UBC Theses and Dissertations
Improving the feasibility of energy disaggregation in very high- and low-rate sampling scenarios Clark, Michelle Susan
Given the world's urgent need to reduce greenhouse gas emissions in order to avoid the most disastrous effects of climate change, efforts must be made to reduce these emissions in every way possible. A large share of these emissions come from the energy consumption in buildings and there are significant opportunities to reduce this consumption through energy saving measures. Energy disaggregation or non-intrusive load monitoring (NILM) is a useful tool that infers the energy consumption of individual appliances or equipment within a building from detailed measurements of the building's total energy consumption. This method is very attractive for providing a detailed breakdown of building energy consumption because it is less expensive and more convenient than measuring the energy use of each appliance individually. A wide variety of NILM methods have been proposed and in this thesis we focus on improving the feasibility of two different classes of NILM methods. We first explore the use of random filtering and random demodulation, two methods closely related to the new and developing field of compressed sensing (CS), to acquire and manage very-high-rate electrical measurements used for NILM. We show that these methods allow us to reduce the required sampling rate and volume of data collected while retaining valuable signal information required for NILM. Second, we switch to the analysis of very-low-rate data for NILM and develop a method to detect interesting patterns in the very-low-rate aggregate consumption signal. These patterns are shown to be responsible for a significant share of the total energy consumption in some buildings and are also related to the outdoor air temperature in some cases. Taken together, the two parts of this thesis allow us to contribute to the field of NILM by improving its feasibility and helping to facilitate its widespread use.
Item Citations and Data
Attribution-NonCommercial-NoDerivs 2.5 Canada