A Guided Pattern Search with a non-intrusive reduced order modeling for oil production optimization: Brugge field case study

We propose a non-intrusive Reduce Order Modeling (ROM) based on an Artificial Neural Network (ANN) and Discrete Empirical Interpolation Method (DEIM) to reduce the computational burden of reservoir simulations. The proposed ANN-DEIM method has been successfully applied to Brugge field that is modeled in a black box reservoir simulator while trying to preserve the dynamics of the reservoir. Another contribution of this paper is to propose a Guided Pattern Search algorithm for oil production optimization. The algorithm is guided by physical properties of the reservoir and intends to reduce the number of function evaluations in the optimization process. The application of the Guided Pattern Search to the case study reduced the function evaluations by 7 times while improving the initial Net Present Value (NPV) more than 23% using a full order model. Furthermore, the Guided Pattern Search coupled with the proposed ANN-DEIM converged to a better production scenario as compared with the solution found by a Pattern Search algorithm. The reason for this observation is the capability of Guided Pattern Search to utilize the reservoir physical properties besides NPV in the optimization process.