Source term estimation in complex urban environments based on Bayesian inference and unsteady adjoint equations simulated via large eddy simulation
Source term estimation in complex urban environments based on Bayesian inference and unsteady adjoint equations simulated via large eddy simulation
复制标题
DOI：
10.1016/j.buildenv.2021.107669
复制
发表时间：
2021-02-12
影响因子：
7.4
通讯作者：
Kikumoto, Hideki
中科院分区：
文献类型：
--
作者：
Jia, Hongyuan;Kikumoto, Hideki
建立准确的源-受体关系对于在复杂的城市环境中识别未知的空气污染源至关重要。现有的源项估计（STE）方法确定这种关系，通过稳定的模拟伴随方程与集成平均流场。然而，湍流扩散对污染物扩散的影响被简化为梯度扩散假设，这可能会导致相当大的STE误差，特别是在复杂的城市地区。因此，我们开发了一种新的STE方法嵌入非定常伴随方程建模通过大涡模拟（LES）到贝叶斯推理框架。非定常模拟的巨大的存储需求，减轻了使用基于小波的压缩方法。所提出的方法的性能进行评估的基础上收集的分散测量在风洞实验中的规则，块阵列的建筑群模型。并与基于平均LES流场的伴随方程稳态模拟方法进行了比较。我们的研究结果表明，基于LES的方法显着提高了建模精度的伴随方程和由此产生的STES。如果后验概率的第50百分位数被视为点估计，在源位置和强度的绝对误差分别减少了89%和99%，与现有的方法相比。
Establishing an accurate source-receptor relationship is essential for identifying unknown sources of air pollution in complex urban environments. Existing source term estimation (STE) methods determine this relationship via steady simulations of adjoint equations with ensembled-averaged flow fields. However, the effect of turbulent diffusion on the dispersion of pollutants is simplified according to the gradient dispersion hypothesis, which may result in considerable STE errors, especially in the complex urban areas. Therefore, we developed a new STE method by embedding unsteady adjoint equations modeled via large eddy simulation (LES) into a Bayesian inference framework. The tremendous storage requirement of the unsteady simulation was mitigated using a wavelet-based compression method. The performance of the proposed method was then evaluated based on dispersion measurements collected in a wind tunnel experiment for a regular, block-arrayed building group model. The estimation results were compared with those derived from an existing method, in which the steady simulation of adjoint equations was employed based on the mean LES flow field. Our results suggest that the LES-based approach significantly improved the modeling accuracy of the adjoint equations and the resulting STEs. If the 50th percentile of the posterior probability was regarded as the point estimate, the absolute errors in source location and strength were reduced by 89% and 99%, respectively, when compared to the existing method.