基于神经网络逆系统的鲁棒主动队列管理算法

通过在中间节点上使用主动队列管理策略来进行有效地拥塞控制,在保证较高吞吐量的基础上稳定地控制队列长度,从而实现了端到端的时延控制和保证QoS需求.在研究中,TCP的流量控制过程被视为二阶非线性时变系统,并通过可逆分析,证明该系统可逆,采用神经网络逆系统这种近年来发展起来的非线性鲁棒控制理论作为控制器的设计方法,设计出一种新的主动队列管理算法.仿真试验表明,这种算法的稳态和瞬态性能都优于与其具有相同实现复杂度的 RED和PI算法,并且在负载扰动和参数变化时具有很强的鲁棒性.神经网络逆系统方法应用于非线性的流量控制过程中有助于系统稳定性和鲁棒性.

Robust algorithm for active queue management based on ANN inverse system

On the intermediate nodes active queue management is used for an effective congestion control policy.Base on guarantee of high throughput,it controls the queue length stabilization,then realizes the end-to-end delay control andensures QoS(quality of service) demands.In this paper,the TCP(transmission control protocol) flow process is modeled as two order nonlinear varying-time system.By analyzing the invertibility of the system,a new AQM(active queue management) algorithm based on artificial neural network inverse(ANNI) system theory that is a newly developed nonlinear theory with good robustness is proposed.The simulation results show that its stability and transient performance are superior to RED(random early detection) as well as PI algorithms.Moreover, this new AQM algorithm possesses high robustness even when network load fluctuates or system parameter changes.Artificial neural network inverse method is helpful to system stability and robustness when it is applied to nonlinear TCP flow process.