Energy usage within buildings in the United States is a very important topic because of the current price of natural gas, steam, and electricity and the ever-increasing depletion of fossil fuels. Therefore, it is necessary to carefully analyze building energy consumption. Currently, there are several methods used for analyzing a solitary building's energy usage. One method involves the use of a neural network (NN) model. The current use of NN for building energy prediction typically requires the collection of hourly energy usage data from a single building using data loggers during several different seasons and occupancy levels in order to create a robust training data set for supervised training of the NN model. Well designed NN energy models are able to predict with a high degree of accuracy (+/- 5 to 10 percent on average), but the upfront data collection can prove to be quite time consuming. Therefore for groups of buildings on a campus or in a city, an alternative method for predicting energy consumption using NNs must be explored. Using readily available monthly energy and weather data from several buildings owned by the city of Rochester, New York, three methods involving NN models are created and validated to find the optimal NN configuration for predicting energy usage. The results of these trials have shown that it is feasible to utilize NNs trained only on readily available data as a warning system for buildings in need of a thorough check and possibly preventative maintenance. Based on the results of the validation trials, it was discovered that the predictive ability of the multiple building trials is poor due to the variability of the data. The usage of data from similar buildings and single buildings improved the predictive ability. The best prediction results occurred by using a single output network trained on data from a solitary building.

Library of Congress Subject Headings

Buildings--Energy conservation--New York (State)--Rochester; Energy consumption--New York (State)--Rochester; Energy consumption--Computer simulation; Neural networks (Computer science)

Publication Date


Document Type


Student Type


Degree Name

Mechanical Engineering (MS)

Department, Program, or Center

Mechanical Engineering (KGCOE)


Margaret Bailey

Advisor/Committee Member

Mark Kempski

Advisor/Committee Member

Edward Hensel


RIT – Main Campus