Analysis of Forecasts for District Heat Production using Different Models for Seasonal Partitions

Sammanfattning: District heating is a common means of space and hot water heating in Sweden. However, the demand for heating is not the same at all times. On a yearly basis more heat is required during winter, while next to none is needed in summer. Since the demand for heat load varies throughout the year, when trying to predict it, using a model that changes with the seasons can give a more accurate prediction. In this study, a forecasting model was tested to change its parameters either yearly, every three months (seasonal), monthly or weekly. The goal was to see which way of partitioning the year would give a more reliable prediction. Using statistical bootstrap to create confidence and prediction bands for the heat load, an analysis was conducted. The results show that a seasonal or monthly approach give a more accurate prediction overall and that the summer was most difficult to predict, relative to the produced heat, although transition seasons, for instance between spring and summer were more prone to large variances overall.