Active Learning in Multi-objective Evolutionary Algorithms for Sustainable Building Design

摘要: Residential and commercial buildings are responsible for about 40% of primary energy consumption in the US. The design a building has tremendous effect on its profile, recently there been an increased interest developing optimization methods that support high performance buildings. Previous approaches either based simulation or training accurate predictive model to replace expensive simulations during optimization. We propose method, suitable multiobjective very large search spaces. In particular, we use Gaussian Process (GP) prediction devise active learning scheme multi-objective genetic algorithm preferentially simulate only solutions informative model's predictions current generation. develop comprehensive publicly available benchmark show GP is highly competitive as surrogate simulations, addition being well-suited setting. Our results our approach clearly outperforms surrogate-based optimization, produces close hypervolume while using fraction time.