SimProxy Decision Support System: A Neural Network Proxy Applied to Reservoir and Surface Integrated Optimization

The development of an oil reservoir consists in drilling wells that maximize revenue. The quest for this configuration is often based on optimization processes that use the net present value as the evaluation function. Determining quantity, location, type, and trajectory of wells in a reservoir is a complex optimization problem, which has a high computational cost due to the continuous use of simulators. Many researchers have proposed the replacement of the reservoir simulator by a proxy, to reduce the computational cost, with promising results. However, in order to analyze all relevant variables and obtain a better solution, a comprehensive proxy model must also consider the conditions of the surface behavior, incorporating the complete integrated system. None of the previous works has developed a model based on the integrated simulation behavior. This article presents the proposal of a new proxy model based on neural networks, called SimProxy, which integrates reservoir and surface behavior, to reduce the computational cost of a decision support system. The proposed model was evaluated in a real oil reservoir. The results indicate that SimProxy can efficiently replace the use of commercial simulators in an optimization process, providing good accuracy with a substantial decrease in computational cost.