Methods for switching time prediction at traffic-dependent light signal systems.

Motorized road traffic is the dominant source of emissions in the German transport sector. In urban areas in particular, stopping, accelerating and braking in front of traffic signals are responsible for avoidably higher emissions. Adjustments to driving behavior with the aim of avoiding these movement states make a contribution to climate protection. So-called GLOSA or traffic light assistance systems (GLOSA: Green Light Optimal Speed Advisory) aim to provide vehicle drivers with the necessary information, such as recommendations on an optimal approach speed to traffic lights. In addition to the vehicle position, the timing of future changes in the signaling states of the respective traffic signal system is required as the basis for generating these driving recommendations. Fixed-time controllers have a deterministic switching behavior, thus their future signaling states are easily predictable. In contrast, the control units of traffic-dependent light signal systems take into account various parameters about the current traffic situation in the intersection environment when selecting and maintaining signaling states, which means that future states cannot be predicted by simply analyzing the temporal switching behavior of the signaling programs.

Within the scope of this project, we pursue the development of a generally applicable, largely automated and modular procedure for the prediction of the switching times of traffic-dependent light signal systems. For the development and evaluation of the method, we use recorded process data of real traffic signal systems from the city of Kassel, which are kindly made available to us by the Road and Civil Engineering Department of the city of Kassel. The project focuses on the following research questions:

• How can a generally applicable concept for switching time prediction look like, which takes into account the system knowledge for signal program design as well as for system operation and which can integrate different types of traffic signal control systems?
• Which data models are useful for which control types to incorporate system knowledge into predictions?
• Which methods are suitable for the predictions of control-relevant parameters?
• What improvement can be achieved by integrating process data from neighboring LSAs?