Well control optimization in waterflooding using genetic algorithm coupled with Artificial Neural Networks

• Published in: Upstream Oil and Gas Technology Vol. 9; p. 100071
• Main Authors: Alfarizi, Muhammad Gibran, Stanko, Milan, Bikmukhametov, Timur
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2022
• Subjects: Artificial Neural Networks; Control optimization; Genetic algorithm; Waterflooding; Waterflooding; Artificial Neural Networks; Control optimization; Genetic algorithm
• ISSN: 2666-2604; 2666-2604
• DOI: 10.1016/j.upstre.2022.100071

Optimum well controls to maximize net present value (NPV) in a waterflooding operation are often obtained from an iterative process of employing numerical reservoir simulation and optimization algorithms. It is often challenging to implement gradient-based optimization algorithms because of the large number of variables and the complexities to embed the optimization algorithm in the simulator solving workflow. Approaches based on repeated model evaluation are easier to implement but are often time-consuming and computationally expensive. This work proposes the use of Artificial Neural Networks (ANN) to replicate the numerical reservoir simulation outputs. The ANN model is used to estimate cumulative oil production, cumulative water injection, and cumulative water production based on sets of well control values, i.e. flowing bottom-hole pressure. Then, the ANN model is combined with the genetic algorithm (GA) optimization (a derivative-free optimization) to find the optimum well controls that maximize the NPV of a synthetic reservoir model. The optimization results of this ANN-GA model were compared against the results of using the traditional approach of applying the genetic algorithm directly on the numerical reservoir model. The ANN model successfully reproduces the results of the numerical reservoir model with a low average error of 1.89%. The ANN-GA model successfully finds optimal operational conditions that are identical to those found by using GA and the original reservoir model. However, the running time was lowered by 96% (43 h faster) when compared to the optimization scheme using the original reservoir model. The optimal solution increases the NPV by 22.2% when compared to the base case.