A Novel Approach to Combine Models to Evaluate Interwell Connectivity in a Waterflooded Reservoir with Limited Injection History

Abstract Waterflooding is one of the most widely implemented enhanced recovery in mature oil fields. In the absence of a reliable reservoir model, waterflood optimization can be a challenge. The availability of continuous recording of production, injection and well data can be utilized to improve reservoir management in this novel approach. This study presents a new approach using Machine Learning (ML) technique through multiple signal analysis to optimize waterflood operation in a brownfield offshore Caspian Sea. To evaluate injection efficiency on oil production, firstly the interwell connectivity between injectors and producers are determined. However, because of the complexities associated with the reservoir and the data, it has been achieved through analyzing various available signal types which are informative and responsive to injection rates. Results obtained from multiple signals are then aggregated to identify the injector-producer pair connectivity. Next, production well performances are evaluated through multiple diagnostic models. Finally, the impact of injectors on oil production rates are analyzed and injector efficiencies are determined to establish a more efficient waterflooding strategy. The proposed methodology has been applied to a reservoir with around 50 producers and 7 injectors. The interwell connectivity between pairs have been identified and ranked. Using data analytics techniques on multiple surveillance data sets, the analysis of the waterflood is achieved more swiftly and accurately. It was observed that for this specific case, the most informative signals that help determine connectivity are the water cut, and water production rate. The identified injector-producer connections obtained from these models were further verified and compared well with additional available surveillance data on tracers for this reservoir. Understanding these leads to devising optimum waterflooding strategies such as diverting more injection water to the more efficient injectors and less injection water to the inefficient injectors. A novel multi-signal analysis using ML techniques is proposed that combines multiple data being collected as part of surveillance. The presented approach can be extended to similar waterfloods to help with optimizing the waterflooding strategy. This new approach helps with current digitization strategies in oil companies that seek to obtain faster and consistent solutions to accelerate decision making and as an alternative to cases especially where reservoir model is poorly defined.