An adaptive PID controller for AQM with ECN marks based on neural networks

Nowadays congestion control problem of the intermediate nodes in the Internet has received extensively attention in networking and control community. In this paper, a novel adaptive PID (Proportional-Integral-Differential) controller based on neural networks for the problem of AQM with ECN marks is presented. Considering a previously developed nonlinear dynamic model of TCP/AQM system and the queue management mechanism of intermediate nodes, the parameters of AQM controller based on neural networks is tuned online by using gradient-descent algorithm, and the packet drop/mark probability is determined adaptively in time to avoid congestion, so that the quality of service (QoS) of network and the transient performance can be improve greatly especially when the network parameters are time-varying. Finally, the proposed algorithm is verified by using NS-2 simulator, and simulation results show that the integrated performance of this proposed controller is obviously superior to those of typical RED and PID controller especially on the queue stability and mean time delay. Furthermore, this AQM algorithm has simple structure and can be implemented easily.