An AQM Algorithm Research Based On Neural Network

With the rapid development of the Internet, network congestion control, as one of the main means to enhance the performance of the network, becomes one of the hot research areas. The aim of congestion control is to maximize the link utilization, minimize the packets delay, distribute the network resources among users reasonably and drop packets as few as possible. As the supplementary means of the TCP end-to-end congestion control, AQM (Active Queue Management) makes the intermediate node join network congestion control, and becomes a hot research area in congestion control these years.As an enhancement mechanism for the end-to-end congestion control, active queue management (AQM) such as red can control queue length efficiently on the basis of keeping higher throughput. But research results show that most AQM schemes are very sensitive to the network states such as the variety of link number, service type and time delay. The incorrect parameter setting can often bring queue oscillation, throughput degradation and delay jitter acuity and so on.This thesis presents the background, advantages and performance evaluation of active queue management, and analyses the existed AQM schemes at first. Then the modeling method of TCP+AQM congestion control mechanism based on control theory is described. introduce the platform NS-2, By using neural network and predictive control, This paper prove a new AQM algorithm that based on neural network and predictive control. The simulation results show that the algorithms improve the transient performance, and increase robustness against uncertainty such as time delay and variety of number of users, and improve network's utilization..