Statistical Model Identification and Variable Selection for Prediction of Heat Exchanger Fouling

Abstract

Fly ash generated during boiler combustion are carried away with hot flue gas and deposited over the heat exchanger surfaces. This study was motivated by the decline in the thermal efficiency of the heat exchangers in thermal power plants due to ash fouling. The paper discusses a new computational procedure that has been attempted to identify the best statistical model for predicting heat exchanger fouling and also to choose the most strongly linked covariates. In this context, three different statistical models including multiple linear regression, sliced inverse regression, and random forests regression are used to estimate the cleanliness factor of the heat exchanger. The variables and model are chosen solely on the basis of analytical procedures. Heat exchanger metal tube temperature that highly influences the cleanliness factor is selected using a measure of importance based on random perturbations. Metal temperature above a threshold is selected as a strongly linked covariate. The R package modevarsel is used here to perform the statistical analysis. The value of root mean square error for the cleanliness factor from sliced inverse regression and random forest is 0.0002, indicating that as more feasible in predicting ash fouling.