Adaptive Sampling For Building Simulation Surrogate Model Derivation Using The LOLA-Voronoi Algorithm

Statistical surrogate models, or meta-models, are used to emulate building simulation models. Their key advantage is the reduction of computational cost. This in particular matters if building design analysis demands to explore a large number of different building designs options as in optimization or uncertainty analysis problems. To derive a surrogate model, a data set consisting of simulation in- and output data is generated. This set is then used to train the surrogate. This process of collecting simulation data may be time intensive and a building designer has to wait until surrogate model is available. In this study we construct a global surrogate model using adaptive sampling to speed up the data collection. In comparison to static sampling, it balances both exploration of the design space while exploiting the iteratively growing information of simulation outcomes. The advantage of adaptive sampling is not only that it can cut simulation time, but also that it rapidly provides a preliminary low-accurate surrogate to the building designer which is sequentially improved while he/she is working with the low accuracy model already.