Abstract
Well placement optimization constitutes a fundamental element in the decision-making process needed for optimizing the field development plans. The use of engineering judgment as well the usual practice of trial and error to solve the challenging problem of well placement may be not sufficient. However, the employment of optimization routines coupled to reservoir simulation models as an automatic aiding tool is a current alternative that could improve the decision-making process. Moreover, despite the successes achieved, automatic optimization strategies still suffer from various drawback such as large computational effort, substantial CPU time consumption due to the expensive simulation model, in addition to the difficulties in handling the realistic field constraints.
In this paper, A hybrid intelligent approach for optimizing well placement under field constraints is presented that use a constrained space-filling experimental design, genetic algorithm (GA), adaptive neuro-fuzzy inference system (ANFIS) surrogate model and one proposed adaptive sampling framework, as its basis for optimization. The hybrid intelligent strategy is initiated by uniformly distributing a limited number of design sites in the constrained search space by space-filling design, then simulations are made to generate a database to fit the ANFIS model. Thereafter, a proposed adaptive sampling routine, which combines an intelligent neighborhood search mechanism with GA is applied to augment the quality of surrogate, enhance the accuracy of the framework, and thus guide the optimization rapidly into the optimal solution.
Via a full-field reservoir case that dealing with a realistic reservoir model based on the Namorado field, Campos basin, Brazil, the proposed technique is assessed against the widely used automatic routines that are known as direct optimization with GA, offline and online surrogate-based optimization approaches. The comparison indicates that the proposed method is more accurate, reliable and efficient than all the other optimization routines presented in this work. The results of this study highlighted the effectiveness of the proposed hybrid approach for solving the real well placement problem with high accuracy in reasonable CPU-time. These auspicious features make it a reliable tool to be used on other real optimization projects.
