Computer Science Colloquia
Thursday, July 31, 2014
Adviser: Kamin Whitehouse
Attending Faculty: Jack Stankovic (Chair), Kevin Sullivan and Worthy Martin
2:00 PM, Rice Hall, Rm. 504
Master's Thesis Presentation
Drapes: A Decision Tree Based Predictive Heating Solution for Smart Home Applications
Home air conditioning is the cause for a substantial portion of total energy usage in many countries. Many home conditioning systems spend much of their energy heating, cooling, or maintaining comfortable temperatures when unoccupied. Allied with an accurate occupancy prediction system, smart conditioning systems could save homeowners a significant amount of money by heating only when there are occupants present. A model to predict occupancy must determine which features of a home are highly correlated with occupancy and which are not. Previous models have used features from a given room or zone, but ignore the relationship between rooms in a home. In this paper, we introduce Drapes, a decision tree based predictive conditioning system that makes use of a holistic view of a residence. We explore Drapes by analyzing seven home occupancy data sets ranging in size from twelve days to eighty. Our experiments and analysis show that Drapes is able to infer zone level occupancy and condition the home accordingly such that, when compared to state of the art heating algorithms without reactive heating components, occupants' discomfort is reduced by approximately 33%, and energy waste is reduced by 15% on average.