Özet:

Today, investments in solar power plants are increasing significantly every day. The variation of solar energy depending on the meteorological parameters makes the planning of energy production relatively difficult. Since this situation can lead to network operating problems, solar energy forecast has become an important topic. In this study, the ability of machine learning methods to predict short-term solar energy was compared. First, a set of data consisted of 1B: Trina power output data and meteorological parameters obtained from the measurement station in the area at a solar energy center in Alice Springs, Australia. The power data is then divided into the subcomponents with the Ampiric mode separation method. In this study, the most commonly used methods for the prediction of solar energy, such as Direct Regression (DR), Support Vector Machine (DVM), Decision Tree Regression (KAR), Gauss Process Regression (GSR) and Community Regression (TR), were selected. Comparative studies have been carried out to assess the predictive performance of these methods in different forecast horizons. In these studies, performance metric results such as R, RMSE and MAE for all machine learning methods were calculated. According to the metric results obtained, DVM has been observed to provide the best forecast outcome.

Anahtar Kelimeler: