Prediction of global solar radiation by machine learning methods: A case study

Today, fossil fuels (petroleum, natural gas, coal, etc.), which meet the majority of energy needs, both harm the environment and are likely to be exhausted in the near future. For this reason, scientists are sought to find alternative energy sources. Due to their sustainable and clean nature, many countries are started to prefer renewable energy sources to meet their energy needs. Solar energy is a rapidly developing energy source among renewable energy sources. The temporal period and amount of the global solar radiation descending to the earth from solar radiation sheds light on the Solar Power Plant (SPP) technologies to be established. With artificial intelligence applications such as machine learning and deep learning methods, predictions can be made by using historical data. For this reason, global solar radiation prediction is realized for the provinces of Artvin, Rize, Batman, Diyarbakır, Mardin, Siirt and Şırnak, which have low and high radiation values. For cases where the prediction horizon is selected 1 to 3 hours ahead, the results obtained from the methods such as Decision Trees (DT), Support Vector Machine (SVM), Linear Regression (LR), Regression Trees (RT), Kernel Approximation Regression (KAR) and Artificial Neural Networks (ANN) are extensively compared. Performance metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE) and Correlation Coefficient (R) were used to evaluate the validity of this thesis study. The prediction of global solar radiation in regional and variable weather conditions is an important parameter in finding the power of the SPP to be established. With this thesis, it is aimed to contribute to making investments and planning more effectively in both short and long term for the SPPs to be established in the determined provinces.