predictive-and-adaptive-heating-controller
Project leader: Nicolas Morel Collaborator: Manuel Bauer, Laurent Deschamps
Partners: CSEM (Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, coordinator of the project; LESO-PB/EPFL, Lausanne; Sauter Inc, Basle; Estia Inc, Lausanne
Duration and funding: 1997-2000, OFEN/BEW (Federal Office of Energy, Switzerland ), Sauter Inc.
Summary
Predictive and adaptive heating controller using a weather prediction elaborated by artificial neural networks. In order to find the optimal command sequence for the next time horizon (typically 6 or 12 hours), a dynamic programming algorithm finds the minimal cost function integrated over the time horizon. The cost function takes into account both the thermal comfort (when the user is present) and the heating energy used. The control algorithm is adaptive, in the sense that the artificial neural networks used for the weather prediction and the building model are progressively adapted to the real climate and the real building characteristics, as measured by the various sensors. Schema.
| Algorithms used |
Artificial neural network model for the weather data (ambient temperature and solar radiation) and the building characteristics (indoor air temperature in function of the heating power history), dynamic programming |
| Work carried out |
Phase 1 –> control algorithm development, simulation study, comparative measurements on two rooms of the LESO building Phase 2 –> slight modifications of the control algorithm, industrialisation study, realisation of a prototype, measurements on a residential building in Basle . |
| Results | The experimental checks have shown an energy saving of about 15 % for the heating, compared to a conventional heating controller, while providing a similar or better thermal comfort. Final report Phase 1. |
Publications
Printed copies of the final report available from leso-pb[at]epfl.ch
Subsequent publications in this field:
PhD Thesis David Lindelöf (2007): Bayesian optimization of visual comfort. http://library.epfl.ch/theses/?nr=3918
PhD Thesis Antoine Guillemin (2003): Using genetic algorithms to take into account user wishes in an advanced building control system http://library.epfl.ch/theses/?nr=2778
Paper Solar Energy Journal:
Morel, Nicolas; Bauer, Manuel; El-Khoury, Mario; Krauss, Jens, Neurobat, a Predictive and Adaptive Heating Control System Using Artificial Neural Networks, Solar Energy Journal vol. 21, p.161-201, 2001.