Phd Oral Exam - Luyang Hou, Information and Systems Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Electric vehicle (EV) continues to grow rapidly due to low emission and high intelligence. This thesis considers a smart infrastructure (SI) as an EV-centered ecosystem, which is an integrated and connected multi-modal network involving interacting intelligent agents, such as EVs, charging facilities, electric power grids, distributed energy resources, etc. The system modeling paradigm is derived from distributed artificial intelligence and modelled as multi-agent systems (MAS), where the agents are self-interested and reacting strategically.

The integration, interaction, and coordination of EVs with SI components will raise various features and challenges on the transportation efficiency, power system stability, and user satisfaction, as well as opportunities provided by optimization, economic, and control theories, as well as other advanced technologies to engage more proactively and efficiently in allocating the limited charging resources and collaborative decision-making in a market environment. A core challenge in such an EV ecosystem is to trade-off the two objectives of the smart infrastructure, of system-wide efficiency and at the same time the social welfare and well-being against agents’ selfishness and collective behaviors. In light of this, scheduling EVs' charging activities is of great importance to ensure an efficient operation of smart infrastructure and provide economical and satisfactory charging experiences to EV users under the support of two-way flow of information and energy of charging facilities.

In this thesis, we develop an advanced mechanism design framework to optimize the charging resource allocation and automate the interaction process across the overall system. The key innovation is to design specific market-based mechanisms and interaction rules, integrated with concepts and principles of computational mechanism design, scheduling theory, and reinforcement learning, for charging scheduling and dynamic pricing problems in various market structures.Specifically, this research incorporates three synergistic areas: (1) Mathematical modelling for EV charging scheduling. We have developed various mixed-integer linear programs for single-charge with single station, single-charge with multiple stations, and multi-charge with multiple station in urban or highway environments. (2) Market-based mechanism design. Based on the proposed mathematical models, we have developed particular market-based mechanisms from the resource provider’s prospective, including iterative bidding auction, incentive-compatible auction, and simultaneous multi-round auction.

These proposed auctions contain bids, winner determination models, and bidding procedure, with which the designer can compute high quality schedules and preserve users’ privacy by progressively eliciting their preference information as necessary. (3) Reinforcement learning-based mechanism design. We also proposed a reinforcement mechanism design framework for dynamic pricing-based demand response, which determines the optimal charging prices over a sequence of time considering EV users’ private utility functions. The learning-based mechanism design has effectively improved the long-term revenue despite highly-uncertain requests and partially-known individual preferences of users.

This PhD dissertation presents a market prospective and unlocks economic opportunities for MAS optimization; furthermore, applies AI techniques to facilitate the evolution from manual mechanism design to automated and data-driven mechanism design when gathering, distributing, storing, and mining data and state information in SI. The proposed advanced mechanism design framework will provide various collaboration opportunities with the research expertise of reinforcement learning with innovative collective intelligence and interaction rules in game theory and optimization tools, as well as offers research thrust to more complex interfaces in intelligent transportation systems, smart grids, and smart cities.