A Novel Neural Network based Model Predictive Controller for Congestion Prevention in IP Networks

Abstract

Design of a neural network based model predictive controller for UDP(User Datagram Protocol) flow caused congestion, in IP( Internet protocol) networks is proposed in this paper. The objectives of congestion control are prevention of congestion collapse, maximum network bandwidth utilization, TCP-friendliness and smoothness for streaming media applications. Various approaches for controlling congestion in networks are present in the literature. Many of these are make use of network models, which are already identified. In this paper a neural network utilizing Levenberg-Marquardt learning algorithm for on-line identification of non-linear plant(network) model is implemented and combined with a model predictive optimization technique using back tracking line search routine over a specified time horizon. Simulations were carried out to prove the effectiveness of the designed controller. Significant increase in the network bandwidth utilization is also established.