Bso 12 Acceleration of Building Design Optimisation through the Use of Kriging Surrogate Models

This paper describes an experiment to test the performance of Kriging surrogate modelling optimisation techniques on a building design problem with discrete design choices. Surrogate modelling optimisation offers advantages over traditional optimisation techniques on design problems with expensive (time consuming) performance evaluation models. The techniques are tested for both single and multi-objective optimisation problems with the objective of minimising both annual CO2 emissions predicted by a dynamic simulation and construction cost. The estimated CO2 emissions and costs of all possible designs were first established through comprehensive analysis using a multi-processor computer, enabling the performance of the optimisation to be assessed precisely against a known single optimum or Pareto front. The performance is compared against an evolutionary algorithm (EA) searching the dynamic simulation model on the same design problem. The results show that for this design problem, Kriging surrogate modelling optimisation is effective at finding estimates of optimum designs. In the case of the single-objective optimisation it is able to find the optimum in fewer simulation calls than the stand-alone EA. In the case of the multi-objective optimisation it is capable of finding a better Pareto front if the total number of simulations is restricted, although the time cost associated with Kriging does not always mean it is worth using.