Ensemble-Based Production Strategy Selection for a Stochastic Gradient Method with Well Number and Priority Control Vectors
Resumo
Decision-making in uncertain scenarios with a large number of decision variables (i.e., control vectors in mathematical formulations) is complex. Open-source simulation and optimization tools can assist with these decisions. However, establishing guidelines is important to reduce computational demand, prevent suboptimal decisions, and, consequently, improve oil recovery and economic return. This work aims to define a practical procedure for tuning the parameters of the optimization algorithm and selecting an ensemble-based production strategy for a stochastic gradient method with decision variables of well number and priority. The number of simulation runs, the stochasticity of the optimization algorithm, and the performance of the procedure are also analyzed during the evaluation. We consider reservoir, geological, operational and economic uncertainties using an ensemble of simulation models and economic scenarios from a public-access synthetic reservoir benchmark case featuring Brazilian sandstone (named UNISIM-I-D). We reparametrized the optimization algorithm (named Stochastic Simplex Approximate Gradient (StoSAG)), which is formulated for continuous variables (e.g., well production and injection rates), for commonly used discrete and binary variables. Consequently, we adapted the well priority for well placement selection, defining candidate well’s positions with a quality map and ranking wells at the start of the process based on the Well Economic Indicator (WEI). The well number and priority decision variables are optimized simultaneously using the robust ensemble-based optimization approach. The optimization algorithm’s parameters (i.e., initial guess, perturbation number, and perturbation magnitude) are tuned using an 8-step assisted procedure subdivided into two decisions. During robust ensemble-based optimization, the Objective Function (OF) is the Expected Field Economic Indicator (EFEI). The results indicate that there is an inverse behavior between the perturbation magnitude and the approximate gradient, provided that both control vectors are tuned in the same direction (ascending or descending) and the other parameters are kept, which is useful for tuning parameters. The optimal strategy achieves a 24.9% increase in economic return compared to the worst strategy. These strategies are obtained only during the 2nd decision, which considers the complete optimization process. Thus, this decision-making procedure is useful for avoiding suboptimal production strategy selection caused by an inappropriate parameter setting. In addition, this work establishes guidelines for selecting an ensemble-based production strategy for a stochastic gradient-based technique with well number and priority decision variables, which also allows: (1) selecting well placement based on well priority, (2) reducing computational demand by proposing steps for setting parameters, (3) improving decision accuracy during oilfield development and management, (4) providing insights that can be applied to a variety of problems (e.g., model-based production strategy selection, other decision variables), and (5) automating processes.
Autores
Oliveira, J. P.; Botechia, V. E.; Santos, A. A. S.; Schiozer, D. J.; Oliveira, R.; R. March.
