Optimizing the volume of water injection in real time and thus realizing delicate water injection, it is of great significance to improve the oilfields recovery with high water-cut in waterflooding. In this paper, an intelligent optimization method and procedure for detailed water injection design based on Random Forest and Multi-Objective Simulated Annealing (MOSA) algorithm is proposed. The method begins with a comprehensive evaluation of the injection effect of each well based on dynamic data, followed by the application of the Random Forest algorithm to qualitatively analyze the volume adjustment direction, and then the MOSA algorithm is used to quantitatively optimize the water injection volume. The method makes full use of test data and the advantages of machine learning algorithms, which can quickly achieve real-time optimization of injection volume for a large number of wells, and has been successfully applied to real oilfields such as Daqing, Jilin, Qinghai, Dagang and so on. Moreover, for wells with a fourth-generation stratified injection technique, the amount of water injected into each stratified section can be optimized by the process proposed in this paper, using the layered injection and production “hard data” monitored at real time by automatic control technology. The stratified section water injection volume of each smart well was optimized and adjusted with this approach in a high-water-cut reservoir in eastern China. After six months, the average water cut decreased nearly by 1% and the average monthly water injection volume decreased by 13%, while the monthly oil production remained basically stable, which verifies that this method can provide guidance for real-time water injection and modulation in reservoirs, thus effectively improving the performance of water-flooding development of heterogeneous reservoirs.