The Research About Network Congestion Control Based On Neural Self-adaptive

As the number of Internet users increased rapidly and the new network application typesappeared constantly, the user demand more and more to Internet service quality. In recent years,network congestion control mechanism based on the intermediate nodes becomes hot spot ofresearch, which Active Queue Management (AQM) is based on it. At present, there are a lot ofAQM algorithms with sensitive parameter or fixed parameters, which cannot adapt to the dynamicnetwork and cannot provide higher stability.This paper mainly studies the AQM algorithm based on neural network and fuzzy control.Detailed analysis is showed of the traditional PI and PID active queue management algorithm. PIand PID algorithm will combine classical control theory and active queue management, but thereexist fixed parameter, not real-time adjustment, unable to adapt to the complex nonlinear network.Neural network control and fuzzy control are two branchs in the field of intelligent control, thispaper combine fuzzy control module and single adaptive neuron module, and proposes an improvedactive queue management algorithm, named based on rate fuzzy control of the neuron adaptive PID(RSNAPID) algorithm. The algorithm, which uses fuzzy control module, adjusts the proportioncoefficient of single neuron on-line, and improves the learning rate of single neuroncorrespondingly. Using NS2to simulate RSNAPID algorithm for its own performance, and at thesame time compare RSNAPID algorithm with PI algorithm traditional PID algorithm and SNAPIDalgorithm, and the simulation results show that RSNAPID algorithm has better convergence,stability and robustness.