Stochastic Modeling and Optimization in Electric Vehicle Networks
Room 2408 (Lifts 17-18), 2/F Academic Building, HKUST

Thesis Examination Committee

Prof Kai TANG, MAE/HKUST (Chairperson)
Prof Danny Hin Kwok TSANG, ECE/HKUST (Thesis Supervisor)
Prof Vincent WONG, Department of Electrical and Computer Engineering, The University of British Columbia (External Examiner)
Prof Chin-Tau LEA, ECE/HKUST



Electric vehicle (EV) networks are the infrastructure system of public EV refueling stations and serve as the vital joint nodes between the modern power and transportation systems. On the one hand, EV networks are large energy consumers, who are motivated to control their charging loads strategically to reduce their operational cost. On the other hand, EV networks are refueling service providers, and hence are required to meet a minimum quality-of-service (QoS) for EV customers. This thesis develops sequential-decision-making models and algorithms for EV networks to best trade off their operational cost and QoS. We primarily focus on addressing algorithmic challenges arising from uncertainties in refueling requests of EVs.


For the coordinated over-night charging of slow charging stations (SCSs), we propose to iteratively solve a static non-linear binary program based on model predictive control. To cope with the computational challenge in solving the large-scale binary programs, we leverage the problem structure to transform them into equivalent linear programs, which are efficiently solvable. For the coordinated charging in public SCSs, we propose an online mechanism, in which all EVs submit their truthful bids, the loss of social welfare is bounded by competitive ratio, and the mechanism can be implemented in polynomial time.


For a battery swapping and charging station (BSCS), we formulate the charging problems as Markov decision processes (MDPs) under both stationary and non-stationary environments. The algorithmic challenge is to solve the MDPs in a computationally tractable manner for large-scale systems. In the stationary case, we discover the order-up-to type structure of the optimal policy and design a subgradient algorithm to search for the thresholds efficiently. Under the non-stationary environment, we develop a fluid-based approximation for the MDP to facilitate the analysis. Managerial insights are observed for both battery purchasing and charging in BSCSs.

Room 2408 (Lifts 17-18), 2/F Academic Building, HKUST