Dynamic Connected Automated Vehicle Trajectory and Traffic Signal Timing Optimization

Abstract

This dissertation addresses the topic of sustainable transportation in the context of traffic signalized networks by optimizing both vehicle and traffic signal operations. From the vehicle side, the research develops a Green Light Optimal Speed Advisory (GLOSA) system also known as an Eco-Cooperative Adaptive Cruise Control system at intersections (ECO-CACC-I) for fixed and actuated traffic signals, using probabilistic traffic signal switching times to minimize vehicle fuel consumption through a computationally efficient A* minimum path algorithm within a dynamic programming procedure. This system explicitly minimizes vehicle fuel consumption while ensuring vehicle safety by preventing red light violations, hard braking, and excessive jerking. A sensitivity analysis is performed to quantify the impact of uncertainty in traffic signal timing predictions on fuel consumption. The research also extends the ECO-CACC-I system to integrate real-time back-of-the-queue estimation using loop detector and probe vehicle data with trajectory optimization, considering uncertainties in actuated traffic signal timings. The system enhances queue length estimates without relying on historical data and significantly improves upon the sole use of shockwave theory for the estimation of queues. On the infrastructure side, a cycle-free dynamic Decentralized Nash Bargaining (DNB) traffic signal controller is developed to optimize traffic signal operations using traffic stream density predictions, a flexible National Electrical Manufacturers Association (NEMA) phasing scheme, and dynamically adaptable control time steps. The DNB controller is benchmarked against fixed-time, actuated, and reinforcement (RL) machine learning (ML) control methods demonstrating its superior performance and simple algorithmic formulation. Furthermore, a two-stage Kalman filter algorithm is developed to predict traffic states for real-time traffic signal control, with the first stage estimating turning movement counts and the second stage estimating queue sizes and traffic stream density on the intersection approaches. This Kalman filter approach is integrated within the DNB controller to predict and optimize traffic signal timings in real time. The development of these vehicle and infrastructure systems aims to reduce vehicle energy consumption and emissions while improving traffic mobility. The proposed ECO-CACC-I system achieves average fuel savings of 37% and 30% for deterministic and stochastic settings respectively, compared to uninformed drivers. Furthermore, the ECO-CACC-I system demonstrates fuel savings of up to 18.89% when considering queue effects. The DNB traffic signal controller reduces average vehicle delay and queue sizes by up to 54% and 63%, respectively, compared to the state-of-the-practice Webster's pre-timed control. Finally, the analysis of the joint DNB-KF system showed benchmarks for the market penetration levels of connected vehicles that are required to achieve significant system performance.