With increasing demand on oil, it is important to improve the recovery factor of oil reservoirs. Naturally fractured reservoirs constitute a major portion of world’s hydrocarbon reserves and are good targets for enhanced oil recovery operation (EOR). Cyclic steam injection is an attractive EOR process for recovering oil from naturally fractured reservoirs. Predicting the performance of different naturally fractured oil reservoirs undergoing cyclic steam injection under varying design parameters is a difficult task. The simulation time and effort required to evaluate such performance for a large number of scenarios is likely to be very high.
Artificial neural networks (ANNs) are mathematical tools designed to map an input domain into an output domain. They are based on observations made in the study of biological systems. Their function is similar to that of a mathematical function. In this work neural network based proxy models are developed for comparative evaluation of cyclic steam injection in various naturally fractured oil reservoirs with constant injection, soaking and production periods. Five different oils with viscosities ranging from 5800 cp to as low as 1 cp at room temperature are used as reservoir fluids in this study and a proxy model is developed for each oil. The proxy models developed are found to be capable of successfully mimicking the reservoir simulation model for above mentioned process within a certain range of input parameters in considerably small computational times.
