Electrical Energy Modeling In Y2E2 Building Based On Distributed Sensors Information

Close to 40% of the primary energy consumption in the U.S. comes from commercial and residential buildings [1]. Therefore, reducing this energy consumption is very important, both economically, and environmentally (due to the amount of CO2 emitted in the process of generating the electricity). Predicting the energy consumption of buildings (energy modeling) is a key component in reducing the building energy consumption. Two traditional types of energy modeling are forward modeling (using purely physical simulations) and inverse modeling (using statistical methods along with expert knowledge, to relate the energy consumption to a set of general inputs such as the outdoor temperature, etc.) [2]. If we have access to sensor data (both inside and outside the building) however, a better alternative is sensor based energy modeling using statistical machine learning techniques. [2] uses seven different machine learning algorithms to predict the total hourly energy consumption of three residential buildings (called Campbell Creek Homes) designed to evaluate the effectiveness of residential construction and efficiency technologies in a controlled environment. In this project, we apply three of these algorithms (Simple Linear Regression, Support Vector Regression or SVR, and Least Squares Support Vector Machines or LS-SVM) to the energy consumption and sensor data from the Y2E2 building on Stanford campus. This rich data-set allows us to model different parts of the total consumption (i.g. lighting, AC, plug loads, etc.) separately, which could lead to more accurate results compared to the case where we are modeling the total consumption as a whole.