Short-term wind power forecasting using long-short term memory based recurrent neural network model and variable selection

Author(s): Umit Cali, Vinayak Sharma
University of North Carolina at Charlotte, Charlotte, NC, USA
International Journal of Smart Grid and Clean Energy, vol. 8, no. 2, March 2019: pp. 103-110
ISSN: 2315-4462 (Print)
ISSN: 2373-3594 (Online)
Digital Object Identifier: 10.12720/sgce.8.2.103-110

Abstract: Increasing concern for the environment has led to governments and companies pushing for renewable power generation. The share of wind power for the supply of electricity has been increasing during the last two decades. Increased penetration levels of intermittent power sources like wind power became a challenge for the power companies. Therefore, the need to make accurate forecasts for wind power generation has become a very critical issue for the power system operators. This paper presents a novel approach to forecasting 1 to 24 hours ahead wind power using Long Short-Term Memory based Recurrent Neural Network (LSTM-RNN). The model is based on knowledge from the data, that both weather and wind power have short-term temporal dependencies. The proposed model is implemented using historical generated wind power and Numerical Weather Prediction (NWP) data for Sotavento, a wind farm in Spain. Input parameters from the NWP data are selected by performing a sensitivity analysis for variable selection technique.

Keywords: LSTM-RNN, wind power forecasting, renewable energy, machine learning, neural network
Full Paper.pdf

Copyright © 2016 International Journal of Smart Grid and Clean Energy, All Rights Reserved