Abstract

A typical urban traffic network is a very complicated large-scale stochastic system which consists of many interconnected signalized traffic intersections. Setting signals at intersections so that the traffic in such a network flows efficiently is a key goal in traffic management. The conventional traffic signal control algorithms assume the traffic system is deterministic; most of them use data aggregation, instead of a mathematical model, and apply off-line, heuristic control strategies which do not respond to the fluctuations of the traffic flows in the network. In this dissertation, the traffic signal control problem is formulated as a decision-making problem for a stochastic dynamical system. Based on Markovian decision theory, a new decentralized optimal control strategy with the embedded platoon dispersion model is developed to minimize the queue length and the steady state delay of traffic networks. A rolling horizon algorithm is also employed to achieve real-time adaptive traffic signal control. Statistical analysis of the computer simulation results for this approach indicates significant improvement over the traditional fully actuated control, especially under the conditions of high, but not saturated, traffic demand.

Abstract

TBD

Abstract

Traditional traffic control has been based on collective stop-and-go movements for over a century. This dissertation explores the potential integration of scheduling for individual vehicle movements as a new paradigm for next-generation traffic control, which can be developed to avoid forced vehicle stoppage and queuing that is inherent in current urban traffic control. Inspired by its proven efficiency and safety in various transportation modes such as railway systems and air traffic controls, individual scheduling shows a promising perspective in urban traffic management to optimize traffic throughput, reduce traffic congestion, and enhance the overall traffic system performance. Supported by the rapid developments in driver assistance technologies and advanced real-time communication systems, such as in-vehicle indication devices and vehicle-to-everything communications, the integration of individual scheduling into urban traffic management holds the potential for improving the traffic efficiency under current traffic control schemes through eco-driving schemes, and ushering in a new paradigm of smart and efficient transportation systems in the future through the Link-based traffic control concept. Furthermore, this dissertation proposes a mathematical model, i.e., the Vehicle Tube Model, for traffic safety analysis under various vehicle behaviours.The individual scheduling is first implemented for various traffic scenarios under traditional urban traffic control management. The scheduled information for individual vehicles includes the speed and time, and each vehicle is guided by an eco-driving vehicle control approach to fulfil its scheduled information. Considering various levels and requirements on vehicle connectivity and control complexity, this dissertation proposes three vehicle control approaches that respectively provide the advisory speed, two-stage advisory speed limits, as well as the optimal acceleration rates to adjust individual vehicle movements. Each approach can be independently implemented for each vehicle to improve the speed and decrease the speed oscillations. Through a set of simulation studies, the dissertation demonstrates significant improvements in vehicle speed, fuel consumption, and emissions reduction, underscoring the benefits of adopting individual scheduling under signalized intersection controls as well as for traffic flows on a freeway after a slowly moving vehicle.With the implementation of scheduled individual vehicle movements, the dissertation introduces the innovative concept of Link-based traffic control, which represents a paradigm in contrast to the traditional node-based control such as the signalized control. The new paradigm further improves travel times and mobility and leads to smoother eco-driving through the development of optimized schemes to schedule movements that use traffic stream gaps. Emphasizing vehicle controls along the traffic links rather than at individual intersections as nodes, the Link-based traffic control schedules each vehicle movements to enable traffic flows from conflicting directions to pass through the intersections within the same period, thereby significantly enhancing the overall traffic throughput and fuel efficiency. This dissertation proposes four Link-based control models to schedule the speed and time for each vehicle when entering the intersection, and the comprehensive simulated results show that the traffic efficiency is dramatically increased with the Link-based control concept.Moreover, the dissertation proposes the Vehicle Tube Model, as a dynamic representation of vehicle movement and theory for analytical traffic safety analysis. By quantifying the risk probabilities associated with potential collisions under current and future traffic scenarios, this framework provides valuable insights into the safety performance of future urban traffic management systems. This dissertation contributes to advancing understanding of the potential benefits and challenges associated with integrating the individual scheduling and innovative traffic management concepts into urban transportation systems, and proposing a way, as a dual perspective of traffic control, for more sustainable, efficient, and safe urban mobility solutions with the intelligent transportation system.

Abstract

The purpose of this study is to develop a methodology for processing vehicle trajectory data which are presented as a series of discrete positions of vehicles recorded over consecutive time intervals. The framework combines vehicle trajectory smoothing and imputation, ensuring that speeds and higher-order derivatives of positions are consistently defined as symplectic differences in positions, while adhering to physically meaningful bounds determined by traffic laws, drivers’ behaviors, and vehicle characteristics.

To remove the outliers and high-frequency noises in speeds and higher-order derivatives, we incorporate some basic principles, including internal consistency, bounded speeds and higher-order derivatives, and minimum MAE between the raw and smoothed positions, based on physical properties and empirical observations. We propose an iterative method. One iteration comprises four types of calculations: differentiation, correction, smoothing, and integration. We adopt the adaptive average method for correction, the Gaussian filter for smoothing, and minimizing the MAEs as the objective in integration. The efficacy of the method is numerically shown with the NGSIM data. However, it is mathematically challenging to demonstrate when the iterations converge or even that the iterations can converge, leading us to develop more mathematically tractable techniques that can either be proved to converge or get rid of iterations.

We then propose a simplified iterative moving average method that makes the ranges of the smoothed speeds, acceleration rates, and jerks align with physical meaning, while preserving the average speeds or total travel distance for a specified time duration segment of a vehicle’s trajectory. Theoretically, we prove that without termination, the speed converges to a constant value after an infinite number of iterations, ensuring the termination of our method and physically meaningful ranges in speeds and their derivatives. Numerically, we demonstrate the advantages of the method in achieving physically and behaviorally meaningful ranges by applying it to the NGSIM dataset and comparing the results with manually re-extracted data and traditional filtering methods.

As another extension of the first smoothing method, We propose a two-step quadratic programming method that incorporates insights into human behavior, particularly the tendency to minimize jerks during motion, and integrates prior position errors derived from pixel length in video images. This method operates without the need for iterative processes, facilitating a single-round solution. Mathematically, we establish the existence and uniqueness of solutions to the quadratic programming problems, thus ensuring the well-defined nature of the method. Numerically, using NGSIM data, we compare the method with an existing approach with respect to the manually re-extracted ones and show the robustness of the method upon the highD data.

In addition, we investigate the scenarios involving missing portions of trajectories. In the last part of this dissertation, we consider segment scenarios where leading and trailing vehicles’ trajectories are obtainable through mobile sensors, while those of intermediate vehicles require imputation based on detected entering and exiting times from loop detectors, and propose a three-step quadratic programming method for longitudinal trajectory imputation of fully sampled vehicles. The method ensures maintaining safe inter-vehicle spacing and adheres to physically meaningful speed, acceleration, and jerk ranges. Using NGSIM and highD data, we demonstrate the great performance of the method in imputing trajectories for three-, four-, five-, and six-vehicle platoons and illustrate its successful application in capturing the true conditions of a mixed-traffic system including 10% connected vehicles (CVs) and 10% CAVs.

Abstract

Traffic advisories to travelers are based upon traffic state information at the link level. This is due to existing infrastructure which sometimes can only provide link-level information. However, the primary justification for providing link-level data is the reluctance of Traffic Management Agencies to consider more detailed traffic state data for operational and safety reasons. However, with the advances in automotive technology, sensing equipment, and the Internet of Things (IoT), we can do better. Research shows that faster and more accurate travel paths can be obtained by using lane data rather than link data. Our contention is that for vehicles to be able to change lanes to improve their travel times, operationally, they would need to enter into Peer-to-Peer negotiations with surrounding vehicles, where they can trade their position in time and space in accordance to their own perceptions of their values of time and satisfaction and possibly in exchange for monetary benefits. Our work is an exploration of this idea. We begin with a simple in-vehicle advisory control policy, partially inspired by the Kinetic theory of traffic. We then move towards an individual-level Peer-to-Peer negotiated lane change framework by first investigating its efficacy by means of microsimulation studies. We then propose an agent-based optimization framework for this system, which minimizes both travel time and the “envy” induced among drivers when they are assigned paths that are inferior to their peers. Numerical results from running our optimization on an illustrative network show that the proposed model converges to both envy-free and system optimum traffic states, even at a net zero budget, meaning this system can be used by transportation agencies without exacting tolls or giving subsidies. Our proposed framework of routing vehicles on a lane to lane basis can only be realized in the field if the mediating agency (TMC, or a mobility service) has accurate information about traffic conditions. We propose multiple algorithms, including a LSTM (Long Short Term Memory) neural network architecture-based framework to estimate traffic states solely using information collected from sensor-equipped probe vehicles, without the need for any other data such as those obtained from traditional embedded loop detectors.