Efficient Procurement and Trading of Electric Vehicle Charging Flexibility

Abstract

This thesis studies a virtual power plant (VPP) that trades the bidirectional charging flexibility of privately owned plug-in electric vehicles (EVs) in a real-time electricity market to maximize its profit. The main contribution of this thesis is the development of scalable and efficient algorithms for the procurement and scheduling of this flexibility. Specifically, to incentivize EVs to allow bidirectional charging, we design incentive-compatible, variable-term contracts between the VPP and EVs. Through deliberate aggregation of the energy storage capacity of individual EVs, we construct an abstraction of the aggregate flexibility that can be provided by the connected EVs. This abstraction is called a virtual battery and its operation is scheduled in real-time by learning a reinforcement learning (RL) policy. This policy efficiently trades the available flexibility, independent of the number of accepted contracts and connected EVs. The proposed aggregation method ensures the satisfaction of individual EV charging requirements by constraining the optimal action returned by the RL policy within certain bounds. We then develop a disaggregation scheme to allocate power to bidirectional chargers. We formulate this as a resource allocation problem, in which the total amount of energy traded in the market is distributed in a proportionally fair manner among the connected EVs. Evaluation on a real-world dataset demonstrates robust performance of the proposed method despite high variability of electricity prices and shifts in the distribution of EV mobility.