UC Berkeley

A smart city is a digitized urban area that aims at improving a quality of life by utilizing abundant electronic datasets. This dissertation is written from the perspective of a control engineer and introduces advanced optimization and control problems uniquely positioned in smart city applications.

We focus on two major components of the smart city: (i) smart mobility and (ii) smart grid. Across the components, we also consider a very important factor: human. Addressing human behaviors and their impact to the system raise technical questions that are yet to be answered. Through the dissertation, we identify unique and unanswered problems that advanced control and optimization theories can solve. Ultimately, this dissertation provides a tool-set of modeling, optimizing, and analysing modern problems in the scope of smart mobility and electrification, with consideration of human behaviors. A brief description on each chapter is as follows.

In Chapter 1, we provide an overview of an emerging concept of the smart city. A high-level overview on technical challenges of the smart city, in particular on the mobility and electrification perspectives, are followed in detail. The chapter concludes with a summary of technical contributions of this dissertation.

In Chapter 2, at a junction of smart mobility and electrification, we explore the effectiveness of a vehicle-to-infrastructure connectivity in terms of an energy management. In particular, we develop a motion planner that saves fuel and energy consumption of a Plug-in Hybrid Electric Vehicle (PHEV) by leveraging traffic light information. The motion planner then considers uncertainties caused by other (human) drivers in actuating dynamic traffic light schedules and in queuing at intersections. Designing the motion planner requires a non-linear program which is typically hard to solve with real-time computation efficiency. Hence, we examine a hierarchical framework and approximation methods that realize a real-time implementation. Importantly, this work performs on-road tests of an automated vehicle controller that communicates with traffic lights and the test results indicate the significant potential of vehicle-to-infrastructure connectivity for saving fuels.

In Chapter 3, we explore human-interactions in urban driving scenarios, especially under a dense traffic. Surprisingly, only few publications in control systems have addressed the dense traffic scenarios albeit their commonness and significance in urban driving. We develop a mathematical framework based upon a receding horizon control scheme, which evaluates human-interactions at each receding control horizon. The resulted framework considers other drivers' cooperative behaviors, e.g., slowing down in response to a lane-changing attempt. A state-of-the-art Recurrent Neural Network design is incorporated to estimate a complex human decision making mechanism. Incorporating neural networks yields additional complexity to the controller, and hence we develop a numerical method that find solutions efficiently. We testify the framework particularly to a mandatory lane-changing scenario in a highly dense traffic, where a cooperation of other drivers is a key to change lanes timely. Although it is validated in urban driving scenarios, this work can find a broad range of applications where interactions between arbitrary agents can be leveraged to facilitate system performances.

In Chapter 4, from a broader perspective, we examine a generic mathematical framework that models human as a stochastic actuator in the system. In particular, we add a new perspective of addressing human behaviors in control systems to the literature, by considering human behavior as an inducible variable. Human behaviors may impede a system operation or, in contrast, may help improve a system performance. And their impact can be adjusted by an incentive that a system operator can determine. Examples include giving a price discount (controllable variable) to induce a purchase on a specific item (induced behavior) and giving a credit (controllable variable) to induce a participant to a regulation service in Energy systems (induced behavior). Simple case studies are included to show the potential of the framework. This work is the first attempt that formally incorporates human behaviors as an ``inducible variable" into control systems.

In Chapter 5, we further explore a potential of inducing human behaviors in the sophisticated application of operating a vehicle charging station. Existing operation strategies of charging facilities typically provide a charging service with a constant charging power level, without exploring either various charging option choices or the chance of losing customers. Hence, in this work, a controller is designed to optimally determine price and charging power policies, considering the consequent impact on the choice of charging services. Additionally, the controller manages an overstay problem that has arisen as a major problem in the urban charging infrastructure. The chapter extends with detailed analysis on the effectiveness of the incentive controls that would attain interests from the readers beyond the scope of energy systems.

In Chapter 6, a summary of contributions of the thesis is documented. Also, perspectives for future works are extended.

In summary, this dissertation identifies and solves uniquely positioned problems that the smart city is faced with. The proposed methodologies and high-level ideas can be very beneficial in the smart mobility systems, operation systems, and energy applications. Each chapter is self-contained and independent (referenced otherwise). Hence, readers may jump to any chapters depending on their interest without having to digest one as a pre-requisite.