TY - JOUR
T1 - Development of universal proxy models for screening and optimization of cyclic pressure pulsing in naturally fractured reservoirs
AU - Artun, E.
AU - Ertekin, T.
AU - Watson, R.
AU - Al-Wadhahi, M.
PY - 2011/12
Y1 - 2011/12
N2 - Economic success of an IOR project primarily depends on the effectiveness of the selected method in the reservoir of interest and the way it is deployed. Due to the heavy computational work required for a large number of simulations, it is rather an arduous task to determine optimum design schemes for a given project. Proxy models that mimic reservoir models can potentially offer an opportunity to reduce the number of simulation runs and allow timely reservoir-management decisions to be made. In this study, a high-performance screening/optimization workflow is presented to narrow the ranges of possible scenarios to be modeled using conventional simulation. The workflow proposed starts with creating a knowledge base with simulations for a wide range reservoir characteristics and design scenarios. Then, neural-network based proxy models are trained with the knowledge base. Finally, the genetic algorithm is used to search for the optimum design scenario by evaluating the objective function via proxy models for increased computational efficiency. This workflow is applied to the cyclic pressure pulsing process with CO2 and N2 in naturally fractured reservoirs. Developed proxies are universal such that the range of their applicabilities is extended to a wide spectrum of reservoir characteristics, including initial conditions, well spacing and reservoir fluid types (heavy, black, and volatile). Proxy models proved to be able to predict critical performance indicators such as cumulative oil production, and oil flow rates within high levels of accuracy (with a correlation coefficient of 0.91 on average). The genetic algorithm is coupled with proxy models to maximize the discounted oil volume produced per injected volume for a specified period of operation. After utilizing the aforementioned workflow, it is possible to significantly reduce the computational time and manpower expended with an initial screening of the cyclic pressure pulsing process. In this way, more effective design strategies can be structured for the reservoir under consideration.
AB - Economic success of an IOR project primarily depends on the effectiveness of the selected method in the reservoir of interest and the way it is deployed. Due to the heavy computational work required for a large number of simulations, it is rather an arduous task to determine optimum design schemes for a given project. Proxy models that mimic reservoir models can potentially offer an opportunity to reduce the number of simulation runs and allow timely reservoir-management decisions to be made. In this study, a high-performance screening/optimization workflow is presented to narrow the ranges of possible scenarios to be modeled using conventional simulation. The workflow proposed starts with creating a knowledge base with simulations for a wide range reservoir characteristics and design scenarios. Then, neural-network based proxy models are trained with the knowledge base. Finally, the genetic algorithm is used to search for the optimum design scenario by evaluating the objective function via proxy models for increased computational efficiency. This workflow is applied to the cyclic pressure pulsing process with CO2 and N2 in naturally fractured reservoirs. Developed proxies are universal such that the range of their applicabilities is extended to a wide spectrum of reservoir characteristics, including initial conditions, well spacing and reservoir fluid types (heavy, black, and volatile). Proxy models proved to be able to predict critical performance indicators such as cumulative oil production, and oil flow rates within high levels of accuracy (with a correlation coefficient of 0.91 on average). The genetic algorithm is coupled with proxy models to maximize the discounted oil volume produced per injected volume for a specified period of operation. After utilizing the aforementioned workflow, it is possible to significantly reduce the computational time and manpower expended with an initial screening of the cyclic pressure pulsing process. In this way, more effective design strategies can be structured for the reservoir under consideration.
KW - Cyclic pressure pulsing
KW - Genetic algorithms
KW - Huff 'n' puff
KW - Naturally fractured reservoirs
KW - Neural networks
UR - http://www.scopus.com/inward/record.url?scp=84855476727&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84855476727&partnerID=8YFLogxK
U2 - 10.1016/j.jngse.2011.07.016
DO - 10.1016/j.jngse.2011.07.016
M3 - Article
AN - SCOPUS:84855476727
SN - 1875-5100
VL - 3
SP - 667
EP - 686
JO - Journal of Natural Gas Science and Engineering
JF - Journal of Natural Gas Science and Engineering
IS - 6
ER -