Abstract
In operating an energy supply system, it is important to determine the operational strategy such as on/off status of operation and energy flow rates for constituent equipment properly based on predicted energy demands. However, actual energy demands may differ from the predicted ones. In this paper, a method of estimating prediction intervals for energy demands based on the density distribution of prediction error is proposed. Absolute and relative prediction intervals are estimated and used selectively to make the prediction intervals narrow so that they are suitable for robust optimal operation. This method is applied to cooling demand prediction of a commercial building, and its effectiveness is ascertained.
