Implementation of a coupled simulation framework with neural network and Modelica for fast building energy simulation considering non-uniform indoor environment

The coupling of computational fluid dynamics (CFD) and building energy simulation (BES) program enables the simultaneous consideration of the indoor environment quality and building energy performance in architecture design. Meanwhile, it improves the energy simulation accuracy for non-uniform indoor environments as compared to a standalone BES. However, a CFD-BES coupled simulation confronts a great challenge in that conducting high-fidelity CFD simulations is time-consuming, making it impractical for scenarios requiring a large number of CFD executions. In this study, an innovative coupled simulation framework is proposed and implemented for fast energy simulation considering non-uniform indoor environments. As a surrogate for CFD, a neural network (NN) for achieving a fast and accurate prediction of the indoor air distribution is coupled with a Modelica-based energy simulation program. The crucial functionality of the coupled simulation framework is validated against the results of a coupled simulation between fast fluid dynamics (FFD) and Modelica. Furthermore, the framework is tested with a more complicated case to explore its feasibility and potential for practical applications. Compared to a standalone BES, the coupled simulation considering a non-uniform environment demonstrates more reasonable results regarding heat transfer calculations and system operations. Moreover, the proposed framework requires only 52 s to execute a simulation with a physical process of 24 h, resulting in an approximately 94% reduction in calculation time relative to a coupled simulation between CFD and BES.

Automated summary

The coupling of CFD and BES enables simultaneous consideration of indoor environment quality and building energy performance in architecture design. In this study, an innovative coupled simulation framework using a neural network for fast and accurate prediction of indoor air distribution is proposed. The framework demonstrates more reasonable results and significantly reduces calculation time compared to traditional coupled simulation methods.