In recent years, building energy consumption has increased, accounting for approximately 40% of total energy consumption in the U.S, approximately half of which is from residential buildings. Given the environmental impacts associated with energy and electricity generation, and the importance of reducing these impacts to minimize climate change, it is important to work towards methods to reduce energy consumption. This work focuses on modeling improvements associated with two aspects of residential buildings that have a significant impact on energy consumption, namely occupants and their energy consuming behaviors, and residential heating, ventilation and air conditioning systems.

In residential buildings, as compared to commercial buildings, energy consumption is more highly dependent on occupants and their energy consuming behaviors. Behavioral energy efficiency is generally considered to be a low-cost method to reduce energy consumption by providing information and feedback to occupants that enables them to understand and change their energy-consuming behaviors. Information provided to occupants typically include energy use trends, as determined through data-driven modeling of historical energy use data to predict the performance of the building. This work improves data-driven modeling methods for residential buildings in two ways – first through improved treatment of outliers, and second, through development and use of a modified sequence of change point modeling methods.

The presence of outliers in energy use data can limit a model’s accuracy, limiting the confidence in the model on the part of the owner, and thus the use of the model to adjust energy consuming behaviors. In this work, three outlier detection methods are used to identify energy use outliers from a diversity of residential buildings. The causes and impact of these outliers are also evaluated for determination whether to keep or remove an identified outlier to improve model performance. Second, a modified sequence of development of an inverse change point model is proposed, to better fit energy consumption trends, as well as several modifications to the modeling method. This includes the addition of (a) a segmented change-point model, and (b) change-point models with relaxed prerequisite criteria in the cooling or heating season. The improved sequence and methods are evaluated across four different locations in the U.S., with results indicating that overall the resulting model fits better with the data and enables a larger range of building types and energy consumption patterns to be represented by a model.

In addition to occupant-dependent energy use, the HVAC system is generally the largest electricity-consuming end use in a residential building in the U.S. Yet despite the HVAC system being a large energy consumer, this HVAC system is not likely to be regularly serviced, as compared to a commercial building, in part because it requires the presence, engagement, and time from the homeowner to do so. The occurrence of an inefficiency in an HVAC system also can develop slowly over time and may not be noticeable to a homeowner, allowing the HVAC system to operate inefficiently over a long period of time before a failure occurs. This research works towards a non-intrusive data-driven assessment tool that uses building assessors data, HVAC energy demand data, indoor environmental conditions, and outdoor weather data to assess the efficiency of operation of a residential HVAC system. The results of this study should prove beneficial for homeowners and for service technicians to help target HVAC systems in homes in need of HVAC service or energy efficiency upgrades, ultimately motivating improved sustainability of residential buildings.