The Research On Active Queue Management Based On Intelligent Control

With the development of Internet technology, the network users and traffic increase rapidly, which make congestion problem worse and worse. How to control the congestion becomes a problem that is necessarily solved.The goal of congestion control is to improve the utilization radio of network and to avoid congestion collapse by taking certain control mechanism on the premise of certain throughout. The active queue management (AQM) mechanism is what the IETF recommends, and it is the essential technology based on the router congestion control.The AQM mechanism combines with the TCP end-to-end congestion control, being a main method to solve the congestion control question of the present Internet.The establishment of TCP/AQM congestion control model makes control theory available to be applied to the AQM algorithm. This thesis designs two kinds of AQM algorithm based on intelligent control theory, which are CMAC-PID parallel control AQM algorithm and single neurons-Smith AQM algorithm for large delay network.There are two problems of PID AQM algorithm.One is that parameter is hard to adjust with time,and the other is that it can't adapt to the complicated network environment .The AQM algorithm based on CMAC and PID parallel control can solve the above issues. This algorithm uses CMAC as feed-forward compensation to insure the tracking error converge fast and reduce the overshoot.At the same time it uses PID controller as feed-backward compensation to insure the stability of the system and restrain disturbance.The results of simulation show that the AQM algorithm based on CMAC-PID paralles control can adapt to the time-vary network environment.It has lower steady state error,much quicker respond speed and better ability of robustness than the conventional PID AQM algorithm.Single neuron-Smith AQM algorithm is designed for large delay dynamic network. Smith predictor can compensate the negative impact on the queue stability caused by the large delay.Single neuron has self-learning and self-adapt abilities as the basic unit of neural network, which can compensate Smith predictor's shortcomings of depending heavily on precise system model.At the same time it also has strong robust to the dynamic network fluid. The new algorithm combines the advantages of Smith control and adaptive single neuron control, which adapts to the congestion control of large delay uncertain networks.The design target and parameter adjusting rule of this algorithm are presented detailed in this thesis.The results of simulations show that the Single neuron-Smith AQM algorithm is effective in the queue management of large delay dynamic network environment.