Estimation of airflow distribution in cubic building group model using POD-LSE and limited sensors

Abstract

Urban airflow has strong turbulence characteristics, and its distribution varies with time. Urban airflow distribution can be measured with sensors or simulated using Computational Fluid Dynamics (CFD). However, CFD is considered still expensive for the rapid calculation of unsteady airflow distribution and the measurement value of sensors is often inferior in terms of spatial resolution. In this study, the high spatial resolution airflow distribution in the cubic building group model was rapidly estimated from limited sensors by combining Proper Orthogonal Decomposition (POD) with Linear Stochastic Estimation (LSE). Validated airflow data simulated by Large-Eddy Simulation (LES) in a cubic building group model were selected as the airflow database. This study was conducted with two main objectives: 1) to verify the basic performance of the POD-LSE; and 2) to analyze the effects of changing the number of sensors and POD analysis regions on the estimation accuracy. By comparing the results of the POD-LSE estimations with the LES data as the true value in the reference region, the following conclusions were obtained: 1) The airflow distribution in the cubic building group model can be reconstructed and predicted by the POD-LSE; 2) nine sensors can provide acceptable estimation accuracy; 3) the velocity value of a point located in the wake region is difficult to estimate; 4) the accuracy does not increase as the POD region expands (without changing the number of sensors); and 5) with an increasing number of sensors, the accuracy of the velocity estimation is improved in reconstruction.