基于神经网络监督控制的拥塞控制算法研究

Study of congestion control algorithm based on neural network supervised control

提出了一个基于神经网络控制的主动队列管理(AQM)算法;研究了TCP/AQM拥塞控制系统的可逆性,并利用一种神经网络监督控制结构进行了AQM算法的设计。算法由一个三层前馈结构的神经网络控制器(neural network controller,NNC)和一个反馈控制器(feedback controller,FC)组成。NNC作为一个前馈控制器,通过FC产生的教师信号进行学习,以建立被控对象的逆动力学模型。仿真结果表明,提出的算法与PI(proportion-al-integral)算法相比,无论在瞬态性能还是稳态性能方面都可以取得比较满意的效果。

This paper proposed a novel AQM algorithm based on neural network control. Firstly, studied the invertibilitv of TCP/AQM system, then designed the algorithm using one form of neural network supervised control, which was composed of a three-layer feedforward NNC and a conventional FC. The NNC acted as a feedforward controller and the FC generates training signal of NNC. Through learning, the NNC can be identified as an inverse model of the plant and simulation results show the proposed algorithm can outperform the PI algorithm.