基于模糊神经PID的主动队列管理方法研究

摘要

PID作为主动队列管理中的重要算法,存在参数不能实时调节、不能适应非线性和动态的网络环境等问题。为了克服这些问题,文中引入基于Mandani规则的模糊逻辑和三层神经网络PID,提出了基于模糊神经PID( FNPID)的主动队列管理算法。该算法利用模糊逻辑计算当前网络学习速率,在神经网络PID中利用加权动量的梯度下降法,根据可变的学习速率来计算相应的丢包率。为使FNPID算法更具有适应性,文中将队列长度和包到达速率作为共同度量,并在丢包率合成时采取动态权重。基于NS2仿真平台,在相同网络环境下对PID算法和FNPID算法的性能进行对比研究。仿真结果表明,相比于PID算法,FNPID算法在稳定队列长度和降低平均时延的同时能使队列长度迅速收敛到期望值,具有较强的适应性和鲁棒性。%As a significant algorithm in active queue management,PID algorithm exists shortcomings that parameters cannot be adjust in real time and cannot be adapted to the dynamic and nonlinear network. For these shortcomings,fuzzy logic based on Mandani rules and 3-level neural network PID are introduced and an active queue management algorithm based on fuzzy neural PID ( FNPID) algorithm is presented. Fuzzy logic part is used to compute the learning rate,then neural network PID calculates the packet loss rate by using weighted momentum gradient learning algorithm. In order to make FNPID algorithm more adaptable,packet arrival rate as same as queue length is taken into consideration,dynamic weight is used in the packet loss rate combination. Under the same network environment,the perform-ance of PID algorithm and FNPID algorithm are compared and researched based on NS2 simulation platform. Simulation results show that FNPID algorithm has stronger adaptability and robustness than PID algorithm,for it can rapidly converge queue length to expected value while maintaining queue stability and reducing the average delay.