Visit the Energy Systems Laboratory Homepage.
Bias in Predicting Annual Energy Use in Commercial Buildings with Regression Models Developed from Short Data Sets
MetadataShow full item record
An empirical or regression modeling approach is simple to develop and easy to use compared to use of detailed hourly simulations. Therefore, regression analysis has become a widely used tool in the determination of annual energy savings accruing from energy conserving retrofits. The regression modeling approach is accurate and reliable if several months of data (more than six months) are used to develop the model. If such is not the case, the regression models can, unfortunately, lead to significant errors in the prediction of the annual energy consumption.
DescriptionIssues relating to bias in regression models identified from short data sets are discussed in this paper. First, the physical reasons for the differences between the predictions of the annual energy consumption based on a short data set model and on a long data set model are discussed. Then, the errors associated with the multiple linear regression model are evaluated when applied to short data sets of monitored data from large commercial buildings in Texas. The analysis shows that the seasonal variation of the outdoor dry-bulb and dew-point temperature causes significant errors in the models developed from short data sets. The MBE (mean bias error) from models based on short data sets (one month) varied from +40% to -15%, which is significant. Hence, due care must be exercised when applying the regression modeling approach in such cases.
Katipamula, S.; Reddy, T. A.; Claridge, D. E. (1994). Bias in Predicting Annual Energy Use in Commercial Buildings with Regression Models Developed from Short Data Sets. Energy Systems Laboratory (http://esl.tamu.edu), Texas A&M University; Department of Mechanical Engineering, Texas A&M University. Available electronically from