基于神经元的自适应拥塞控制技术研究

发布时间：2018-08-02 21:44

【摘要】：随着互联网的普及，现在所提供的尽力而为服务已经不能满足多个用户的要求，以及多用户、多业务对不同服务质量的要求。因此在目前研究领域中，提高网络资源的利用率、给予用户提供更好的服务质量，非常具有活力。主动队列管理机制对于拥塞控制有着很好的表现，可以提供好的服务质量。但是目前很多主动队列算法，虽然基本指标表现不错，但是仍然存在一些问题，如参数设置敏感，无法适应网络的动态变化，鲁棒性和稳定性表现不佳等。本文研究了几种主动队列管理算法如RED、ARED、RIO-C等，评价了各个方法的特点。在区分服务算法中，对RIO-C进行了改进，使其具有自适应性。并重点研究了基于控制理论设计的PID队列管理方法，并在此基础上，，联合神经网络控制和模糊控制技术，提出了一种基于神经元的自适应控制方法，该方法采用模糊控制模块，在线调整单个神经元的比例系数，并相应地改进单个神经元的学习速率。仿真结果表明改进方法在延时、丢包率和吞吐量上表现良好，比传统的PID控制方法具有更好的收敛性和稳定性。
[Abstract]:With the popularity of the Internet, the best effort service provided now has been unable to meet the requirements of multiple users, as well as the requirements of multi service and service quality for different services. Therefore, in the current research field, it is very dynamic to improve the utilization of network resources and provide better service quality for users. The system has a good performance for congestion control, and it can provide good quality of service. However, many active queue algorithms, although the basic indicators are good, still have some problems, such as the sensitivity of the parameter setting, unable to adapt to the dynamic changes of the network, the performance of robustness and stability, and so on.
This paper studies several active queue management algorithms, such as RED, ARED, RIO-C and so on, and evaluates the characteristics of each method. In the differentiated service algorithm, the RIO-C is improved to make it self-adaptive. And the PID queue management method based on the control theory is emphatically studied, and on this basis, the joint neural network control and fuzzy control technique are combined. An adaptive control method based on neurons is proposed. The method uses a fuzzy control module to adjust the ratio coefficient of individual neurons on line and improves the learning rate of individual neurons accordingly. The simulation results show that the improved method performs well in time delay, packet loss rate and throughput, and is better than the traditional PID control method. Good convergence and stability.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06;TP183

【相似文献】