Department of Building Science, School of Architecture, Tsinghua University, Beijing, China.
Institute of Industrial Science, University of Tokyo, Tokyo, Japan.
J Hazard Mater. 2018 May 5;349:68-78. doi: 10.1016/j.jhazmat.2018.01.050. Epub 2018 Jan 31.
The estimation of source information from limited measurements of a sensor network is a challenging inverse problem, which can be viewed as an assimilation process of the observed concentration data and the predicted concentration data. When dealing with releases in built-up areas, the predicted data are generally obtained by the Reynolds-averaged Navier-Stokes (RANS) equations, which yields building-resolving results; however, RANS-based models are outperformed by large-eddy simulation (LES) in the predictions of both airflow and dispersion. Therefore, it is important to explore the possibility of improving the estimation of the source parameters by using the LES approach. In this paper, a novel source term estimation method is proposed based on LES approach using Bayesian inference. The source-receptor relationship is obtained by solving the adjoint equations constructed using the time-averaged flow field simulated by the LES approach based on the gradient diffusion hypothesis. A wind tunnel experiment with a constant point source downwind of a single building model is used to evaluate the performance of the proposed method, which is compared with that of the existing method using a RANS model. The results show that the proposed method reduces the errors of source location and releasing strength by 77% and 28%, respectively.
从传感器网络的有限测量中估计源信息是一个具有挑战性的反问题，可以将其视为观测浓度数据和预测浓度数据的同化过程。在处理建成区的排放时，预测数据通常通过雷诺平均 Navier-Stokes（RANS）方程获得，该方程可提供建筑物分辨率的结果；然而，在气流和扩散的预测方面，基于 RANS 的模型不如大涡模拟（LES）。因此，探索通过 LES 方法改进源参数估计的可能性非常重要。本文提出了一种基于 LES 方法和贝叶斯推断的新源项估计方法。源-受体关系是通过求解基于 LES 方法的时均流场的伴随方程获得的，该 LES 方法基于梯度扩散假设。利用单点源下风的单个建筑物模型的风洞实验来评估所提出方法的性能，并将其与使用 RANS 模型的现有方法进行比较。结果表明，该方法分别将源位置和释放强度的误差降低了 77%和 28%。
