Raft consensus algorithm is a reliable consistency algorithm implementation in distributed systems, but the traditional Raft algorithm is easy to make efficiency low because of malicious attacks. Aimed at the problem that the leader identity in Raft algorithm may be detected by attackers in the process of execution and then attacked, the randomness and non-interactive characteristics of verifiable random functions are utilized for improving the original Raft algorithm, and hiding the leader identity in the leader election process, avoiding the multi-round election problem in the original algorithm. This paper designs an implementation of verifiable random functions in detail, and incorporates it into the leader election process. The paper also analyzes the system security and stability of the improved scheme, compares the performance of different consensus algorithms in various requirement scenarios. The simulation experiments show that the improved algorithm hides the identity of the elected node in the leader election process, improves the system security on the premise of ensuring system stability, and also performs still more in improving the time efficiency of leader election with the increase of the number in system nodes.