Dynamic multilevel traffic simulation for urban traffic considering human driver behavior

Abstract

Traffic simulation serves as a valuable tool for evaluating and predicting transportation systems, and its evolution is influenced by key issues. In this study, particular emphasis is placed on the `human factor' and `urban traffic' as important considerations in the development of the simulation. Dynamic traffic phenomena, such as hysteresis, stop-and-go traffic, stability, and capacity drop, are closely associated with traffic safety and congestion. While previous studies suggested that human behavior caused these phenomena, existing models have been unable to fully replicate these dynamic traffic phenomena. Research in this area has primarily focused on microscopic simulation, particularly in the car-following models that simulate the longitudinal movement of vehicles. On the other hand, urban traffic differs significantly from highways due to complex road structures, including intersections and traffic signals. Additionally, an integrated evaluation that considers the impact on overall traffic, as well as the target area, requires a large-scale simulation. To obtain detailed results for a target area through large-scale simulations, hybrid or multilevel traffic simulations have been devised, integrating different levels of simulation. Existing multilevel traffic simulations generally take a static format where the simulation level remains fixed. However, this approach is inefficient in urban areas where targets can be easily dispersed or shifted. To address this limitation, a dynamic multilevel traffic simulation is proposed, allowing for the dynamic adjustment of the simulation level during the simulation, thereby expanding the utility of urban simulation. Therefore, this study aims to develop a car-following model that incorporates human factors, a mesoscopic traffic simulation model for urban traffic, and a dynamic multilevel traffic simulation that combines microscopic and mesoscopic levels. First, we develop an asymmetric repulsive force model (ARM), a novel car-following model that considers the driver's asymmetric behavior and psycho-physical properties. The ARM is evaluated in diverse aspects and the results describe that the ARM accurately describes vehicles in stop-and-go traffic. In particular, the ARM's ability to explain asymmetric behavior has been demonstrated through results in close agreement with the observed data in the speed-spacing and spacing-relative speed domains. In addition, a platoon simulation replicates the flow reduction after a stop-and-go wave, consistent with previous studies on the capacity drop. In the second phase of the research, we present the urban cell transmission model in mesoscale (UCTM), which extends the cell transmission model into a mesoscopic form. The UCTM incorporates the concept of agents to account for lane changes and intersection-related behaviors. Extensive simulation studies verify and validate the UCTM with similar results to the actual data in various aspects, although there are still areas to be improved in the future, such as lane changes, human driver behavior in urban traffic, and discrete traffic flow. On the other hand, the UCTM exhibits compatibility advantages in multilevel models due to its morphological characteristics, which integrate the concept of an agent with a macroscopic model. Finally, a dynamic multilevel traffic simulation is developed by combining microscopic and mesoscopic traffic simulations. A simulation framework and data structure are proposed to ensure compatibility and consistency between different levels of simulations. Temporal and spatial interfaces are modeled and verified for proper functioning, including the preservation of vehicle information and consistency in traffic dynamics. The performance of the simulation is evaluated in terms of computational expenses and accuracy using two demand scenarios. The dynamic multilevel simulation dramatically reduces computation time compared to the microscopic simulation while showing higher accuracy than the mesoscopic model. This dissertation contributes to developing a dynamic multilevel traffic simulation for urban traffic, considering human driver behavior and addressing current challenges faced in traffic simulation. The findings of this study are expected to have practical applications and contribute to the development of innovative solutions for addressing complex urban traffic issues.