A hybrid metamodel-based method for quick energy prediction in the early design stage

Abstract

It is generally acknowledged that the decisions made in the early design stage of a building critically affect its performance. To promote sustainability, building simulations are widely used in the early design stage to calculate the building performance and optimize the design. However, the high computational burden limits the application of building simulations in daily design work. This paper proposes a hybrid metamodel-based method to facilitate the rapid assessment of a building's energy demand in the early design stage. This method (1) decomposes a complex building mass into several zones as different metamodel variants and characterizes the total energy demand by summing up the energy demands of all the metamodels; (2) uses the received solar radiation values as metamodel input parameters to describe the surrounding shading and reflection effects; (3) employs GPU acceleration technology to accelerate the radiation calculation and reduce the simulation time; and (4) implements a machine learning (ML) algorithm screening framework to enhance the accuracy of the metamodels. Case studies were conducted to demonstrate the proposed method. The results showed that this method could predict the energy demands of buildings in high-density urban environments with acceptable accuracy in a short period of time, which would allow designers to obtain feedback on the building energy demand immediately in the early design stage and opens up more possibilities for achieving low-energy buildings.