Research On Active Queue Management Of Network Congestion Control

With the rapid development of modern communication networks represented by the Internet,more and more users send and get information through the Internet, the user demand has becomeincreasingly rich and diverse, the end-to-end congestion control mechanism has been difficult toavoid or alleviate network congestion phenomena. Therefore, the problem that research oncongestion control technology within the network needs to be solved in network management. Asan important technology of congestion control in network intermediate routing node, active queuemanagement（AQM）has gradually become a hot research in network congestion control filed.Currently, a major approach that solve network congestion problem is that implement AQM strategyin the routing node combined with congestion control mechanism based on transmiss ion controlprotocol（TCP）.This paper mainly studies the AQM algorithm in the routing node of communications networkmodel. Firstly, the paper describes in detail the theory of congestion control, then focuses onanalyzing AQM algorithm, including composition, design rules, performance indicators, fairness,and analyzes further several classic AQM algorithms, then optimizes proportional integralderivative (PID) controller, proposes a modified self-adaptive PID neural network（MAPIDNN）algorithm. MAPIDNN algorithm introduces neural network controller that applies gradient descentmethod based on additional momentum factor to adjust the weights, and utilizes quadratic functionto adjust packet loss rate adaptively in the arrival time of each packet to adapt to changes of theactual queue length, simultaneously, introduces the stateless traffic matching mechanism to enhancethe viability of TCP self-adaptive traffic. Simulation results show that MAPIDNN algorithm hasbetter transient performance, high throughput and bottleneck link utilization, minimal packetqueuing delay and delay jitter, while enhancing the viability of TCP self-adaptive traffic comparedwith PID, PIDNN algorithm.