An active queue management scheme using neural network based predictive control

Active queue management (AQM) is a crucial and attractive theme in congestion control for IP network. Though control theory like proportional integral (PI) controller has been applied in AQM, the quality-of-service (QoS) of IP networks cannot always be guaranteed due to their nonlinear, time-varying and uncertain characteristics. In order to provide better QoS, a novel AQM algorithm, namely NNPC-AQM, is proposed based on predictive control, which requires less model accuracy. A predictor is constructed using two-layer linear neural network (NN) to predict the future queue length and a controller is composed of two-layer nonlinear NN to optimize the next drop probability. The performance of the proposed algorithm is verified and compared with PI controller by simulations. The results show that NNPC-AQM is robust against network parameters like number of TCP sessions and round trip time, and its performance has advantages over PI controller significantly.