Project Team
Students
Simon Pezalla
Energy Engineering
Penn State Abington, Penn State University Park
Faculty Mentors
Renee Obringer
Penn State University Park
Earth and Mineral Sciences
Project
https://sites.psu.edu/mcreu/files/formidable/2/real-final-MCREU-poster-horizontal.pdf
Project Video
Project Abstract
Despite the recent advances in battery technology, society does not have a system capable of efficiently storing large quantities of excess power. Because of this, many powerplants must constantly match real-time demand or cities risk largescale power outages. To make matters worse, the warming climate will bring unexpected changes to household energy consumption that could cause generation facilities fall behind on production. While there is research on implementing machine learning algorithms for predicting energy use, this project is unique because it is focused on understanding climate dependent air-conditioning (AC) use on a household level which could then easily be scaled up. We started by cleaning and combining the data into one set consisting of AC use (kWh) and eight climatic variables. However, at this point it was clear our sample size was not large enough to represent a population. Nevertheless, we decided to continue and tested three commonly used machine learning algorithms: Generalized Additive Model (GAM), Multivariate Adaptive Regression Splines (MARS), and Random Forest (RF). Our results show that GAM was able to make predictions with the lowest error (RMSE/kWh = 8.827). However, the real relationship is likely more complex and couldn’t be captured with the small sample size. These findings demonstrate the promising results received from leveraging modern machine learning techniques and add to the literature explaining the best algorithms for representing the relationship between AC use and climate conditions.
Evaluate this Project
Use this form link to provide feedback to the presenters, and add your project evaluation for award(s) consideration.