1. Dataset

Dr. Frederick Maier and Chris Barrick in the Institute of Artificial Intelligence have been working together since late 2016 to build a dataset, so as to train machine learning models to predict solar irradiance. This dataset consists of two major parts. The North America Mesoscale Forecast System (NAM) which is one of the components of this dataset, is one of the major weather prediction models run by the National Center for Environmental Prediction (NCEP), for producing weather predictions using a variety of weather parameters such as temperature, precipitation, turbulent kinetic energy, etc. Here, all the non-planar features from the NAM dataset are discarded, since we are only concerned with the weather conditions near the Earth’s surface. Georgia Power operates a solar farm in the University of Georgia, Athens, and has provided near real-time data from several pyranometers (by multiple manufacturers) in the farm. We focus on observations from a particular pyranometer installed on a particular fixed array. These observations shall act as the ground-truth irradiance observations in our regression problem, and thus, is the second major component of our dataset.