Waterflood Performance Diagnosis and Optimization Using Data-Driven Predictive Analytical Techniques from Capacitance Resistance Models CRM

Abstract

Abstract Quick evaluation of reservoir performance is one of the main concern in decision making. Time consuming data preparation and processing, and data uncertainty (geological, petrophysical, reservoir engineering) limit the use of numerical simulators in addition to long term response for reservoir management. Effective reservoir management needs quick action regarding injected fluid distribution to improve areal and vertical sweep efficiency during secondary and tertiary recovery processes. Therefore, simpler methods that provide quick results to complement or substitute reservoir simulation are important for reservoir monitoring and management. Capacitance Resistance Model (CRM) is one proven method to address the above challenges. The CRM model is based on the hypothesis that reservoir performance can be inferred from analyzing production and injection data and a simplified analytic model structure. CRM is an input-output and material balance based model that use continuity equation to quantify the connectivity among injector and producer wells, time constant and productivity indices. In this study, a CRM model was generated to build a forecasting waterflood model by to matching historic production and injection data. After history matching, the model can be used to: (1) effectively forecast oil and water production, (2) evaluate reservoir and production-injection data uncertainties, and (3) optimize injection settings to improve oil recovery factor. CRM methodology is used as a cooperative approach to reservoir simulator generated streamlines. Different case studies were designed to investigate the robustness of CRM by providing different levels of complexities as compared with numeric simulation results. In the first case, a constant flowing bottomhole pressure and injection rate was considered, and in the second case, changes to flowing bottomhole pressure, injection rate and productivity index per each time step were considered. The CRM was able to replicate the synthetic historic data generated using numerical simulator within a satisfactory error, i.e. less than 7% in the worst case. The CRM injector producer allocation factors were also in line to those computed using streamlines. In both cases, the CRM model showed a higher allocation factor for the injector producer pairs which had better connectivity allowing the evaluation of waterflood performance by pattern while identifying the poor sweep efficiency areas. In a later optimization exercise, a dry oil production objective function was maximized subject to the constraints in the injection system. Thus, strategies derived from CRM model increased production in the range of 12% with associated water cut reduction just by reallocating injection rates.