Heat load forecasting using adaptive spatial hierarchies

District heating is an efficient method of distributing heat in densely populated areas at a low cost. The heat is usually produced at central production plants and then distributed to consumers through large networks of pipes. However, district heating is gradually becoming more decentralised with additional heat sources, e.g. heat pumps, solar thermal farms, and industrial waste heat connected to the network. Therefore, the system is changing from a system with centralised heat sources to a more decentralised system with several different heat sources within the network, including also still a large production area. Operationally this is more complex than the previous setup, especially in terms of temperature optimisation. Typically, the temperature must be adjusted for each area in order to work efficiently with the decentralised heat sources, so a forecast of the local heat load is required. It is relatively easy to make a forecast for each area, but they are usually made independently and are therefore not necessarily coherent. In this paper, we propose a methodology to spatially reconcile hierarchies of individual localised heat load forecasts with a coherency constraint. This results in coherent reconciled forecasts. Enhancing forecast accuracy and making them coherent are essential for future decentralised systems as temperature and production optimisation need accurate information to yield optimal operation. We will use two different case studies to illustrate the proposed method. One case study has a few areas, while the other case study will have more areas, and here it is proposed to add a new level of aggregation to the hierarchy to increase accuracy. The results in this paper show that the reconciled forecast, where information is shared between areas through the spatial hierarchy, improves forecast accuracy by 1% to 20%, depending on the prediction horizon.