Abstract
Inferential or soft sensing techniques have been gaining momentum recently as viable alternatives to hardware sensors in various situations, e.g. continuous emission monitoring systems. Dynamic neural networks are used in the present work to develop soft sensors for the NOx and O-2 emission due to combustion operation in industrial boilers. A simplified structure for the soft sensor is obtained by grouping the input variables, reducing the input data dimension and utilizing the system knowledge. The principal component analysis (PCA) is used to reduce the input data dimension. The genetic algorithm (GA) is used to estimate the system's time delays by optimizing a linear time-delay model. Real data from a boiler plant is used to validate the models. The performance of the proposed dynamic models is compared with static neural network models. The results demonstrate the effectiveness of the proposed models. (C) 2008 Elsevier Ltd. All rights reserved.