On-line fuel identification using optical sensing and Support Vector Machines technique

Abstract

In this paper, Support Vector Machines (SVM) technique was used to identify fuel types. Flame oscillation signal were captured by a three-cell flame monitor. Thirty flame features were extracted from each flame signal. Then Principal Component Analysis (PCA) was used to choose the principal components of each features vector that represent over 99 percent variations of the features vector. An SVM was deployed to map the principal components, size-reduced flame features, to an individual type of fuel. PCA can reduce the data dimension and ultimately the training time of SVM. The data of eight different types of coal obtained from a combustion test facility demonstrate that the SVM technique was effective for identifying the fuel types, and the average success rate was 96.1% in twenty trials.