Domestic heating accounts for 78% of residential energy use. In addition, domestic heating is becoming increasingly electrified (using heat pumps), in part driven by a political drive to wean off carbon-based energy sources. This poses two challenges: (1) electricity production from renewable sources fluctuates, and so does the price, and (2) current day electricity grids are limited in capacity.
In this talk, Peter proposes a combination of digital-twin based technologies (model estimation and controller learning) used in tandem to reduce consumer cost. Importantly, consumer cost is dominated by the efficiency of the heat-pump, the energy price, and the energy grid transport cost – thus a reduction of cost alleviates the two posed challenges.
For a single-family house, a predictive digital twin of the house can be inferred from measured data. This digital twin estimation process provides the necessary component for doing model-based reinforcement learning to (safely) train a controller to reduce energy costs. By repeatedly doing model estimation and controller synthesis, Peter demonstrates how consumer flexibility and substantial energy cost savings (up to 40%) can be achieved.
Peter Gjøl Jensen
Peter Gjøl Jensen is an Assistant Professor at the Department of Computer Science, Aalborg University
How does one create safe, optimal, and robust controllers for cyber-physical systems? This has been the focus of Peters research – a line of work that entails both theoretical, academic, and recently practical applications of state-of-the-art technology for automatic controller construction.
Currently he is migrating his state-of-the-art methodology for heat-pump control from controlled laboratory setting towards commercialization and general adoption via a InnoExplore grant from Innovation Fund Denmark.
The aim is to improve the efficiency of heat-pump control and unleash the flexibility potential of domestic heating systems with a clear aim of reducing (1) consumer cost, and (2) the Co2 footprint of residential heating systems.