Book Description

Recent years, automated vehicle (AV) technology, which is expected to solve critical issues, such as traffic efficiency, capacity, and safety, has been put a lot of efforts and making considerable progress. There is another technology called connected vehicle (CV) which connect vehicles through dedicated short-range communication devices. Combining the AV technology and CV technology leads to the more comprehensive connected and automated vehicle (CAV) technology. Although some of the car industry companies, such as Tesla, Waymo, has made great progress in developing CAV, it is still hard to realize commercial use due to the safety issue and cost issue. It seems CAV is not the solution for autonomous in near future. Thus, another innovating technology has been brought into the public's view which is connected automated vehicle highway systems (CAVH). CAVH provides a safer, more reliable, and more cost-effective solution by redistributing vehicle driving tasks to the hierarchical traffic control network and roadside unit (RSU) network. But the cost of a full CAVH system is still too high for commercial use. As a result, a new system has been brought into discuss which is the Partially Instrumented CAVH (PI CAVH). The PI CAVH network facilitates sensing, prediction, decision making for low automated level vehicles (Level 2 CAV) in the areas which involving heavy weaving activities, on/off ramp, work zones, etc. The PI CAVH is considered as a feasible solution for the commercial use of autonomous driving. However, even with the implementation of PI CAVH, human driving vehicles (HDV) will still dominate the road in the near future. Therefore, to find a proper platoon level car following strategy for CAVs under PI CAVH will be a challenging problem. Due to the lack of empirical data, we have to simulate the scenarios under PI CAVH. The current simulation platform cannot reproduce realistic HDV trajectories (especially of different driving styles). The deep learning techniques have demonstrated promising capability in traffic trajectory generation. Neural Networks are widely applied in the research of the car-following model. Among those networks, long short-term memory neural networks (LSTM) is the most used and has great potential for car following behavior modeling. This research focuses on establishing a car following model that can represent various driving styles and generate large numbers of realistic HDV trajectories with the help of deep learning techniques. The proposed model will help us to determine the performance of different car following strategy for CAVs under PI CAVH. This dissertation first reviews on car-following models and CAV control algorithms. Then a unidirectional interconnected LSTM car following model with heterogeneous driving style is established to generate numerous trajectories to simulate scenarios under PI CAVH. Several experiments are carried out to analyze the performance of different car following strategies.