Neural networks for the adaptive control of disruptive nonlinear network traffic

A study has been made of nonlinear behaviour found within the traffic profile and physical components of distributed communications networks, with particular attention given to the chaotic dynamics produced at the extremes of such behaviour. This analysis has led to the development of an algorithm describing how an arbitrary traffic flow may be monitored and characterised by a discrete sequence of analytical and statistical metrics. These metrics are then employed by a network-wide distributed flow control mechanism based on adaptive traffic routing, using a neural processor. The processor is configured to retain an imprint of recent flow characterisation to build a projection of likely future behaviour. Results show that when applied to networks operating across a wide variance of traffic conditions, the flow management system improves end-to-end delay through maintaining and/or regaining traffic flow stability