Abstract
Rigorous flow simulations to obtain estimates for recovery are infeasible given many combinations of reservoir and development scenarios. This motivates an alternative approach to calibrate likelihood of recovery using reservoir datasets. A proxy model relating reservoir, well, and operational factors to the ultimate recovery factor could guide subsequent field-scale flow simulations. There are two reservoir datasets that we have used for the classification and estimation of ultimate recovery: the Tertiary Oil Recovery Information System (TORIS) database for oil reservoirs and the Gas Information System (GASIS) database for gas reservoirs.
For oil reservoirs, 19 predictor variables were used to estimate the ultimate recovery factor. Cluster analysis was followed by linear regression analysis within the identified clusters that provided a reliable deterministic model for predicting the recovery factor. The linear regression model was compared with the empirical correlations given by Arps et al. (1967) and Guthrie et al. (1995). Analysis showed that geological and engineering parameters are correlated and both are important to the prediction of recovery factor. Later, we used a naive Bayesian approach on principal scores of the predictors for the calibration of recovery factor likelihood. The likelihood function of the recovery factor for oil reservoirs provided the uncertainty in recovery. It was computed to be multimodal and non-Gaussian.
In case of the gas reservoirs, good deterministic estimation was achieved by introducing cluster analysis to the data. The robustness of the linear regression model was tested on tight gas reservoirs. The likelihood of recovery factor was multimodal and non-Gaussian. It also indicates that gas reservoirs are less complex.