The application of artificial neural networks to the control of a road traffic monitoring system
This thesis describes the development of a new Road Traffic Monitoring (RTM) system designed by a team at the University of Aberdeen, to provide low cost traffic monitoring that maintains the privacy of the driver and provides high quality traffic data for the authorities. The RTM system communicates with vehicles equipped with satellite transceivers attached to Global Positioning System (GPS) units which provide data about position and velocity, enabling a picture to be built up of traffic flow conditions. The author developed an innovative system to maintain the anonymity of the driver in order for the system to be attractive to both authorities and drivers. The thesis focuses on the author's contribution to the project - the design of the major protocols, including polling and 'system wide' strategies, an in-depth study of the statistics of the capacity of the RTM system and the development of a protocol to limit the number of responses from vehicles to a manageable level. The RTM system was implemented as a simulation and all designed protocols and parameters thoroughly tested. It is currently being used by the European Space Agency in a pilot study as the first phase of implementation. The author then envisaged an intelligent control system that would monitor and control the RTM system in real time, optimising parameters dynamically in order for the system to run efficiently and accurately. A hybrid 'Artificial Intelligence' (AI) system composed of a rule based system and a number of Artificial Neural Networks (ANNs) is proposed. A new rule extraction system developed by the author to counter the 'black box' effect of ANNs is then presented, making use of a hierarchical system of Self Organising Maps (SOMs). Following this, an extension to the system is described, using the Quantisation Error (QE) to enable the real time detection of 'unusual' data, or data from outwith the current operating environment, enabling retraining or the return to a previous control situation, to take place. Such a system of ANNs is ideal for use in both monitoring and control of a dynamic real time system such as RTM.