Kaukolämmön ja -jäähdytyksen lyhyen aikavälin kysynnän ennustaminen käyttöpaikkakohtaisesti koneoppimismenetelmiä hyödyntäen

Heating and cooling of buildings account for one third of global final energy consumption and greenhouse gas emissions. Combined with urbanization, increasing energy demand, and an urgency to mitigate global warming, district heating and cooling play a crucial role in achieving a sustainable energy system. In Finland, district heating is the most common form of heating, especially in urban areas, while demand for district cooling has also consistently increased. To build a sustainable and optimized district heating and cooling system, accurate load forecasts are needed. District heating and cooling are increasingly supplied by multiple small heat sources utilizing variable renewable and local energy sources instead of a single large plant. With metering-point-level forecasts, demand can be predicted at various levels of aggregation, enabling regional optimization of the more fragmented production. Furthermore, these forecasts provide a deeper understanding of demand patterns at each individual metering point. In this thesis, hourly metering-point-level district heating and cooling loads are predicted on a short-term horizon using machine learning. The performance of six different methods for predicting total and metering-point-level energy demand is compared. Furthermore, the effect of weather forecast errors on prediction accuracy is studied. The results indicate that the machine learning models can predict the metering-point-level heating and cooling demands with reasonable accuracy while obtaining good performance on the total energy demand. For the aggregated total demand, an NRMSE of 5.5% and 20.5% are obtained for heating and cooling, respectively. Weather forecast errors are shown to significantly impact forecast accuracy. While prediction accuracy varies considerably across individual metering points, the results demonstrate that metering-point-level forecasting is a promising approach that can be utilized for many practical applications.