INVERSE IDENTIFICATION OF MULTIPLE POLLUTANT SOURCES

Proceedings of BS2015: 14th Conference of International Building Performance Simulation Association, Hyderabad, India, Dec. 7-9, 2015.

INVERSE IDENTIFICATION OF MULTIPLE POLLUTANT SOURCES

Yun Wei, Tengfei(Tim) Zhang* ,

Shugang Wang School of Civil Engineering, Dalian University of Technology (DUT) 2 Linggong Rd, Dalian 116024, China

* Corresponding email: tzhang@dlut.edu.cn

ABSTRACT

In case of an accidental release of indoor airborne pollutants, it is critical to know information of the pollutant sources. Current inverse modeling concentrates on identifying a single pollutant source or multiple pollutant sources in simplified puff or constant release forms. This investigation proposes an inverse model to precisely identify the multiple pollutant source number, locations and their temporal release rate profiles simultaneously. The model implements Tikhonov-based inverse matrix operation to obtain the release rate profiles for each candidate source with temporal concentrations provided by sensors. A candidate source with a solved release rate close to zero is excluded as an active source, and then the pollutant source number and locations with nonzero release rate can be determined. The above strategy was applied to identify two sources released by passengers in a three-dimensional aircraft cabin. The results show that the proposed method is able to determine the source number, locations and their temporal rate profiles correctly.

INTRODUCTION

In case of an accidental release of indoor airborne pollutants, it is critical to know the number, locations and temporal release rates of the pollutant sources. Current sensors can provide pollutant concentrations locally but cannot tell where and how the pollutants have released. To obtain the above pollutant source information, a viable method is to conduct inverse modeling, i.e., to infer the sources from certain detected consequences. Inverse modeling in indoor environments is majorly concentrated on identification of pollutant source locations. The adopted strategies can be divided into two categories (Zhang et al., 2015), namely, directly reversing transport equations to track a source and solving forward transport equations to match a source. For example, Zhang and Chen (2007a) applied the first strategy and solved a quasi-reversibility (QR) equation to identify a single pollutant source location via directly reversing the time-marching direction of the pollutant transport equation. Bady et al. (2009) also implemented a similar reversed time-marching method in an urban environment. The QR method was further extended to locate a particulate source after accounting for particle settling effect due to the gravity (Zhang et al., 2012). Except for reversing the time-marching direction, solving the convective pollutant transport in a reversed flow field using the so-called pseudo-reversibility method may also locate a single pollutant source (Zhang and Chen, 2007b). The first strategy does not require much priori source information assuming as known but may suffer from numerical stability. Most researchers turn to a stable solution by applying the second strategy. If a candidate source is found providing concentration response highly comparable with a monitoring sensor, it is determined to be the source. Liu and Zhai (2008, 2009) solved the adjoint equation of pollutant source location probability to determine a single gaseous source location. Vukovic et al. (2010) and Bastani et al. (2012) applied neural network to match the pollutant source concentration to find a pollutant source. Wang et al. (2013) matched a source with the concentration response provided by the state-space matrix. The second strategy is stable but has a large computing burden in solving and matching concentration. All possible source information including the release location and temporal rate profiles must be assumed to be known in advance. In addition to pollutant source location, some studies also addressed identification of pollutant release rates. For a gaseous source released in an instantaneous form, a linear scaling method (Zhang and Chen, 2007b) is valid for determining the total pollutant release amount. However, for a dynamically released source, identification of its rate profile is more challenging. Zhang et al. (2013) proposed a matrix inversion method to determine the temporal release rate profile of a gaseous source with the concentration monitored at a point. However, the pollutant source location must be assumed to be known before inverse identification. For more comprehensive source information, it is necessary to identify both the pollutant source location and its temporal release rate profiles simultaneously. Sohn et al. (2002) applied a Bayesian model to predict the location and release amounts of a pollutant source. All candidate pollutant source locations, release forms and amounts were presumed to be known. Zhang et al. (2015) combined an inverse matrix method with a Bayesian probability model to identify a single pollutant source location, temporal rate profile and the sensor alarming time. The required inputs are concentrations at two different points in a space. Cai et al. (2013) presented a model to identify locations and emission rates of multiple sources, but the sources must be in the very simple constant-release types. To date, no inverse modeling was demonstrated to simultaneously identify the number, location and temporal release rate profiles of multiple dynamically-released pollutant sources. This investigation presents an inverse model for determination of the above source information with several temporal concentrations as inputting. A case with two sources dynamically released in a threedimensional aircraft cabin was solved for illustrating the methodology.

READ the ARTICLE --->