We applied data-based recurrence CFD (rCFD) to model pollutant dispersion in near-field flow configurations. In case of complex topologies, the global-domain version of rCFD fails to account for local recurrent flow features. We therefore developed a novel island-based version of rCFD, which partitions the computational domain to isolate islands of high recurrence prominence, and subsequently defines a distinct recurrence path for each of these islands. We applied island-based rCFD to pollutant dispersion for two side-by-side cubical buildings with three different gap widths in between them and a real urban environment. We showed that numerical predictions of pollutant dispersion by island-based rCFD were in excellent agreement with full CFD simulations, thus outperforming the global-domain version of rCFD. In both applications, island-based rCFD simulations ran three orders of magnitude faster than corresponding full CFD simulations. In the second application, this speed-up enabled real-time simulations on a computational grid of 10 million cells.

我们应用基于数据的递归 CFD (rCFD) 来模拟近场流配置中的污染物扩散。在复杂拓扑的情况下，rCFD 的全局域版本无法考虑局部循环流特征。因此，我们开发了一种新的基于岛的 rCFD 版本，它划分了计算域以隔离高复发突出的岛屿，并随后为这些岛屿中的每一个定义了不同的复发路径。我们将基于岛的 rCFD 应用于两个并排的立方体建筑的污染物扩散，它们之间具有三种不同的间隙宽度和真实的城市环境。我们表明，基于岛的 rCFD 对污染物扩散的数值预测与完整的 CFD 模拟非常吻合，因此优于 rCFD 的全局域版本。在这两个应用程序中，基于岛的 rCFD 模拟的运行速度比相应的完整 CFD 模拟快三个数量级。在第二个应用程序中，这种加速支持在 1000 万个单元的计算网格上进行实时模拟。