Thermal building model identification using time-scaled identification methods

Auteurs: Paul Malisani, François Chaplais, Nicolas Petit, Dominique Feldmann, 49th IEEE Conference on Decision and Control, pp. 308-315, December 15-17, 2010, Atlanta, USA DOI: 10.1109/CDC.2010.5717975
The aim of this paper is to propose a robust and accurate method for the parametric identification of the thermal behaviour of low consumption buildings. These buildings are known to have a two-time scale structure, which, if not handled properly, results in poor conditioning of the parametric identification.
We compare three identification methods, one uses the data on the whole frequency domain (ARX) when the other methods use the same data but separated on local frequency domain (time scaled methods).
All three methods identify a reduced second order model. Robustness is tested by corrupting the input and output before the identification, and comparing the simulation results for the various models and the original uncorrupted input. The numerical results clearly show that the time scaled methods are superior both in accuracy (noise free identification and simulation) and robustness (when identification is performed on corrupted data).
Download PDF
BibTeX:
@Proceedings{,
author = {Paul Malisani, François Chaplais, Nicolas Petit, Dominique Feldmann},
editor = {},
title = {Thermal building model identification using time-scaled identification methods},
booktitle = {49th IEEE Conference on Decision and Control},
volume = {},
publisher = {},
address = {Atlanta},
pages = {308-315},
year = {2010},
abstract = {The aim of this paper is to propose a robust and accurate method for the parametric identification of the thermal behaviour of low consumption buildings. These buildings are known to have a two-time scale structure, which, if not handled properly, results in poor conditioning of the parametric identification.
We compare three identification methods, one uses the data on the whole frequency domain (ARX) when the other methods use the same data but separated on local frequency domain (time scaled methods).
All three methods identify a reduced second order model. Robustness is tested by corrupting the input and output before the identification, and comparing the simulation results for the various models and the original uncorrupted input. The numerical results clearly show that the time scaled methods are superior both in accuracy (noise free identification and simulation) and robustness (when identification is performed on corrupted data).},
keywords = {Buildings, Data models, Mathematical model, Numerical models, Optimization, Robustness, Temperature measurement}}