Forecasting and hierarchical reconciliation for renewable energy production and consumption

In response to the accelerating climate changes the energy systems of the future will need to use exclusively renewable or low-emission energy sources. To achieve this the penetration of renewable energy sources will need to increase greatly in the coming years. Herein lies a significant challenge, since the majority of the renewable energy sources that are available to us are highly weather dependent. In short, wind turbines only produce energy as the wind blows. Thus, it is crucial to have accurate forecasts of how much energy will be produced or consumed at any given time. This information will allow the operators of the energy infrastructure and other actors on the energy market to make better decisions regarding operation, planning, trade, etc.

This thesis examines the methods that are in use for forecasting energy production and consumption. It is based on four papers, each of which explore different aspects of the process of forecasting.

Paper A provides an overview of the methods used for generating forecasts for wind and solar power production. We highlight methods for both point forecasting and probabilistic forecasting, and show the importance of optimally combining or reconciling forecasts.

Paper C examines an important step that must be taken prior to forecasting, namely proper calibration of weather forecasts to the local environment. For a case study on heat load forecasting in an urban environment, we show how localizing forecasts can correct for phenomena such as the Urban Heat Islands
effect.

Paper B and Paper D examine the process of reconciling hierarchies of forecasts to make them coherent. This is still a relatively new approach, but has shown great promise in improving the accuracy of the resulting forecasts significantly. In the papers we propose adjustments and modifications to the method, that make it more adaptive and better suited to cases with energy consumption and production.