Multivariate Statistical Modeling of Crude Oil Viscosity and Mole Percent Components for Reservoir Fluids

Abstract

This study presents multivariate statistical modeling of PVT properties in a Hydrocarbon Reservoir. Traditionally, these properties are modeled by correlating them with the changes in pressure and temperature associated with them. However, these changes are often a direct corollary of the changing composition as each component’s presence and/or absence contributes to the overall property of the fluid system. In this work, Multiple Linear and Multiple Nonlinear regression models were used to develop a correlation for the crude oil viscosity and the mole percent of the components as they change with pressure and temperature. A fluid system was modelled using the PVT Package of Integrated Production System Modelling (IPM) Suite. The model was then used to perform a differential liberation test simulation to predict the changes in composition of the crude oil. The composition included three lumped crude oil fractions. After the PVT modelling, the generated PVT Composition was used to perform the Multivariate statistical analysis (MVA) and modelling.Two MVA techniques were compared – the Multiple Linear Regression (MLR) and Multiple Non-Linear Regression (MLNR).Analysis of residuals generated from the prediction run based on both techniques showed that the multiple linear regression method had trending residuals, contrary to the law of parametric estimation upon which it is based. However, the multiple nonlinear regression yielded 100% correlation with adequate residual trend and is thus recommended for adaptation to distinct fluid systems. It is imperative to note that the accuracy of this adaptive PVT modeling and prediction is hinged primarily on the accuracy of the PVT model used to estimate the compositional variation.