Development of K- Means Based SVM Regression (KSVMR) Technique for Boiler Flue Gas Estimation

Abstract

This paper presents development of a Support Vector Machine (SVM) regression, driven by a Radial Basis Function kernel for obtaining the composition of boiler flue gas mixtures. The frequency components of various gas mixtures were first processed by Floyd K – Means algorithm and the data with class labels were utilized to build a multi-class SVM regression model for discrimination of the flue gas constituents and subsequent composition finding. The Meta parameters (C, e and kernel) are optimized using grid search technique to obtain appropriate support vectors to train the network. After ascertaining the performance of proposed technique through volatile organic component (VOC) data acquired from quartz crystal microbalance (QCM) type sensors used by earlier researchers, detailed studies have been carried out to study the discriminating and estimation capability of the proposed technique for real time flue gas data acquired from two different analyzers namely ORSAT and KANE. Exhaustive studies clearly indicate the exceptional performance of the proposed SVM model in classifying and estimating the flue gas components in machine (Analyzer) independent manner.