English abstract
Finding the optimal well placement scenario under geological uncertainty is a challenging task
due to its expensive computational costs. A viable solution is introduced in this study, where a
deep learning multi-input artificial neural network with convolutional layers and fully
connected layers was integrated to build a proxy model that partially substitutes a full physics
reservoir simulator to find the optimal well locations in a one-year field development plan. The
plan consisted of installing an injection and a production well for two periods in a reservoir
exhibiting permeability uncertainty. The methodology started with acquiring data from the
synthetic reservoir egg model by running reservoir simulations for two periods of 6 months
each, then a multi-input deep learning neural network-based proxy model was built to estimate
the potential cumulative oil production at different well locations, and finally, Particle Swarm
Optimization algorithm was coupled with the proxy model to search for the optimal well
locations for each period. The neural network model was trained to input an image of the
permeability realization and a vector of well locations to output cumulative oil production in
𝑠𝑚3
. The data size for the first period was 7500 and 4000 for the second period. Both datasets
were split into 60% training, 20% validation, and 20% testing. The neural network model
yielded a testing coefficient of determination value of 0.89 and 0.73 for the first and second
period, respectively. Furthermore, it was able to suggest a well placement scenario that yielded
a cumulative oil production value that is 96% of the optimal value obtained from conventional
methods for both periods. The proposed solution also dramatically reduced the computational
time, as it required only 25.3% of the time it would take using reservoir simulators for the first
period and only 14.6% for the second period. Thus, the coupling of a deep learning proxy model with particle swarm optimization was found to be reliable for making effective decisions
regarding field development plans under geological uncertainty.
