The use of neuro-fuzzy proxy in well placement optimization

Optimal well placement is crucial step in oil filed development but it is a very sophisticated process on account of different engineering and geological variables affect reservoir performance and they are often nonlinearly correlated. This study presents an approach where a hybrid optimization technique based on genetic algorithm (GA) and a Neuro-Fuzzy system as proxy was created and used to determine the optimal well locations regarding net present value (NPV) maximization as the objective. Neuro-Fuzzy system was used as proxy to decrease the numbers of costly and time consuming-simulations. Such a system has supplanted a conventional technology in some scientific applications and engineering systems, especially in modeling nonlinear systems. Neuro-Fuzzy modeling is a flexible framework, in which different paradigms can be combined, providing, on the one hand, a transparent interface with the designer and, on the other hand, a tool for accurate nonlinear modeling. The rule-based character of Neuro-Fuzzy models allows for the analysis and interpretation of the result. Within Hybrid Genetic Algorithm (HGA), a database of the completed simulations is made. This database is used to construct of Neuro-Fuzzy network. Then this network is used to estimate the fitness function at points that no simulations have not been done. This proxy is also able to get better during the optimization each time a new point is verified and visited points database is updated. A synthetic reservoir was tested and comparisons made among HGA, simple GA and non-proxy using approaches. Results showed that Neuro-Fuzzy system is very reliable proxy to estimate fitness function so the HGA will have a good chance to obtain the optimal place for the well in minimum possible duration.