Optimization of Oil Production Using Reduced Order Modeling in Hydrocarbon Reservoir Simulation

Production-injection optimization has been the subject of various researches due to its complicated and expensive computations. The main reason for this complexity is number of reservoir simulation runs is needed to predict reservoir performance. These numerical reservoir simulations are computationally expensive and time consuming. Therefore, finding a way to reduce the computational burden of reservoir simulation will facilitate the optimization process. One of the methods for reducing the complexity of reservoir simulation is Reduced Order Modeling (ROM) which has been recently introduced for improving efficiency of open source reservoir simulators. In this paper, an ROM method based on Artificial Neural Networks (ANN) and Discrete Empirical Interpolation Method (DEIM) is proposed to resolve the curse of dimensionality while simulating reservoir dynamics with acceptable accuracy. This method is also applicable to black box reservoir simulators. The performance of the suggested ANN-DEIM algorithm has been investigated on a case study on Brugge field. The reduced model well represent the reservoir dynamic behavior while reducing run time by a factor of eight comparing with that of a full order reservoir simulator. ANN-DEIM also has been applied in production-injection optimization of Brugge filed using a Pattern Search optimization algorithm. The proposed method can reduce optimization time by 7 times while leading to %11 improvement in Net Present Value (NPV) over the initial solution used in the optimization process.