Comparative analysis of model reduction methods applied to building simulation benchmarks

Jordan Gauvrit (1) , Guillaume Ansanay-Alex (1)

1 CSTB - Centre Scientifique et Technique du Bâtiment

Many research and development studies use virtual buildings represented by building thermal models at a detailed level. Such models need relatively long computation times for one-year simulations: the use of these tools is then often too time-consuming when it comes to compare technical solutions during the design stage. Model reduction methods allow models with shorter computation times to be synthesized. In this paper, the linear state-space representation of a whole model is extracted and a balanced truncation method is applied to it. The detailed models are built from the SIMBAD library, a Simulink library of building modeling components developed at CSTB. Both linear state-space and reduced order models ensure shorter computation times than the full detailed model. However, the choice of the order of the reduced model has an impact on the final results. The main strengths and weaknesses of using the linear state-space and reduced models built from the same detailed model are investigated. To this end, the physical descriptions of idealized test buildings provided by ASHRAE standard 140 for building simulation tools assessment are used, and the results obtained through the different approaches for computation time reduction are compared.

Mots clés: thermal model; model order reduction; computation time; spectral analysis

Domaines: Sciences de l'ingénieur [physics]

Jordan Gauvrit, Guillaume Ansanay-Alex. Comparative analysis of model reduction methods applied to building simulation benchmarks. SSB2014, 9th International Conference on System Simulation in Buildings, Dec 2014, Liège, Belgium. 20 p. ⟨hal-01102181⟩