Sensitivity Analysis, Calibration and Uncertainty Quantification for Heavy Oil Viscosity Estimates with a Corresponding States Model

The main goal of this work is to assess heavy oil viscosity estimates by a Corresponding States Principle (CSP) model using a Bayesian approach in an efficient way. To determine and select relevant parameters for model calibration, an enhanced Elementary Effects method is used to evaluate sensitivity measures of CSP tuning parameters. With the combination of sensitivity analysis and Bayesian calibration, a unified procedure to automatically tune CSP viscosity model while reducing the number of tuning parameters is devised. Moreover, the Bayesian approach provides additional information on CSP model uncertainties and credible regions inherited from experimental data. To evaluate such uncertainties in CSP viscosity model, it was used five heavy oil samples available in the literature. The viscosity curves constructed by 50th-percentile from Monte Carlo realizations for the CSP calibration show good agreement when compared with classical Least-Squares regression (deterministic), demonstrating the potential of the sensitivity assessment for both Bayesian and deterministic approaches. However, when Bayesian calibration is used, limitations of CSP viscosity estimates are detected through violation of credible regions, suggesting that heavy oil viscosity estimates for relatively low pressure conditions can be insufficiently accurate for the CSP model considered in this study.