Data-driven technique estimates skin factor and average pressure during oil-flowing periods

Abstract

We propose a data-driven workflow to address the problem of estimating skin factor, reservoir average pressure and productivity index during well production, in opposition to the conventional shut-in dependent approach. To that end, we demonstrate that bottom-hole pressure (BHP) and liquid flow-rate recorded during production form a two-dimensional feature space for regressions whose parameters are learned from past shut-ins. The proposed models learn how to map from BHP and flow-rate to skin factor, via non-linear mapping, and average pressure, via linear mapping, as they are trained using a few previous well shut-ins. Once trained, the regressors can be used during the production period to estimate skin factor and reservoir pressure. The theoretical groundings for the data-driven method are presented. We discuss the regression model limitations caused by the non-stationarity of reservoir and operational conditions and propose guidelines on model training so as to have the models adapt as the system evolves. The application of the method is demonstrated in field examples featuring non-stationary reservoir and operating conditions. The results suggest that regressors can be used to produce skin and reservoir pressure estimates during production, while a single shut-in can help update the function mapping inputs to prediction after changes in reservoir and operating conditions take place. Additionally, we show how the method can be used to detect anomalies in the wellbore. Practicing engineers involved in reservoir management and PDG monitoring for the purpose of formation damage control in oil producers in developed fields might find value in the content presented in this work.