Predicting and mobilizing energy flexibility in intelligent buildings

Abstract

In response to climate changes caused by green-house emissions, the energy systems are transitioning to an increased reliance on intermittent renewable energy. Due to this transformation, energy flexible assets that help balance of supply and demand side will be increasingly valuable in the coming years. One such potential source of energy flexibility can be found when considering the building stock. As buildings become increasingly digitized, connected and controllable, the possibility to unleash the capacity for energy flexibility inherent in the buildings arises. With buildings producing vastly more data on energy consumption, indoor environment quality and occupant behavior, new methods can be used to estimate the potential for flexibility. In a similar manner, data-driven methods can quantify the flexibility in actual consumption when the potential is mobilized for e.g. peak-shaving or demand response. The state-of-the-art suggests model-based methods for control of flexible assets, however the modelling efforts can be an obstacle to practical implementation in buildings. In addition, the occupants pose constraints to the flexibility that can be utilized, and thus increasingly human-centric approaches are needed. This PhD thesis aims to address challenges to flexible consumption in buildings by data-driven and model-free methods for both analysis and control. We adopt a holistic view, considering both the occupants and the energy systems context, by anchoring our research in the living laboratory of EnergyLab Nordhavn. We show that real-world investigations are required in order to understand the real constraints and limitations that flexible energy consumption in buildings face. Further, we contribute to development of the living-lab infrastructure with new algorithms for real-time control. To transform measurement data from the many sensors and meters to meaningful information, we adopt methods from machine and deep learning. We use the methods to analyse energy consumption data from buildings in a flexibility perspective, producing models that can be used for forecasting. To account for the thermal preferences of individual occupants, we analyse the rich datasets collected in the living-lab and propose novel data-driven methods. We utilize the living-lab for novel studies of demand side management in a real-world setting. The research documents the potential of buildings as an important source of flexibility for the district heating systems in the quest for carbon-neutral heating supply. We show how context-aware control is a framework to move buildings towards a more prominent role in future energy systems, by enhancing the responsiveness towards both the energy systems context and the needs of the occupants. We conclude by proposing model-free reinforcement learning to train context-aware agents capable of controlling the technical systems in the buildings without the need for elaborate modelling efforts.