Date of Award
Master of Science in Engineering Management
Department of Systems Engineering and Management
Torrey Wagner, PhD
United States Air Force energy resiliency goals are aimed to increase renewable energy implementation among its facilities. Researchers at the Air Force Institute of Technology designed, manufactured, and distributed 37 photovoltaic test systems to Air Force installations around the world. This research uses two types of modeling techniques, multivariate linear regression and random forest machine learning, to determine which technique will better predict power output for horizontal solar panels. Many previous solar panel prediction studies use solar irradiation data as an input. This study does not use irradiation as an input and aims to predict power output with input variables that are more readily available. If power output of a horizontal solar panel can be predicted using available weather data, then assessing the possibility of utilizing horizontal panels in any global location becomes possible. Input variables used for each model was latitude, month, hour, ambient temperature, humidity, wind speed, cloud ceiling, and altitude. The variance each model accounted was used as a comparison measure. The multivariate linear regression model accounted for 56.2% of the variance in a sample validation dataset. The random forest machine learning model accounted for 65.8% variance. The random forest model outperformed the multivariate linear regression model by accounting for 9.6% more variance. The most important variable in reducing the random forest model mean squared error was the month of the year, closely followed by cloud ceiling. Wind speed was the least important variable in reducing model error. More predictor variables are needed to increase predictability of horizontal solar panel power output if irradiation is not present as an input.
DTIC Accession Number
Pasion, Christil K., "Modeling Power Output of Horizontal Solar Panels Using Multivariate Linear Regression and Random Forest Machine Learning" (2019). Theses and Dissertations. 2348.