A Machine Learning Based Application for Predicting Global Horizontal Irradiance

The adoption of residential solar energy solutions has become a popular alternative for many families as the need for alternative energy resources begins to increase and this is causing a fluctuation in the estimated electrical energy needs forecasted by energy companies. This is a problem because electric companies base their production amounts on regional estimated energy needs and might easily overproduce or underproduce energy unless they can better estimate the availability of solar energy in each region and create a solution for monitoring its residential adoption across the same region. This study investigates the former by using ensemble machine learning algorithms to build a more accurate general predictive model which can be used predict the amount of usable solar radiation available at any given hour of the day and at any location. This method is a newer approach and is based on the reliability of a given combination of weather features. The established success metric for the study was to meet or exceed a .900 R-2 value with the lowest RMSE possible. Four predictive models were created using data from the National Solar Radiation Database which represented various meteorological and solar measurements taken over a given period. Training data was created from one year of normalized observations taken every half hour; testing data was sampled from an additional year of data taken a decade later. Of the four methods studied a Cubist Model Tree performed best with a .959 R-2 value and an RMSE of .0667. This new model was then incorporated into a GPS-based mobile application that could be used to predict the amount of usable solar radiation at any given location and at any given time of day. This application is part of a more comprehensive decision support system that addresses the overproduction of electrical energy through the adoption of new technological solutions and a proposed system for monitoring and/or controlling consumer energy consumption.