Intelligent Traffic Signal Control Framework in a Connected Vehicle Environment with the Development of a Real-Time Traffic State Estimation Model using Deep Learning

Abstract

Traffic signal control systems were developed to regulate the traffic flow at roadway intersection to improve the traffic mobility and safety. Over the past years, traffic signal control systems have experienced major developments due to the advancement in the communication and computation areas. Connected Vehicle (CV) technologies, such as vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication systems, are emerging technologies envisioned to develop much safer transportation and intelligent traffic control systems to improve the traffic mobility, safety, and environmental sustainability.A new generation traffic signal control system, MMITSS, was developed to provide priority to special modes of vehicles including emergency vehicles, transit vehicles, and trucks based on connected vehicle technologies. The current state of the MMITSS does not directly mitigate the negative impacts on regular vehicles or address safety concerns while making priority control decisions. MMITSS requires that the host controller run in “free” mode so that the external NTCIP commands (e.g., CALL, HOLD, FORCE-OFF, and OMIT) can be applied to make strategic priority control decisions. Due to this, traffic engineers are often reluctant to deploy MMITSS since they cannot run their popular traffic control tool coordination. This dissertation focuses on the enhancement of MMITSS to explore the use of connected vehicle data to improve safety by protecting trucks that are trapped in the dilemma zone when emergency vehicle preemption is requested, to develop a coordination algorithm to replicate traditional traffic signal controller behavior, and to explore the use of artificial intelligence (AI) approaches to estimate traffic state in real-time to explicitly mitigate the negative impacts on regular vehicles in the optimization model when the connected vehicle penetration rate is low. The existing mathematical model (mixed-integer linear programming (MILP)) of MMITSS is enhanced to explicitly consider multiple priority requests from the emergency vehicles and heavy vehicles that are trapped in the dilemma zone. The optimization model provides an optimal signal timing schedule that minimizes the total weighted priority request delay and dilemma zone request delay, as well as to provide some provide flexibility to adapt to other vehicles in real-time. The simulation experimental results indicate that the algorithm can reduce conflicts between the emergency vehicles and trucks that are trapped in the dilemma zone as well as the negative impact on regular traffic on the minor-streets. The MMITSS emergency vehicle priority system has been implemented in the Maricopa County SMARTDrive ProgramSM test bed in Anthem, Arizona since September 2020. Priority-based coordination was developed that provides preferential treatment to the vehicles traveling along a coordinated route. The mixed-integer linear programming model (MILP) of MMITSS is enhanced to consider coordination as a form of priority along with the multi modal priority for eligible emergency vehicles, transit vehicles, and freight vehicles, and provide dilemma zone protection to freight vehicles during emergency vehicle preemption in a connected vehicle environment. The mathematical model generates an optimal signal timing schedule that minimizes the total weighted delay of the coordination requests, priority requests, dilemma zone requests, and maximizes the flexible implementation of the optimal schedule. The simulation experiments, statistical analysis, and field results show that priority-based coordination can replicate traditional coordination behavior of a traffic signal controller. The MMITSS multi-modal priority with priority-based coordination system was implemented in the Maricopa County SMARTDrive ProgramSM test bed in Anthem, Arizona and in Portland, Oregon. An intelligent traffic signal control (I-Sig) system was developed that optimizes the travel time delay for all vehicles travelling through a signalized intersection. A traffic state estimation model (DNN-TSE) is developed using deep learning techniques to estimate traffic state in real-time based on the connected vehicle trajectory data and connected intersection SPaT data. The optimization model of I-Sig is dynamic programming based and developed to take the output (arrival table) of the DNN-TSE model as input. The I-Sig algorithm follows a rolling horizon approach to accommodate new traffic information that collected over time. The simulation experimental results indicate that DNN-TSE learns about the traffic flow distribution and can estimate the traffic state of the non-connected vehicles in the queueing region and slow-down region quite well when market penetration rate is low (e.g., 10%). The numerical experiments demonstrate that I-Sig can reduce the travel time and delays of regular vehicles while providing priority to the special types of vehicles.