Predictive models development using gradient boosting based methods for solar power plants

Abstract

Being able to predict the power to be generated by solar power plants in a smart grid, microgrid or nanogrid with high accuracy and speed brings a lot of advantages in the decisions to be made for these systems. Making power generation forecasts, which are strictly dependent on the dynamic energy management of these grids, influences many factors from the amount of energy to be stored to the cost of energy. In this study, the development and analysis of three gradient boosting machine learning-based methods for power prediction are carried out. Innovative and fast predictive models are designed with XGBoost, LightGBM and CatBoost algorithms. These models, which have a training set consisting of several meteorological features, offer considerable benefits such as high accuracy and fast learning. Further, the performances of these models are compared and their applicability is discussed.