Energy Modeling Using Feed-Forward Artificial Neural Networks

Buildings are responsible for over half the energy use in this country. Building energy use can be reduced by a few actions and one is energy conservation. This can be accomplished by knowing what the building energy use is currently and even better is what the energy use is going to be tomorrow, or next month, or even next year. By use of building energy models this will help monitor performance to ensure buildings are operating at peak performance and yield non-HVAC energy use. There has been many studies in this area and one of the many that have been done is by American Society for Heating, Refrigeration, and Air-Conditioning Engineers, also known as ASHRAE, called ‘The Great Energy Predictor Shootout’ in this study many different models were used to predict or model energy usage most notably was Artificial Neural Networks (ANN), Linear, and Non-Linear Regression. From the study ANN was shown to consistently predict energy data and has been the model chosen for energy building modeling here at Boise State University. Here at Boise State the Office of Campus Sustainability has been monitoring building energy data for the last 2 years, this will give us more than enough required for the data set for ANN modeling. To date, the Office of Campus Sustainability has been implementing programming scripts to run the ANN model and have been discovering the best variables to be used as inputs to the model. Current buildings being studied are the Mathematics and Geology Building and the Barnes Towers Residence Hall. Using statistical significance these buildings the research has shown an 80 percent agreement, R-squared value, with the modeled data and the actual data. Research is still ongoing to refine this significance to 90-95 percent agreement