Multi-criterion optimisation control framework for intelligent network traffic agents

arch question investigated the feasibility of machine learning techniques, in particular pattern matching, to augment or replace the Computer Network Quality of Service (QoS) targets, as set by human experts in responding to the variable context of traffic demand arising from services carried over TCP/IP networks. The thesis presents a feasibility modelling of both black and white-box computational techniques. These complete a feature extraction, cluster analysis (as distinct from discriminant analysis where the observations are assigned to pre-arranged groups), and classification of internet working traffic not by any a priori human designation, but resulting from the inherent characteristics of the traffic. This enabled a post-process marking of these packets, using existing prioritisation techniques such as Differentiated Services Code Point, with the appropriately re-prioritised traffic being injected back into the network for onward transmission. Accordingly this research programme has completed the development and performance evaluation of the laboratory deployment of a proto-agent. The thesis then builds on this to examine a more promising agent paradigm, telo-agents, able to interact with their environment; sensing their situation and the environmental state-of-affairs; building an understanding so as to support improved performance and provide predictive analytical reasoning. These agents may enable decision-making, to the extent that the machine-learning traffic management would be better capable of modelling, the human network manager's experientially derived expert knowledge, thus mitigating organisational risk from areas such as complexity, human error and corporate memory loss; resulting from staff mobility.EThOS - Electronic Theses Online ServiceGBUnited Kingdo