Predicting Residential Heating Energy Consumption and Savings Using Neural Network Approach

Date of Award


Degree Name

Ph.D. in Mechanical Engineering


Department of Mechanical and Aerospace Engineering and Renewable and Clean Energy


Advisor: Kevin P Hallinan


Upgrading and replacing inefficient energy-consuming equipment in both the residential and commercial building sectors offers a great investment opportunity, with significant impacts on economic, climate, and employment. Cost effective retrofits of residential buildings could yield annual electricity savings of approximately 30 percent in the United States. This obviously could reduce greenhouse gas emissions in the U.S. significantly. Further, investment in energy efficiency can create millions direct and indirect jobs throughout the economy for manufacturers and service providers that supply the building industry. Unfortunately, the prediction in savings, which has been generally based upon energy models, has been circumspect, with energy savings typically over-predicted. Investor confidence as a result can degrade. An enabler for this research is a collective grouping of over 500 residential buildings used for student housing owned by a Midwestern U.S. university. These residences offer significant variation in size, ranging from a floor area of 715 to 2800 square feet, in age, ranging from the early 1900s to new construction, and energy effectiveness, the latter occurring mostly as a result of improvements made gradually over time to some residences over the past fifteen years. The historical monthly natural gas and electricity energy consumption for these houses is available. Additionally, in the summer of 2015, energy and building data audits were completed on a total of 139 residences. Documented in these audits were the amount and type of insulation in the walls and attic, areas and types of windows, floor heights, maximum occupancy, appliance (refrigerator, range, oven) specifications, heating ventilation air-conditioning system specifications, domestic hot water equipment specifications, and the presence of a basement. Finally, county auditor real estate information was relied upon to obtain detailed features of each residence, including the age of the house, number of floors, floor area of each level, and total floor area.Using this data, a data mining approach based upon an artificial neural network (ANN) model was shown to be effective in estimating the annual heating energy savings from a variety of measures for a large number of houses for which energy characteristics are known and energy consumption data is available. In combination with cost models for implementation of the measures, the cost effectiveness of every measure considered for each residence was estimable. This preliminary study provides the starting point for the research presented here. With knowledge of the individual cost effectiveness of all measures within a collective grouping of residences, it becomes possible to adopt a strategy for energy reduction based upon a `worst to first' methodology. The economic impact of adoption of this methodology is then determined using an economic-input-output (EIO) approach. Considering only those measures that are economically viable and extrapolating the results from this study to the entire Dayton region yields with the initial energy efficiency investment of {dollar}26.1M can result in a total local economic impact of {dollar}41.2M (i.e. summation of direct, indirect, and induced) and additional economic impacts stemming from the annual energy savings of {dollar}2.21M for the lifetime of the considered EE measures.


Mechanical Engineering, Energy, Artificial neural network model, Heating energy prediction, Residential buildings, Prioritized reduction, Worst to First Strategy

Rights Statement

Copyright © 2019, author