PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Spatial statistical calibration of CFD modelling of street canyons flows
Serge Guillas, Nina Glover and Liora Malki-Epshtein
In: 5th International Building Physics Conference, 28-31 May 2012., Kyoto, Japan.

Abstract

CFD methods are increasingly used to model wind flows around buildings in street canyons. The k-epsilon turbulence model is a popular choice in CFD modelling due to its robust nature and the fact that it has been well validated. However it has been noted in previous research that the k-epsilon model has problems predicting flow separation as well as unconfined and transient flows. The k-epsilon model contains empirical model constants whose standard values are sometimes empirically adjusted to the situation being modeled. Here, we calibrate these constants jointly against wind tunnel observations of turbulent kinetic energy distributed over a two-dimensional cross section of a regular street canyon. A Latin hypercube design covers the space of parameters. To overcome the prohibitive computational burden of calibration against all the observations over space, we use a sequential design to pick locations. We employ a fully Bayesian calibration framework. As a result, we narrow down the set of parameter values that provide the best match between the CFD model outputs and the observations, and quantify the uncertainty of the turbulent kinetic energy outputs resulting from both uncertainties in the CFD parameterization and observation errors.

EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
Learning/Statistics & Optimisation
Theory & Algorithms
ID Code:8842
Deposited By:Serge Guillas
Deposited On:21 February 2012