Adaptive Learning and Optimal Control Methods for Developing Intelligent Transportation Systems

In recent years, intelligent transportation systems (ITSs) have a significant impact on human life as ITSs improved transportation safety and mobility. The development of ITSs lies in many different aspects. In this dissertation, an adaptive based learning algorithm and optimal control schemes are used in ITS development. The two major areas of focus are the hybrid electric vehicle (HEV) fuel-saving problem and the traffic signal control optimization problem, where both vehicle level and traffic system level are considered in the design.

HEVs have been an effective solution for improved vehicle fuel efficiency and reduced emission pollution. However, the optimization design for HEVs is complicated due to the presence of nonlinear dynamics and complicated integration of the HEV systems. Moreover, there is a trade-off between fuel optimization and emission reduction. In Chapter 2, a co-optimization scheme is proposed to optimize fuel efficiency for HEVs. The proposed optimization scheme uses vehicle to vehicle (V2V) and vehicle to infrastructure (V2I) information as the basis to optimally tune control parameters for the existing powertrain control system. Moreover, the speed of catalyst temperature to reach its light-off level in the exhaust emission system is also considered as an additional optimization constraint to reduce emission. It has been shown that a further 9.22% fuel savings can be achieved on average for a Toyota Prius test model.

Traffic signal control is important for intersection safety and efficiency. However, most traffic signal control methods are designed for individual intersections or corridors. Although some adaptive control systems have been developed, the methods used are often proprietary and not published, making it difficult to evaluate their effectiveness. The goals of our research are to identify the unknown traffic dynamic and then develop some efficient control schemes to achieve minimal traffic delay time and smooth traffic. To met the above goals, in Chapter 3 to Chapter 5, three different approaches are developed. In Chapter 3, an adaptive linear-quadratic regulator (LQR) is designed to minimize both traffic delay and incremental changes in the control input, which is based on linear system approximation. In Chapter 4, a traffic signal optimal control scheme with an adaptive on-line learning scheme using multiple-model neural networks is designed to achieve traffic delay minimization. In this work, a non-linear neural network model is used to represent the unknown traffic dynamics. In Chapter 5, the above problem is solved from another aspect, where the goal is to have the distributions of every intersection follows a target distribution. To this end, a stochastic traffic signal control algorithm is designed. The above three methods are not only practically meaningful for traffic signal control design but also have some theoretic contributions to the field of adaptive learning techniques.