RED algorithm is a kind of implementation which is relatively simple, and better solutions to optimize the network congestion problem, but the RED algorithm of control for the network changes is relatively sensitive, So in practical applications may be to enhance the practical utility of RED algorithm by artificially setting control parameters. In this paper, analysis the effect of maximum drop probability on the queue and average queue based on the discrete TCP/RED feedback network model. And discuss when the queue length stabilized at the desired target, the relationship of the maximum drop probaility value with other network parameters. Based on theoretical analysis, proposed an improved RED algorithm, based on NS-2 simulation results show that this algorithm has good average latency and packet loss rate in a complex network environment, superior FRED and BLUE algorithm.