Abstract
In order to operate energy supply systems optimally from the viewpoint of energy and cost savings, it is necessary to predict the energy demand accurately. In this paper the parameters of a neural network model are identified with a global quasi-optimal method, and the method of predicting the energy demand was examined. In addition to the past energy demand, as an input to a model, weather conditions, such as an actual measurement and predicted values of air temperature, relative humidity, the sensible temperature, and the discomfort index, were taken into consideration. The prediction method was applied to the cooling demand in a building used for a benchmark test of a variety of prediction methods, and its validity and effectiveness are clarified.
