Prediction of molar fractions in two-component gas mixtures using pulse-echo ultrasound and PLS regression

Abstract

The composition, and thus the energy content and monetary value of natural gas and biogas, vary considerably depending on the source. Present energy measurement techniques are not suitable for online use. We show with experiments on mixtures of ethane (C2H6) and oxygen (O2) how partial least squares regression (PLSR) can be used to predict the molar fraction of ethane in the mixtures, given spectral data from ultrasonic pulse-echo measurements. The PLSR technique is compared with the standard principal component regression (PCR), and we show that PLSR yields better predictive performance.