Date of Award
Santa Clara : Santa Clara University, 2020.
Doctor of Philosophy (PhD)
Residential and commercial buildings consume more than 40% of energy and 76% of electricity in the U.S. Buildings also emit more than one-third of U.S. greenhouse gas emissions, which is the largest sector. A significant portion of the energy is wasted by unnecessary operations on heating, ventilation, and air conditioning (HVAC) systems, such as overheating/overcooling or operation without occupants. Wasteful behaviors consume twice the amount of energy compared to energy-conscious behaviors. Many commercial buildings utilize a building management system (BMS) and occupancy sensors to better control and monitor the HVAC and lighting system based on occupancy information. However, the complicated installation process of occupancy sensors and their long payback period have prevented consumers from adopting this technology in the residential sector. Hence, I explored a method to detect the presence of an occupant and utilize it to reduce energy wasting in residential buildings.
Existing methods of occupancy detection often focus on directly measure occupancy information from environmental sensors. The validity of such a sensor network highly depends on the room configurations, so the approach is not readily transferrable to other residential buildings. Instead of direct measurement, the proposed scheme detects the change of occupancy in a building. The new scheme implements machine learning methods based on a sequence of human activities that happens in a short period. Since human activities are similar regardless of house floorplan, such an approach may lead to readily transferrable to other residential buildings. I explored three types of human activity sensor to detect door handle touch, water usage, and motion near the entrance, which are highly correlated with the change of occupancy. The occupancy change is not only based on one single human activity, it also depends on a series of human activities that happen in a short period, called event. As the events have different durations and cannot be readily applicable to existing machine learning models due to varying input matrix sizes. Hence, I devised a fixed format to summarize the event regardless of the total duration of the event. Then I used a machine learning model to identify the occupancy change based on the event data. The saving potential of occupancy driven thermostat is about 20 % of energy in residential buildings. However, the actual saving impact in any given house can vary significantly from the average value due to the large variety of residential buildings. Existing building simulation tools did not readily consider the random nature of occupancy and users’ comfort. For this reason, I explored a co-simulation platform that integrates an occupancy simulator, a cooling/heating setpoint control algorithm, a comfort level evaluator, and a building simulator together. I explored the annual energy saving impact of an occupancy-driven thermostat compare with a conventional thermostat. The simulation had been repeated in five U.S. cities (Fairbanks, New York City, San Francisco, Miami, and Phoenix) with distinctive climate zones.
Wang, Chenli, "Cost effective and Non-intrusive occupancy detection in residential building through machine learning algorithm" (2020). Engineering Ph.D. Theses. 32.