In view of the limitations of performing closed-loop reservoir management with single model, a novel approach for closed-loop management based on multi-geological realizations is proposed with the combination of reservoir numerical simulation technique. A set of realizations are sequentially updated by the ensemble kalman filter (EnKF) method to honor the measurement data. Then taking the expected NPV of the updated realizations as the objective function, the production optimization process is implemented by the simultaneous perturbation stochastic approximation (SPSA) algorithm. The example application indicates that the proposed method for closed-loop reservoir management can decrease the actual geological uncertainty and provide a reasonable estimate reservoir model, and the calculated optimal controls significantly improve the effect of waterflooding with a higher economic benefits and lower developing risks than normal operation strategies.