Deep learning-based daily solar irradiation prediction for Adana region

Nature suffers because of the negative effects and insufficiency of fossil energy resources such as oil, coal, and natural gas, which is why renewable energy generation has received significant attention. To replace harmful fossil fuels with clean, safe, and sustainable energy, the photovoltaic (PV) market has grown rapidly in the last decades. A high-accuracy daily solar global horizontal irradiation (GHI) prediction model is a critical tool for the operation, maintenance, and optimization of PV systems so that developing an intelligent and robust GHI prediction model has become one of the most captivating topics among researchers in recent years. In this thesis, meteorological data of Adana region were obtained from MERRA-2 via introducing calendar data, then the dataset was wrangled to treat missing and erroneous data by using Python programming language in order to form a cleansing dataset that is normalized and split into 80% and 20% for training and testing dataset respectively. Moreover, four different feature selection scenarios were created depending on Pearson's correlation and the threshold limits were specified as 0, 0.10, 0.25, and 0.5. After day-head GHI prediction models were carried out using different deep learning-based models with a statistical and a traditional artificial neural network algorithm, namely multiple linear regression (MLR), multilayer perceptron (MLP), recurrent neural network (RNN), long short-term memory (LSTM), gated recurrent unit (GRU), bidirectional LSTM, and bidirectional GRU. The prediction results were evaluated by utilizing a variety of performance metrics, including R-squared (R2), normalized mean absolute error (nMAE), mean absolute percentage error (MAPE), and normalized root mean squared error (nRMSE). The Bi-LSTM model has been observed to outperform the other models in terms of R2 (85.495%) and MAPE (19.901%). Key Words: Daily Global Horizontal Irradiation Prediction, Deep Learning, Recurrent Neural Networks, Long Short-Term Memory, Gated Recurrent Unit, Adana