Using Data- Driven Building Performance Modeling using Artificial Neural Network Model

Developing countries face the challenge of finding new ways to conserve energy usage in new construction as most of the construction is still going on, while developed countries like the United States face the challenge of the implement or optimizing energy conservation in existing building. For this prediction of building energy using simulation program and regression models has proven to be instrumental in optimizing set operating schedules, evaluating retrofit options and other energy conservation measures. However, studies have shown that the predicted energy at operational stage does not always match the actual energy usage. Thus, the aim of the research is to improve the accuracy of building energy prediction techniques. Assumptions about occupancy schedule, equipment load and use of weather data different from site conditions are few of the factors affecting the accuracy of prediction. Taking this into consideration, for an office building located in downtown Los Angeles, actual occupancy, weather data and sub-metering is measured. The measured data is used to perform calibration of energy model. To find the impact of occupancy, weather and sub-metering data, the measured data is inputted in the energy model individually and all together. It is found that by using the measured occupancy the prediction results were 5% more accurate than others and when all the measure data is used the results are 7% more accurate. Further, a time series regression model with artificial neural network (ANN) is proposed. Different network structures are studied. It is found that ANN can predict energy usage at 10% accuracy.

University of Southern California

© 2016 by Joon-Ho Choi. Proudly created with

  • Facebook Clean Grey
  • Twitter Clean Grey
  • LinkedIn Clean Grey