改进神经元PID无线拥塞研究

Research on improved neural network PID congestion control in ad hoc network

神经元PID算法能较好地控制瓶颈节点的队列长度,但当网络环境发生较大变化时,其控制效果往往难以保证。根据Ad Hoc网络环境参量时变的特点,推导了无线TCP/AQM离散模型,在神经元算法的加权系数中引入二次型性能指标。另外神经元增益K是系统敏感参数,而恒定的K值不易适应时变的无线自组织网络,据此设计了一种改进二次型性能指标神经元PID的AQM。仿真结果表明:在动态拓扑、突发流及链路容量变化的Ad Hoc网络中,该改进算法优于PI算法。

Neuron PID can regulate the queue size of a bottleneck node successfully,but its control effect is worse when the network state changes.For ad hoc network’s time-varyingi,t deduces discrete math model of TCP/AQM control object in wireless network.It introduces quadratic performance index to weighting coefficient modification for neuron PID controller.Fur-thermore neuron gain K is system sensitive coefficient.So constant K cant’ be fit for time-varying ad hoc network.Finally,a new improved quadratic performance index neuron PID AQM is proposed.Matlab and NS-2 simulations demonstrate that the performance of new algorithm is better than PI under different link capacity,dynamical topology and sudden flow condition.