基于Kalman算法及神经网络预测的网络流量控制  

Flow Control Based on Kalman Algorithm and Neural Network Prediction in Networks

作　　者：沈伟[1] 冯瑞[1] 邵惠鹤[1] 

机构地区：[1]上海交通大学自动化研究所,上海200030

出　　处：《计算机研究与发展》2003年第8期1162-1167,共6页Journal of Computer Research and Development

基　　金：国家"九七三"重点基础研究发展规划项目 (G19980 3 0 415 )

摘　　要：针对通信网络的传播时延会给基于速率反馈的流量控制带来极大的不利影响 ,提出了基于Kalman算法的反馈控制和神经网络在线预测补偿相结合的复合控制 ,对ATM网络的ABR流量进行控制 ,较好地克服了时延对流量控制的快速性和稳定性所产生的不利影响 仿真研究表明 :本方案能使信源的发送速率快速响应网络状态的变化 ,有效地避免拥塞的发生 ,并使链路带宽得以充分利用 与PID控制方法相比 ,信元的丢失率更低、链路的利用率更高以及所需的缓冲容量更小 .The propagation delay in networks has a great adverse effect on rate-based flow control. Proposed in this paper is the composite control based on Kalman algorithm feedback control and neural network predictive compensation on line for ABR communication in ATM networks, which can better overcome the adverse effect caused by the delay on the control rapidity and stability. The simulation shows that the scheme can make sources respond to the changes of network status rapidly, avoid the congestion effectively, and utilize the bandwidth sufficiently. Compared with PID control, much lower cell loss rate, much higher link utilization rate, and much smaller buffer capacity can be achieved.

关 键 词：ATM网络 流量控制 KALMAN算法 神经网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术] TP273.5[自动化与计算机技术—计算机科学与技术]