Renewable energy production forecasts for Turkey by artificial neural networks and time series

Renewable energy sources are preferred because they minimize continuous or environmental pollution due to the increasing population and developing industry and changes in living conditions over time. In the study, the rate of electricity generation from renewable energy sources in Turkey should be determined appropriately in order to prevent unnecessary energy loss in the coming years. In order to prevent this loss, the dataset of electricity generation rates from renewable energy sources in Turkey between 1960-2019 was processed using time series analysis and artificial neural network estimation techniques. In addition, estimation was made from 2020 to 2025. For the performance evaluation of the estimation techniques used, the data between 1960-2014 was taken as training data and the data between 2015-2019 was taken as test data and examined on mean absolute error (MAE) and root mean square error (RMSE) values. Although it turned out that the technique that gave the best results was neural network autoregression (NNAR), it was not preferred because the values obtained in the estimation process remained constant. Instead, it turned out that ARIMA (2,2,1) model and cubic smoothing spline techniques can be preferred. In this study, it has been determined that renewable energy sources should be foreseen and more efficient planning and organization is required for this. Key words: Renewable Energy, Artificial Neural Networks, Time Series Analysis, ARIMA, Cubic Smoothing Spline, Neural Network Autoregulation Method