Emission monitoring using multivariate soft sensors

For combustion processes, it is important to monitor gases such as NO, in exhaust streams. Traditional approaches for such emission monitoring use analytical instruments, which are usually very expensive to install. Soft sensor techniques can provide a lower cost alternative to analyzers. In this paper we discuss using neural network partial least squares (NNPLS) and nonlinear principal components analysis (NLPCA) to build soft sensors for emission monitoring using data from an industrial heater. Several issues which are very important for the soft sensor approach are discussed, such as variable selection, sensor validation, and missing sensor replacement.