Analyzing the Predictive Power of Solar Photovoltaics Using Three Machine Learning Models

The meteorological data corresponding to Vellore Institute of Technology, Chennai is downloaded by pinpointing the institute's latitude and longitude in the National Solar Radiation Data Base viewer. The photovoltaic power generated by the panel is calculated using the downloaded meteorological parameters due to unavailability of power generation data. The collected data is pre-processed, relevant features for power prediction are selected, split into training and testing datasets. Three machine learning models - namely, Cubic Support Vector Machine, Bagged Trees and Rational Quadratic Gaussian Process Regression are trained to forecast the power generated from a photovoltaic panel. The models were evaluated using error metrices. The Coefficient of Determination, Root Mean Square Error and the Mean Average Error of all the three models was found during validation phase. The testing phase showed that the latter model is the better out of the three machine learning models under consideration.