基于混合神经网络的主动队列管理算法

Research of active queue management algorithm based on mixed neural network

针对高速互联网中拥塞控制的问题,在主动队列管理算法模型基础上,提出了一种基于混合pi-sigma神经网络的动态管理机制．其模型可以方便地在线修正前提参数(隶属函数)和结论参数,适合网络系统拥塞预测和控制．仿真表明,该算法能够保证缓存器中队列长度的稳定性,而且在网络突发流量较大时,在短时间间隔内可以使流量的抖动变得平缓,对网络动态的、不精确的、突发性的环境具有较强的自适应能力．

On the basis of the active queue management algorithm model, a dynamic management system is presented based on mixed pi-sigma neural network to solve the congestion problem in high-speed intemet network. The model can modify premise parameters （ subject functions） and conclusion parameters on line conveniently, suitable for congestion forecast and control of the network system. Simulation shows that this algorithm can ensure the stability of queue length in buffer storage, and make the flux fluctuation become mild in short time interval when the network burst flux happens. This algorithm possesses strong adaptation in dynamic, imprecise and sudden environment.