Research On Congestion Control Based On Nerual Network

As the new network application types increasing, especially tripe play is going to realize, QoShas become increasingly demand.In recent years, active queue management (AQM) has become ahot issue in network congestion control. Network congestion control is aimed to minimal packetloss rate, lower latency, more fair bandwidth allocation, lower network jitter, larger throughput.Many AQM algorithms can not adapt to dynamic network because of fixed parameters. Thisthesis studies the network congestion control methods based on nerual network. This thesis analyzestraditional AQM algorithms, such as RED algorithm, ARED algorithm, PI algorithm and PIDalgorithm, they all have fixed parameters, and can not adapt to non-linear dynamic networks. BPneural network has a lot of features,such as " approach". This thesis proposes a congestion controlalgorithm combined queue length with rate based on neural network and fuzzy control——RSPIDalgorithm. This thesis also proposes a algorithm based on neural network and CHOKe congestioncontrol——CNRPID algorithm. Using NS2to simulate the two proposed algorithms andtraditional AQM algorithms, simulation results show that the two proposed algorithms have betterrobustness, convergence and stability compared to traditional AQM algorithms.