Building Electricity Demand Forecasting with Temporal Hierarchies

Accurate building electricity load forecasts play a major role in the energy transition, as they facilitate flexibility deployment, grid stability and overall reduce costs and CO2 emissions. This research leaverages forecasting and reconciliation of temporal hierarchies to achieve coherent and accurate forecasts for different time scales. Two different hierarchical structures and their variants were studied, a daily and hourly hierarchy. Four different forecasting algorithms were implemented and compared, as well as a variety of reconciliation methods, their improvements of base forecasts were quantified and assessed, with up to ten percent accuracy gains for certain levels of the hierarchical structure. This thesis constitutes a comprehensive comparison of and the developed tools for temporal hierarchical forecasting and reconciliation.