A comparative study of dynamic building simulation and machine-learning within a two-stage multi-objective stochastic optimization framework

To tackle increasingly serious environmental issues and satisfy different stakeholders, trade-offs between different goals should be considered in the building planning phase. However, especially during early stages of decision-making, uncertainties are inevitable and should be modeled. Moreover, in real-world decision-making processes, decisions are often not made at the same time. Thus, it is of interest to model the trade-off-aware decision-making process that takes uncertainties and different decision stages into account in building planning. In this paper, we conducted a comparative analysis of machine-learning-based and dynamic-building-simulation-based methodologies of generating operational energy within a multi-objective stochastic optimization (MOSO-II) framework. Results show that the Pareto-optimal solutions with the machine-learning model can withstand a higher degree of uncertainty, which can be explained by higher use energy amount and a larger uncertainty range in the applied dataset caused by occupant behavior or other effects.