy Reports

Sara Imran Khan¹  Hafiz Suliman Munawar²  Fadi Al-Turjman³  Erfan Khalaji⁴  M.A. Parvez Mahmud⁵  Zakria Qadir⁶  Abbas Z. Kouzani⁷  Khoa Le⁷ 

[1]Corresponding author.

[2]Computer Engineering Department, Middle East Technical University, Northern Cyprus Campus, 99738 Kalkanli, Guzelyurt, Mersin 10, Turkey

[3]Faculty of Built Environment, University of New South Wales, Kensington, Sydney NSW 2052, Australia

[4]Faculty of Chemical Energy, University of New South Wales, Kensington, Sydney NSW 2052, Australia

[5]Research Center for AI and IoT, Artificial Intelligence Department, Near East University, Nicosia, Mersin 10, Turkey

[6]School of Computing Engineering and Mathematics, Western Sydney University, Locked Bag 1797, Penrith, NSW 2751, Australia

[7]School of Engineering, Deakin University Locked Bag 20000, Geelong, VIC 3220, Australia

关键词: Smart grids;    Regression models;    Feature selection;    Prediction accuracy;    Renewable energy system;    ANN;   

DOI  :  

来源: DOAJ

【 摘 要 】

In the current technological era, predicting the power and energy output based on the changing weather factors play an important role in the economic growth of the renewable energy sector. Unlike traditional fossil fuel-based resources, renewable energy sources potentially play a pivotal role in sustaining a country’s economy and improving the quality of life. As our planet is nowadays facing serious challenges due to climate change and global warming, this research could be effective to achieve good prediction accuracy in smart grids using different weather conditions. In the current study, different machine learning models are compared to estimate power and energy of hybrid photovoltaic (PV)-wind renewable energy systems using seven weather factors that have a significant impact on the output of the PV–wind renewable energy system. This study classified the machine learning model which could be potentially useful and efficient to predict energy and power. The historic hourly data is processed with and without data manipulation. While data manipulations are carried out using recursive feature elimination using cross-validation (RFECV). The data is trained using artificial neural network (ANN) regressors and correlations between different features within the dataset are identified. The main aim is to find meaningful patterns that could help statistical learning models train themselves based on these usage patterns. The results suggest that opting feature selection technique using linear regression model outperforms all the other models in all evaluation metrics having to mean squared error (MSE) of 0.000000104, mean absolute error (MAE) of 0.00083, R2 of 99.6%, and computation time of 0.02 s The results investigated depict that the sustainable computational scheme introduced has vast potential to enhance smart grids efficiency by predicting the energy produced by renewable energy systems.

【 授权许可】

Unknown