Stable reduced-order models for pollutant dispersion in the built environment

Abstract

In order to limit the impact of accidental releases of hazardous pollutants into the atmosphere, there is a need for an accurate near-range atmospheric dispersion modeling approach that is suitable for on-line risk management. Computational Fluid Dynamics (CFD) has proven to be a promising tool for atmospheric dispersion studies at the near-range. However, the relatively long computing times currently prohibit the use of CFD for real-time purposes. Therefore, we present in this work an effective model reduction method that is based on the projection of the original model, which solves the transient advection-diffusion equation on a steady background velocity field, onto a Krylov subspace by means of the Arnoldi algorithm. This allows to construct a reduced order model (ROM) from an accurate CFD model that is guaranteed to be stable. The algorithm is formulated in such a way that the ROM can be derived using any CFD software package, commercial or non-commercial. The accuracy of the ROM is illustrated by performing a series of simulations of a time-dependent pollutant release at the Doel nuclear power station, located to the north of Antwerp (Belgium). A comparison between the results obtained using the ROM after initialization, and the original CFD model shows a reduction of a factor of 2500 in computational time, leading to a ROM that runs 25 times faster than real-time without a significant loss in accuracy. In terms of computational cost, the ROM is a factor 105 less expensive than the original CFD model.