Evaluation of solar power forecasting using deep learning: a case study in Izmir, Turkiye

Abstract

Countries’ ambition to achieve independence from foreign energy sources, coupled with the need for future energy production forecasts based on reliable information, not only enables the safe operation of electrical networks, but also enhances the economic efficiency of these systems designed to utilize energy resources. Therefore, the prediction of energy production from renewable energy sources has emerged as a highly researched topic of considerable interest. Deep learning algorithms, such as Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and One-Dimensional Convolutional Neural Networks (1D-CNN), have been demonstrated efficacy in diverse forecasting tasks, including economic time series and computer vision. However, their application to energy production forecasting from renewable energy plants has only recently seen a significant surge. This study examines LSTM, GRU, and 1D-CNN based time-series forecasting experiments for predicting solar power generation in ˙Izmir, the third largest city in T¨urkiye. The predictions have undergone comparative analysis using various statistical calculations, and the results are depicted visually through graphs. The primary objective of these computations is to deliver an optimized academic outcome, potentially necessary for the development of new solar energy fields. This could significantly contribute to the amplified usage of solar energy, a sustainable and cleaner energy source, in T¨urkiye.