Development of an adaptive surrogate model for production optimization

Recently production optimization has gained increasing interest in the petroleum industry. The most computationally expensive part of the production optimization process is the evaluation of the objective Function performed by a numerical reservoir simulator. Employing surrogate models (a.k.a, proxy models) as a substitute for the reservoir simulator is proposed for alleviating this high computational cost. In this study, a novel approach for constructing adaptive surrogate models with application in production optimization problem is proposed. A dynamic Artificial Neural Networks (ANNs) is employed as the approximation function while the training is performed using an adaptive sampling algorithm. MultiANNs are initially trained using a small data set generated by a space filling sequential design. Then, the state-of-the-art adaptive sampling algorithm recursively adds training points to enhance prediction accuracy of the surrogate model using minimum number of expensive objective function evaluations. Jackknifing and Cross Validation (CV) methods are used during the recursive training and network assessment stages. The developed methodology is employed to optimize production on the bench marking PUNQ-S3 reservoir model. The Genetic Algorithm (GA) is used as the optimization algorithm in this study. Computational results confirm that the developed adaptive surrogate model outperforms the conventional one-shot approach achieving greater prediction accuracy while substantially reduces the computational cost. Performance of the production optimization process is investigated when the objective function evaluations are performed using the actual reservoir model and/or the surrogate model. The results show that the proposed surrogate modeling approach by providing a fast approximation of the actual reservoir simulation model with a good accuracy enhances the whole optimization process. (C) 2015 Elsevier B.V. All rights reserved,