Building and Environment ( IF 7.4 ) Pub Date : 2021-02-23 , DOI: 10.1016/j.buildenv.2021.107728 Ruijun Zhang , Parham A. Mirzaei
Coupling of building energy simulation (BES) tools with computational fluid dynamics (CFD) technique offers the ability to include the commonly neglected, but significantly important, neighbourhood effect on the local airflow patterns and thus buildings’ energy demand. Amongst various coupling approaches, the fully dynamic coupling is considered as the most accurate technique although not a practical one for medium-to-long-term simulations due to the associated high computational cost.
This study, therefore, aims to propose a novel framework of virtual dynamic BES-CFD-artificial intelligence (AI) coupling to prevent intensive computational calculations. The prediction is performed by artificial neural network (ANN), which is trained over a series of fully dynamic BES-CFD coupling results to replace the local flow characteristics, in particular, convective heat transfer coefficient (CHTC). Furthermore, a case study of a city block performed in a typical hot month (September) in Los Angeles is undertaken to assess the proposed framework.
The predictions of the local CHTCs on the external surfaces are found satisfactory with an accuracy of 0.88. Moreover, 10 is found as the effective size of days to train the neural network tools for a one-month simulation. The proposed approach results in saving approximately 2/3 of the required computational time using an ordinary approach.
中文翻译:
计算流体动力学的虚拟动力耦合-建筑能耗模拟-人工智能:城市邻里对建筑能耗的影响案例研究
建筑能量模拟(BES)工具与计算流体力学(CFD)技术的耦合提供了将通常被忽略但非常重要的邻域效应包括在局部气流模式以及建筑物的能源需求中的能力。在各种耦合方法中,全动态耦合被认为是最精确的技术,但由于相关的高计算成本,因此对于中长期仿真而言并不实用。
因此,本研究旨在提出一种虚拟动态BES-CFD-人工智能(AI)耦合的新型框架,以防止进行密集的计算。预测是通过人工神经网络(ANN)进行的,该人工神经网络通过一系列全动态BES-CFD耦合结果进行训练,以替代局部流动特性,特别是对流热传递系数(CHTC)。此外,还对洛杉矶典型的炎热月份(9月)进行的一个城市街区进行了案例研究,以评估拟议的框架。
发现外表面上局部CHTC的预测令人满意,准确度为0.88。此外,发现10天是训练神经网络工具进行一个月仿真的有效天数。所提出的方法使用普通方法可以节省大约2/3的所需计算时间。