Towards an efficient regression model for solar energy prediction

Abstract

This paper describes a model for forecasting the daily solar energy. The features used in this model include precipitation, flux (long-wave, short wave), air pressure, humidity, cloud cover, temperature, radiation (long-wave and shortwave). These features along with previous data for daily solar energy received for the years 1994-2007 has been used for forecasting. The data for the features comes from a grid of sites in the United States and the data for previous years' daily solar energy comes from 98 sites in Oklahoma, United States. Two algorithms have been used for forecasting - Linear Least Square Regression and Gradient Boosting Regression. Gradient Boosting Regression has shown to be around 2.5% more accurate as compared to Linear Least Square Regression.