Application of Multivariate Statistical Models to Prediction of NOx Emissions from Complex Industrial Heater Systems

Industrial fired heaters are a major source of nitrogen oxides (NOx). On-line analyzers, kinetic models with computational fluid dynamics, and empirical predictive models have been developed to monitor NOx emissions and analyze excessive NOx emissions. However, previous approaches have been applied only to single heater systems, not to large-scale multiheater systems. This paper proposes a hierarchical monitoring and diagnosis procedure to monitor NOx emissions from large-scale multiheater systems and to identify the root causes of excessive NOx emissions as well as heater malfunctions. The procedure provides a functional three-layer hierarchy: (1) prediction of NOx concentrations; (2) estimation of the influence of individual heaters on the predicted NOx; and (3) identification of the root causes by examining the detailed contributions of process variables to variations of the heater identified in step 2 as being the principal source of NOx. An integrated multiblock partial least-squares (PLS) model, created by combining standard PLS and multiblock PLS, is employed to predict the NOx emissions and estimate the influences of the heaters on the emissions. The validity of the proposed method is demonstrated through its application in two case studies.