To deal with the long delayed control process of dissolved oxygen in sewage, a control strategy is proposed by combining directly predicting variables based on curve fitting and directly searching for optimization control variables. This proposed method uses a least square algorithm to fit the online data and obtains the predictive values after τ time delay. An optimization method based on the golden section points is used to determine the boundary of the target control variables. The search interval length is gradually reduced till the target control target is reached. Simulation and experimental results show that the proposed control strategy is more effective and efficient for control process of a large time delay. Comparing with traditional method like expert control, the proposed approach can improve convergence accuracy up to 50% and can be used to control the target to ±1%.