ISSN :2687-6418

Short Term Wind Speed Forecast By Using Long Short Term Memory


Number of pages:


While wind is one of the most difficult variables to predict; well-configured and high-resolution numerical weather prediction (NWP) models can be quite accurate. On the other hand; the accuracy of the wind speed forecasts can be significantly increased by coupling of NWP models and machine learning algorithms. Even though machine learning algorithms can also be used directly for forecasting the energy generation of a wind farm without forecasting the wind speed; wind power generation forecast can be done by firstly forecasting the wind speed and then using the power curve of the related wind turbine. This study will be focused on short term spatio-temporal wind speed forecast of the five airports where are located at Istanbul, Izmir, Mugla, Tekirdag and Eskisehir provinces of Turkey by using their own historical wind speed data and long short-term memory (LSTM) which is an artificial recurrent neural network. Furthermore, two years of Meteorological Aerodrome Reports (METAR) have been used as the data of the chosen airports. Besides, Auto-regression which uses only wind speed data of the related airports has been used and the previous twelve hours have been used for each initial state in order to gain information from the training of LSTM model. Besides, Two and three hours ahead forecasting has been done in this study and Pearson correlation, MAE, RMSE, and nRMSE have been used as statistical performance metrics to evaluate and compare the results.


Article Statistics

Number of reads 1,262
Number of downloads 750


Journal of Research in Atmospheric Science (JRAS)
E-Mail Subscription

By subscribing to E-Newsletter, you can get the latest news to your e-mail.