Design of neural model predictive controller for active queue management

Abstract

Recently many active queue management (AQM) algorithms have been proposed to address performance degradations of end-to-end congestion control. However, these AQM algorithms show weaknesses to detect and control congestion under dynamically changing network situations. In this paper, we use a neural model predictive controller (NMPC) for AQM in the Internet based on neural networks. Which uses P/sub 1,2/ Pade's approximation model of bottleneck network to provide robust and predictive congestion avoidance. Based on a fluid theoretical model of a network, a neural network is trained to control the traffic flow of a bottleneck network router. Our simulation results show that this scheme has better robustness, short response time, and more desirable tradeoff than RED and REM, especially under highly dynamic network and heavy traffic load.