Engineering Mechanics Institute Conference 2013

Full Program »

Big Energy Data Identifies Retrofit Potential in Residential Buildings

DOC Abstract
View File
doc
663KB

Mohammad Javad Abdolhosseini Qomi
MIT
United States

Marta Gonzalez
MIT
United States

Franz-Josef Ulm
MIT
United States

Arash Noshadravan
MIT
United States

Abstract:
Almost 50% of national end-use energy consumption in United States is spent for space heating, ventilation and air conditioning (HVAC) in residential and commercial buildings. This places buildings as the largest greenhouse gas emission sector nationwide, solely responsible for 39% of national CO2 emissions. Therefore, identification of key driving parameters contributing to energy consumption and quantifying their impact can improve our understanding of building energy consumption at the use-phase. In this work, a new mechano-statistical approach is proposed based on conventional building heat transfer modeling and existing ideas in machine learning techniques. Using Monte Carlo sampling technique, a comprehensive global sensitivity analysis is performed in the domain of building parameters such R-values, infiltration rate, temperature set points, occupation level and so on. All variables are ranked subsequently using Spearsman Rank Correlation Coefficient method, identifying the most critical parameters in the design space to describe the energy consumption. By choosing a proper set of essential random variables and excluding deterministic parameters, a simple surrogate function is proposed that emulates the behavior of complex building energy consumption models. This surrogate function is trained and normalized under meteorological data so it can screen out the loading response of buildings under climate change. This method is applied to investigate energy bill data of 17000 households in Cambridge Massachusetts for a period of three years, 2007-2009. Given a realistic set of scenarios ranging from poorly- to well-isolated building, machine-learning techniques identify which scenario best suits the actual energy consumption of a building. We particularly assign the best scenario in reduced building parameter space to describe the building energy consumption. This produces a spatial map of buildings over the city, which can potentially benefit from public and private sector retrofit programs. With minimal use of public and private data, this method produces a smart map that can be directly used to effectively reduce the CO2 and energy footprint of cities.

 

Powered by OpenConf®
Copyright ©2002-2012 Zakon Group LLC