基于预测控制的时滞网络主动队列管理  

Active queue management of delay network based on model predictive control

作　　者：王萍[1] 陈虹[1] 卢晓晖[1] 

机构地区：[1]吉林大学通信工程学院,长春130022

出　　处：《吉林大学学报（工学版）》2011年第4期1089-1095,共7页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家杰出青年科学基金项目(60725311)

摘　　要：针对Internet网络这种典型的延迟系统,基于模型预测控制理论,提出了一种新的主动队列管理算法——MPAQM算法。首先考虑时滞系统的因果性定义预测输出,根据预测模型和状态估计器,预测瓶颈链路队列的未来动态特性;然后将拥塞控制的控制要求转化为优化目标函数,求解优化问题得到丢包率。在单瓶颈和多瓶颈网络拓扑中进行仿真验证,结果显示MPAQM算法能快速平稳地收敛于队列长度的期望值,丢包率小且具有较好的稳定性和鲁棒性。Based on the theory of model predictive control,a novel Active Queue Management（AQM） algorithm is proposed for typical Internet systems with large delay,called MPAQM.First,the predicted output is defined by considering the causality of time-delay system.Then,according to the predictive model and state estimator,the future dynamic queue length in data buffer is predicted.Finally,the control requirement of congestion is converted to optimal control objective function,and drop probability is obtained by solving the optimal problem.The performances of MPAQM are evaluated through a series of simulations in simulator NS2 with single and multiple-bottleneck network topologies.Simulation results show that the proposed MPAQM algorithm outperforms RED and PI algorithms in terms of stability,disturbance rejection and robustness.

关 键 词：自动控制技术 控制理论 网络拥塞控制 主动队列管理 模型预测控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]