Wind speed forecast in Bilecik province using deep learning

In recent studies, deep learning-based models are widely used for short-term prediction of wind speed. Within the scope of this thesis, LSTM (Long Short-Term Memory) model, which is one of the most suitable deep learning methods, was used since wind speeds are time series type. The LSTM neural network has all the advantages of a deep learning RNN network, but also has strong nonlinear processing capability as it overcomes the vanishing gradient problem of the RNN network and is suitable for unsteady wind speed prediction. In the study, the missing wind speed data in the hourly wind speed raw data obtained from the Bilecik Meteorology Directorate were completed by preprocessing. Then, 10 years (2010-2019) processed hourly wind speed data obtained from Bilecik Meteorology Directorate were put into training in our deep learning model designed for various input sizes, and the best accurate prediction of wind speeds at certain time intervals was realized. For this, 10 years of hourly wind speed data was trained by using 67% of the data for training and 33% for testing in the 3-layer LSTM model we created. Forecasts for 1 day, 2 days, 1 week and 1 month of January were made. In order to obtain better forecast values, two different types of LSTM models, Univarate Bidirectional and Multi-Step Vector Output, were used and it was investigated which one is more suitable for hourly wind speed forecasting at certain intervals. Another result of the study is the effect of different sizes of 1-year and 10-year hourly wind speed data used in LSTM models on the accuracy of the models and the results are compared.