Medium and Long-term Probabilistic Heat Demand Forecasting in District Heating

摘要

The common practice in industry and academia regarding heat demand forecasting in district heating systems is to use point forecasting models, which provide single valued estimations of heat demand for every period of the forecasting horizon. Heat demand is strongly correlated to the outdoor temperature, therefore, as temperature can be predicted with high accuracy for a couple of days ahead, these point forecasting models can also make good short-term forecasts of heat demand. However, as the forecasting horizon is getting longer, the accuracy of the temperature prediction is deteriorating which results in worse predictions of the heat demand. Therefore, for forecasting horizons ranging from few weeks to few years ahead, a point forecast may not be so useful to operators of district heating systems who would like to have more information regarding the probability of future heat demand to be inside some intervals. This is what probabilistic forecasting does in the form of quantiles or probability density functions.