Temporal Fusion Transformer for thermal load prediction in district heating and cooling networks

Accurate forecasting of thermal loads is a critical factor for operating district heating and cooling networks economically, efficiently with minimized emissions. If are known high accuracy in advance, use renewable energies can be maximized, fossil generation, particular peaking units, avoided. Machine learning has already proven to an efficient tool time series this context. One recent advancement machine the "Temporal Fusion Transformer" (TFT), which shows especially good results area forecasting. This paper examines performance TFT concrete context load networks. First, brief summary differences between other methods given. Secondly, it described how method adopted train model The data evaluate neural network based on 8 years hourly made available from city Ulm Germany. presented technique used produce 72 hours forecasts three different grids Ulm. compared that have been previously as part publicly funded research project "deepDHC", order if improvement further reduce uncertainties.