神经网络自校正预测拥塞控制算法研究

传输速率、处理速度和节点缓存容量的饱和非线性特性、传输延迟的随机时变性、用户接入的随机性以及高优先级业务的突发性,使得网络中存在严重的不确定性,由此给异步传输模式(ATM)网络拥塞控制系统的分析与设计带来极大的困难。为此设计了鲁棒神经网络自校正拥塞控制算法。其优点在于:(1)最大限度地减小了测量误差和随机干扰的作用,有效地补偿了时变不确定非线性的影响;(2)保证了闭环系统的稳定性、收敛性和公平性,增强了系统对随机延迟等不确定性的鲁棒性。仿真分析进一步验证了该算法的有效性。

Self-tuning predictive congestion control based on neural networks

There are severe uncertainties in the high-speed networks due to saturated nonlinearity on transmission rate, processing speed, buffer capacity, randomness of accessing users and burst of traffic with higher priority, which increase the difficulty in analyzing and designing congestion control systems for ATM networks. A self-tuning predictive congestion control algorithm is presented based on neural networks. The advantages of the method lie in : (1) Measurement error and stochastic disturbance are farthest reduced, and time-varying nonlinear uncertainties are effectively compensated. (2) The stability, convergence and fairness of the algorithm are guaranteed, and the robustness of the systems is enhanced with respect to stochastic transmission delay. The effectiveness of the proposed methods is demonstrated by the simulation results.