基于队列长度和速率的拥塞控制神经网络方法

文中研究了网络拥塞控制问题。PID控制器是实现网络拥塞控制非常有效的方法，能够实现对网络的主动队列管理。文中根据队列长度和变化速率，利用神经网络实现传统的比例微分积分器（PID）功能，从而提出了基于队列长度和速率的拥塞控制神经网络方法（RSPID）。该方法利用神经网络的加权动量梯度学习算法，自动调节控制参数，克服了传统PID控制方法由于控制器参数固定带来的适应性和稳定性问题。仿真结果表明，RSPID算法的鲁棒性和队列长度性能要优于PID算法。

Method of Congestion Control Neural Network Based on Queue Length and Rate

Study the problems of network congestion control. PID controller is an effective method of realizing the network congestion control, which completes the active queue management for network. According to the queue length and rate, using the traditional PID function implemented by neural network, a new network congestion control algorithm （RSPID） based on the queue length and the rate is proposed. The new algorithm adjusts the control parameters by using weighted momentum gradient learning algorithm in neural network, to overcome the adaptability and stability problems in traditional PID control caused by constant controller parameters. The simulation results show that performance of the RSPID algorithm is superior to PID algorithm.