Risk in traffic systems

Abstract

Safe and efficient traffic systems constitute a cornerstone of modern life. This thesis provides a comprehensive treatment of the topic of risk in traffic systems. Based on an interdisciplinary research approach, novel models and methodologies are developed to investigate the impact of individual driving style, technological innovation, and traffic system design on safety and efficiency. For this purpose, perspectives and techniques from traffic modeling and insurance mathematics are combined.

First, we develop a microscopic traffic model that can describe the occurrence of traffic accidents due to random misperception, a type of error that is relevant for both human drivers and sensors of autonomous vehicles. The model allows us to characterize the real-world tradeoff between safety and efficiency in case studies.

Second, we generalize the microscopic traffic model to study the important scenario of an unsignalized urban intersection, a particularly accident-prone area. We apply the concept of random misperception to model the occurrence of accidents and discuss the numerical solution of the random ordinary differential equations involved using state-of-the-art methods. In case studies, we analyze the impact of driving styles on different types of conflicts; we also consider the important case of heterogeneous traffic participants where human drivers and autonomous vehicles coexist.

Third, we devise a methodology that makes microscopic traffic models accessible for a statistical study of traffic accidents and corresponding financial losses. The approach enables comprehensive risk management for traffic systems: We study the impact of changes in the design of vehicles and transport systems on functionality and road safety, and price insurance contracts that cover residual risks.

Fourth, we complement the above microscopic approaches with a macroscopic perspective: We investigate stochastic cell transmission models of traffic networks. The performance of traffic systems is evaluated based on preference functionals and acceptable designs. The numerical implementation combines simulation, Gaussian process regression, and a stochastic exploration procedure.