With large-scale electric vehicles（EVs）integrated into the power system，how to optimize EV charging power in real time to avoid incremental peak load and achieve valley-filling will be important.The regular real-time charging optimization as well as the vehicle charging prediction method is reviewed at first.Since vehicle charging（especially vehicle fleet charging）is predictable，a new real-time charging optimization model is proposed on the basis of the regular method with the consideration of the prediction model of incoming EVs.The new model is solved with receding horizon optimization，obtaining the current optimized charging power of plugged EVs.Through case study with various prediction accuracies，it can be seen that the proposed method always approaches the optimal solution to valley-filling.Moreover，compared with regular method，the proposed method would make a greater improvement of optimization result in the case of more diverse EV plug-in time.