Solar Power Forecasting with Model Selection Analysis

Solar power has been growing at an exponential rate during the recent years. Due to the increasing share of solar energy in the grid and the associated uncertainties, the task of solar power forecasting is very crucial and challenging. In this study, a specific PV site in the state of Florida is considered. Various forecasting models are constructed with combinations of multiple training lengths and feature selections. Using the past instances of power output and selected features such as temperature and wind speed, the forecasting accuracy is compared for the linear regression model, gradient boosting model, and neural network model.