Deep reinforcement learning controller runs a real home heat pump after digital-twin training
An arXiv case study reports training a DRL agent in a digital twin and deploying it in an occupied residential building without online fine-tuning.
In plain words
Researchers tested an AI controller that learns by trial and error to run a home's underfloor heating and heat pump. They first trained it in a computer replica of the building, then installed it in the real house. It kept rooms within the set comfort range while adjusting water temperature and valve settings.
What was announced
Controlling underfloor heating paired with a heat pump is difficult because buildings have high thermal inertia and these systems respond slowly, the paper's abstract states. Deep reinforcement learning, a data-driven method, can in principle optimize energy use and thermal comfort at the same time, but deployments in occupied buildings under real conditions have remained limited. The authors addressed this gap with an end-to-end case study covering development, evaluation, and real-world deployment of a controller for an integrated underfloor heating and heat pump system.
The workflow began with a simulation-based pilot comparing different DRL algorithms, control strategies, and configurations. The team then built a digital twin of a real residential building and used it to train agents under simulated operating conditions. The best-performing agent was deployed in the physical building without additional online fine-tuning. According to the results, the controller maintained indoor temperatures within prescribed comfort limits while dynamically adjusting the heat pump supply temperature and valve positions. The authors say this demonstrates the feasibility of such deployments and highlights the challenge of transferring control policies from digital environments to real heating systems.
What it does not show
- The source is the paper's abstract only; detailed results, metrics, and study duration are not available in the supplied passages.
- The abstract reports the authors' own findings, and no independent confirmation of the deployment results is provided.
- Quantitative energy savings or comfort performance figures are not included in the supplied passages.
For security and IT teams
- Scope
- The reported deployment concerns a single residential building with an integrated underfloor heating and heat pump system, as described in the case study.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.