Current workflows for well completion and production optimization in unconventional reservoirs require extensive earth modeling, hydraulic fracture simulation and production simulation. This methodology is data intensive and a time-consuming and often not rigorously accomplished due to lack of skillset and time. In addition, a complete understanding of the impact of each variable and the relations between them is very difficult to achieve. Considering the potential value of these studies on well completion design optimization, overcoming the above-mentioned limitations become critical. To do so, conducting fracture and reservoir simulations in the cloud and analyzing the results using data analytics and machine learning algorithms can help to develop a powerful solution that creates "proxy" models for fast and effective completion optimization.
In the present work, three reservoir models were calibrated for different landing zones and a sensitivity work was carried out changing stimulation design. For each run, hydraulic fracture geometry and productivity was obtained. The numerical simulations were run in the cloud, making approximately 500 runs in five days. The results of all the runs were analyzed and evaluated using machine learning and data mining techniques, providing a better understanding of the effect of completion variables on fracture geometry and well productivity. This in turn, translated into a better comprehension and use of the numerical simulations.
Finally, a machine learning model was fitted (using gradient boosting technique) to obtain the productivity of the wells from the stimulation design parameters and the reservoir properties, without the need of additional runs and in a few seconds instead of weeks.
Current workflows for well completion and production optimization in unconventional reservoirs require extensive earth modeling, hydraulic fracture simulation and production simulation. This methodology is data intensive and a time-consuming and often not rigorously accomplished due to lack of skillset and time.
In addition, a complete understanding of the impact of each variable and the relations between them is very difficult to achieve. Considering the potential value of these studies on well completion design optimization, overcoming the above-mentioned limitations become critical. To do so, conducting fracture and reservoir simulations in the cloud and analyzing the results using data analytics and machine learning algorithms can help to develop a powerful solution that creates "proxy" models for fast and effective completion optimization.
