【作者】 刘志新；

【导师】 关新平；

【作者基本信息】 燕山大学 ， 控制理论与控制工程， 2003， 硕士

【摘要】 综合业务数字网是未来通信技术的发展趋势，ATM网络已被国际电信联盟作为一项典型传输技术加以推广。在ATM网络中，信息的拥塞及丢失是影响网络业务服务质量的主要原因。其中ABR业务是唯一一种可采用反馈机制进行流量控制的业务(因此网络拥塞控制问题引起了广大控制和通信学者的关注)，ABR业务流量的控制和管理问题近年来也成为一个研究的热点。通讯网络是一个庞大的复杂系统，ATM网络拥塞控制研究对控制、通信而言均具有重要的理论意义和实用价值。本文正是以此为出发点，将控制理论引入到网络通讯中，解决可控流的拥塞控制问题。本文首先从ATM网络通信基础知识开始，介绍了ATM网络的基本原理，ABR业务的反馈机制，给出了ATM网络单瓶颈节点模型，并在此模型基础上，将PID控制引入到网络控制当中，设计出适用于ATM网络模型的PID控制器，给出了保证系统闭环稳定的充分条件。进而利用前馈控制环节降低带宽波动对输出队列的影响。为消除队列输出饱和特性对控制器的影响，采用了虚队列机制，同时为加快系统的响应速度，设置了速率提升因子和速率下降因子。在以上各种方法中，均能保证系统队列输出是稳定的。时延变化及带宽波动始终是影响系统稳定及动态性能的重要因素。本文在原有文献的基础上提出了两种改进方法，分别应用内模控制和Smith预估方法解决ABR业务拥塞控制问题，可以实现系统稳态无静差跟踪给定值，在网络可用带宽大幅波动的情况下，算法仍能保证输出队列长度稳定在一定范围之内。最后将神经网络智能控制方法引入到网络拥塞控制之中。利用神经网络的自组织、自学习能力，实现对可用带宽的预测、对交换机队列模型的建模及用神经网络控制器实现队列控制。针对具有ARMA、FARIMA等不同特性的可用带宽时间序列，预测网络都可以实现较为精确的预测，在此基础上进行的PERICA算法、神经网络控制算法、公平算法都取得了较好的控制效果。更多还原

【Abstract】 Integrated Service Digital Network(ISDN) is considered as the tendency of the communication technique in the future. Asynchronous Transfer Mode(ATM) is adopted as a typical technique by the International Telecommunication Union(ITU) and is spreaded . But the loss of data and the congestion of information are main reasons that affect the quality of service. ABR service is the only one type of traffic that can be controlled using the feedback mechanism. So, in the recent, the control and management of ABR service becomes a hot subject. The communication network is a large and hybrid system, the research on congestion control of ATM network has great significance in theory and practice. The paper is just based on it and the control theory is used to solve the congestion of controllable flow.The paper begins with the foundation of ATM network, and introduces the basic work principle of networks and the feedback mechanism. Then, the model of single bottleneck node is set up and the PID control method is introduced into the net control based on the model. The controller for the special network model is designed and the sufficient condition is given which can guarantee the stability of closed loop. The feed-forward control is adopted to reduce the effect caused by oscillation of available bandwidth. To remove the effect of the saturation feature of the queue length, the virtual queue is set in the switch. At last, the increase factor and decrease factor are utilized to quicken the response. All the method mentioned above can keep the queue output stable.Delay variation and oscillation of available bandwidth are two important elements that affect the stability and dynamic performance of the system. Two improved methods are presented base on the existing literature. They solve the problem of ABR congestion control using the inner model control and Smith predictor, respectively. The system can keep the queue length stable in a certain region when the available bandwidth has large oscillation.In the final, the intelligent control method of neural network is introduced in the congestion control of network. The neural network can realize the<WP=6>prediction of available bandwidth, the modeling of queuing model of the switch and the queue control. The predictive networks can achieve the accurate predicted value for the processes such as ARMA and FARIMA time sequence. Based on it, the PERICA algorithm, neural network control algorithm and fairness algorithm achieve better control performance.更多还原