PAT soft sensors for wide range prediction of key properties of diesel fuels and blending components for the oil industry

Abstract

In an oil refinery, all decisions about the immediate destiny and processing of oil streams and blended products depend on the preliminary determination of their properties. These are obtained through standard methods, which are time consuming, expensive and require specialized personnel. Furthermore, their outcomes only become available after a considerable delay, significantly affecting plant activities. In this work, we present soft sensors based on Process Analytical technology (PAT) for accurately predicting eleven critical diesel properties in a wide variety of diesel fuel fractions and blends collected from an industrial refinery. The soft sensors were developed from FTIR-ATR spectra, through the optimization of preprocessing, band selection, spectral resolution tuning and bilinear modeling, under real plant operating conditions (GALP oil refinery plant). Quality parameters are estimated in a short cycle time with comparable accuracies to reference methods, enabling faster response times and decision making, while lowering the experimental overload in the laboratory.