Application of ant based routing and intelligent control to telecommunications network management
This thesis investigates the use of novel Artificial Intelligence techniques to improve the control of telecommunications networks. The approaches include the use of Ant-Based Routing and software Agents to encapsulate learning mechanisms to improve the performance of the Ant-System and a highly modular approach to network-node configuration and management into which this routing system can be incorporated. The management system uses intelligent Agents distributed across the nodes of the network to automate the process of network configuration. This is important in the context of increasingly complex network management, which will be accentuated with the introduction of IPv6 and QoS-aware hardware. The proposed novel solution allows an Agent, with a Neural Network based Q-Learning capability, to adapt the response speed of the Ant-System - increasing it to counteract congestion, but reducing it to improve stability otherwise. It has the ability to adapt its strategy and learn new ones for different network topologies. The solution has been shown to improve the performance of the Ant-System, as well as outperform a simple non-learning strategy which was not able to adapt to different networks. This approach has a wide region of applicability to such areas as road-traffic management, and more generally, positioning of learning techniques into complex domains. Both Agent architectures are Subsumption style, blending short-term responses with longer term goal-driven behaviour. It is predicted that this will be an important approach for the application of AI, as it allows modular design of systems in a similar fashion to the frameworks developed for interoperability of telecommunications systems.