Screening and optimization of CO2-WAG injection and fish-bone well structures in low permeability reservoirs using artificial neural network

Low permeability oil reservoirs yield low production rates and recovery factors because of their low porosity, low permeability, and low oil mobility. The low production from low permeability reservoirs prohibits many reservoirs from being economically producible. For those reservoirs to be considered for production, advanced production techniques are needed to increase the recovery rates. Previous research efforts concluded that utilizing advanced well structures and Carbon Dioxide based Water Alternating Gas (CO2 -WAG) injection are the most successful techniques to improve oil production from low permeability oil reservoirs. However, evaluating the implementation of a combination of those two production techniques to a low permeability reservoir is an expensive process due to the long run time needed to acquire the results using numerical simulation. Furthermore, optimization and history matching processes for these complicated production plans require more time and high computation power. In this study, we developed a reliable Artificial Neural Network-based toolbox (ANN-Based toolbox) to evaluate, optimize and history match the implementation of CO2 -WAG injection to a reservoir including a fish-bone structure producing well in low permeability oil reservoirs. The developed ANN-Based toolbox provides the results faster than the Numerical Simulation (few minutes compared to several hours) and requires lower computational power. This speed in acquiring the results helps to study more scenarios to optimize the production schemes. The ANN toolbox is developed to perform three tasks; production forecasting, production schemes design, and history matching. The production forecasting proxies are employed as a screening and forecasting toolbox to predict the time-based production rates for the reservoir under consideration when different production schemes are applied. The production schemes prediction proxies are employed to predict the production scheme that would fulfill desired production rates. Finally, the history matching proxy is developed to update the reservoir characteristics based on the production data. The developed proxies are validated using an extensive blind testing dataset, and the results of the testing dataset are presented to validate the capabilities of the developed toolbox. Furthermore, the developed toolbox was used to optimize a CO2 -WAG design scheme for a case study reservoir from Sirri A field. The results provided by the ANN toolbox showed the capabilities of the proposed EOR technique to increase the production in Sirri field by 10%.