Congestions and rear-end crashes are two undesirable phenomena of freeway traffic flows, which are interrelated and highly affected by human psychological factors. Congestions on freeways increase rear-end crash risk, and rear-end crashes can initiate or aggravate congestions. Since congestion are everyday problems, and crashes are rare events, congestion management and crash risk prevention strategies are often implemented through separate research directions. As a result, traffic control studies focusing on increasing efficiency may increase the risk of rear-end crashes, whereas those focusing on improving traffic safety may not necessarily be desirable from efficiency perspective. Both freeway traffic flow and safety management will be more challenging in the era of connected driving. In connected environments, the role of human psychological factors on the traffic flow dynamics and traffic safety will be much more pronounced.
The motivation behind this Ph.D. research is to pave the way for traffic management scenarios that result in more efficient and safer freeway traffic in the era of connected driving. As such, the research aim is to develop a robust understanding of freeway traffic flow dynamics and their safety implications with respect to human psychological factors. This research selects the continuum framework to understand traffic flow dynamics and proactive safety assessment framework to understand the safety implications of traffic flow dynamics.
A comprehensive and critical literature review is conducted to understand the state-of-the-art of continuum traffic flow models. The review effort aims to obtain a robust knowledge about existing discussions and debates over continuum models’ analytical properties and real-world performances. A major part of the review explores the research gaps in continuum models for the era of connected and automated vehicles (CAVs). It is found that none of the existing continuum models can describe the role of complex human psychological factors (e.g., risk perception) on traffic flow dynamics. As well, the critical review revisits the properties and issues with continuum models for both conventional and CAV traffic flows. The review aims to take a close look on a wide range of theoretical, practical, and behavioral issues that must be kept in mind when developing new continuum models.
Next, this research conducts a comprehensive benchmarking study on single-pipe continuum models by using traffic data from the German A5 autobahn. Model families are examined based on the review effort, and suitable representative models are selected within the families. A set of benchmarking criteria is designed, ranging from the operational measures (e.g., delay and travel time) as well as the complex traffic phenomena (e.g., scattering, oscillations, capacity drop, and hysteresis phenomena). Suitable real-world traffic scenarios are carefully selected concerning the benchmarking criteria, and the selected models are comprehensively assessed for real-world scenarios.
Based on the understanding obtained from the review and benchmarking efforts, a novel behavioural continuum model (Non-Equilibrium Traffic model based on Risk Allostasis Theory, i.e., NET-RAT) is developed. NET-RAT fills a huge gap in the literature, that is, lack of behavioural continuum models with respect to a well-established human factor theory. Perceived risk and preferred response time are selected as the major human psychological factors of the conventional and connected environment. Perceived risk is defined in terms of the proportion of stopping distance which is a car-following safety measure. The selected human psychological factors are incorporated into the full velocity difference car-following model (FVDM). A novel continuum model is derived from the extended FVDM, and the interplay between preferred response time and perceived risk is formulated within the well-established risk allostasis theory. The analytical properties of NET-RAT are explored in detail, and its relationship with the notable existing continuum models are discussed. NET-RAT’s performance against real-world traffic is investigated comprehensively and compared with the models studied in the benchmarking effort.
The results demonstrates that NET-RAT outperforms the existing continuum models in the benchmarking effort regarding some aspects of real-world traffic, e.g., travel time estimation and qualitative propagation of jam front. It is shown that drivers that have higher perception of risk can aggravate congestions by increasing shockwave speeds. Such drivers, however, can stabilize traffic flow regardless of traffic conditions due to responding quickly to initial perturbations. On the other hand, those with lower perceptions of risk can reduce congestions, but can initiate traffic instabilities in the intermediately congested states due to increased response time. This research also explores the implications of NET-RAT for the environments, where drivers are provided with information about traffic and their risk perception may be affected.
Next, this dissertation proposes a hybrid methodological framework combining probabilistic and machine learning models to develop the relationships between safety and macroscopic state variables within a flexible conflict-based safety assessment framework. Time spent in conflict is introduced as the total time spent by all vehicles in rear-end conflicts, where the conflict instances are defined based on the proportion of stopping distance and a flexible threshold. The proposed hybrid framework can assess the time spent in conflict for all underlying car-following interactions using only macroscopic state variables, and thus, overcoming the need for trajectory data. Besides, it provides an endogenous safety dimension to the fundamental relations of freeway traffic flows that can be utilized to balance freeway traffic flow efficiency and safety. For instance, control studies can utilize the proposed framework to minimize total travel time while also minimizing total time spent in conflict for crash-prone situations such as shockwaves and traffic oscillations.
Finally, the proposed safety assessment model is utilized in conjunction with NET-RAT to investigate the safety implications of various driving behaviours in relation to risk perception. It is shown that drivers with low perception of risk spend more time in conflict rear-end conflicts during critical situations from safety perspective, e.g., when entering shockwaves and when undergoing stop-and-go waves.
The proposed methodology in this Ph.D. is a pathway for unifying freeway traffic flow modelling and rear-end crash prevention in the era of mixed traffic, when human factors play significant role on both congestion and rear-end crashes.