Derin öğrenme algoritmaları kullanarak küresel yatay ışınlamanın çok değişkenli tahmini

Increasing photovoltaic (PV) panel instalments jeopardise the electrical grid frequency, especially in island countries, such as Cyprus. For a continuous growth in the PV instalments in Northern Cyprus as well as minimal usage of conventional energy sources in power generation, it is of utter importance for a grid manager to possess information on the energy production of PV panels, hence knowledge on received radiation, i.e. Global Horizontal Irradiation (GHI). Therefore, the prediction of GHI plays an essential role in the growth of renewable energy in Northern Cyprus. This study focuses on forecasting long-term and short-term GHI for Kalkanlı, Northern Cyprus. For long-term forecasting, a dataset is obtained from NASA while the short-term GHI prediction is carried out with a dataset recorded at METU NCC. Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) algorithms are employed for the long-term GHI forecasting. Support Vector Regression (SVR) is employed in addition to CNN and LSTM algorithms in the short-term GHI estimation. For both datasets, hybrid and stand-alone models are constructed, and their performances evaluated extensively. Additionally, seasonal forecasting is carried out for the short-term GHI estimation with a hybrid model of CNN, LSTM and SVR.