Recent progresses in computer science and parallel-processing have opened new frontiers in reservoir simulation applications. New powerful computers can run full field reservoir models faster and with higher accuracy, making reservoir simulator-based optimization feasible. In this study, genetic algorithm is used to estimate the optimal values for design variables to maximize the net present value in a water-flooding project. Surrogate-based optimization has shown promising results in all fields of science. In this work, multiple artificial neural network-based surrogate models, having the capability of on-line recursive adaptation, are presented for optimization purposes. Several genetic algorithm-based approaches have been developed to execute the necessary optimization tasks. A set of simulation test studies are conducted on a synthetic reservoir model to evaluate comparatively the performance of different approaches.