For consensus control of multi-agent systems under malicious attacks, a fast and effective secure consensus algorithm is proposed. In each status update process, normal agents select the median state value in an arranged value sequence collected at the same time from their neighbor agents. The computation complexity is reduced. The strong network connectivity and vast resource needed for information storage of the traditional algorithm are also reduced. The proposed algorithm makes the whole system much simple, flexible, and cheap. By using iterative learning and convex hull conditions, we create a virtual network topology with the same connectivity condition as the original system directed graph. Under the condition that the system satisfies certain network topology, it is proved that the system is safe and consistent. Simulation results verify the effectiveness of the proposed algorithm.