A Hybrid Model Approach to EOR Polymer Injection Pattern Ranking in a Mature Waterflood Field in South Oman

Abstract

Abstract In mature reservoirs, the sweep efficiency of waterflooding is frequently improved by polymer injection. The effectiveness of polymer injection can vary depending on the complexity of the reservoir and chosen injection pattern. As against the traditional reservoir simulation approach that is typically used to evaluate patterns, a novel approach is presented to combine connectivity (obtained from production/injection data history) and remaining oil saturations. This novel workflow evaluates polymer injection patterns and predicts their impact on oil production. This study proposes a hybrid model that combines reduced physics models - Capacitance Resistance Model (CRM), the Buckley Leverett (BL) oil model and machine learning model (ML) model to assess and rank polymer injection patterns for injection optimization strategies. More specifically, interwell connectivity between injection and production wells are obtained by combining physics-based reduced order models (CRM) and machine learning (ML) models. The hybrid combination of CRM and ML helps to provide confidence in results, eliminating shortcomings associated with just using CRM models. Finally, the results obtained from injection connectivity are combined with BL to model oil production rate to evaluate the impact of polymer viscosity on oil production. These first order oil prediction results are further used to rank patterns for polymer injection. The proposed workflow is applied to a complex, large waterflood field in Oman with more than 150 production and 43 injection wells. The hybrid combination of CRM and ML models used to obtain interwell connectivity compares well with tracer data resulting in higher confidence in the results. Using ML model, results obtained from multiple signals are aggregated to further identify and verify the injector-producer pair connectivity. The combination of CRM with BL model helps to evaluate additional oil recovery based on the changing viscosity of the displacing fluid. A combination of remaining oil saturation maps of producers and the impact of displacing fluid viscosity from the model helped rank 43 polymer injection patterns quickly and accurately. Hybrid models that combine CRM and ML to obtain connectivity address shortcomings of individual models such as CRM. The combination of interwell connectivity with remaining oil saturations to rank patterns for polymer injection is unique. The identification of waterflood patterns to choose for injecting polymer by novel use of hybrid models accelerates decision making and is directly applicable as part of Well and Reservoir Management (WRM) for large mature fields in the Middle East with waterflood in patterns and especially, where polymer injection is being considered. The presented approach can accelerate reservoir surveillance and reservoir management decision making as compared to traditional reservoir simulation workflows.