Resilient Traffic Management

This thesis explores ways to improve the self-organised urban traffic management system Organic Traffic Control by means of forecasting traffic developments and by applying machine learning techniques. Current traffic control systems typically rely on suboptimal human designed signal plans which can not respond to changing traffic demands and become outdated over time. The goal of this thesis is to transform the reactive control cycle into an anticipatory and more resilient one. To achieve this goal, the observer/controller architecture for organic technical systems is extended with a forecast module for time series. Based on monitored sensor values, forecasts of the future traffic developments help to enrich the understanding of the complex dependencies within the traffic network. Furthermore, machine learning techniques are used to improve the performance of the Organic Traffic Control system at runtime. In detail, the contributions of this thesis are as follows. To introduce resilience into technical systems, the reference design model for organic computing systems is extended by a module for forecasting of time series. This contribution enables the transformation of Organic Traffic Control from reactive to proactive adaptation. On the basis of traffic flow forecasts, an anticipatory signalisation is developed. Two routing protocols for urban road traffic guidance are transformed into time-dependent route guidance protocols including both current sensor values and forecasts for future points in time. A rule-based machine learning technique, a variant of an extended classifier system, is adapted to the problem of congestion detection, both for highways and for urban areas. Finally, approaches for a fully self-adaptive management process are outlined in which the system is enabled to react autonomously to congestion alarms