Carbon Dioxide (CO2) injection is one of the more effective EOR methods in oil reservoirs. Continuous CO2 injection is a viable technique that could yield high oil recoveries when employed effectively in naturally fractured reservoirs. Screening, evaluating, and designing optimized continuous CO2 injection projects require extensive computational power for the concerned reservoirs. Although reservoir fluid composition is the most important parameter when it comes to EOR, it can be particularly challenging to account for different reservoir fluid compositions in any workflow. In this study, a new workflow is developed that helps evaluate and design CO2 injection in naturally fractured reservoirs for different reservoir fluid compositions. Reservoir rock properties are also evaluated as part of this study to assess the impact on recovery. Continuous CO2 injection schemes with variable injection design parameters are constructed using a two-well, two-layer, dual porosity, miscible compositional reservoir simulation model. Representative scenarios are then used to train and develop two Artificial Neural Network (ANN) based proxy models: 1) A performance prediction proxy that predicts reservoir performance for a specific injection design, 2) An inverse ANN for injection applications that determines injection design parameters for the desired performance. ANN's predictability is evaluated and measured through its accuracy to predict blind testing data sets. The proposed proxies are found to be capable of predicting within an acceptable degree of error against the blind testing scenarios. The developed proxies have the potential of overcoming existing limitations and setting up a new benchmark for future studies.

You can access this article if you purchase or spend a download.