Optimal Reservation Scheduling and Online Admission Control for Electric Vehicle Charging Stations in Microgrids

Abstract

The rapid adoption of electric vehicles (EVs) used worldwide introduces an environmental benefit by reducing the emission of greenhouse gasses. Presently, the most significant drawbacks of encouraging further EV adoption are long charging hours and insufficient charging facilities. This problem can be tackled by strategically planning charging station locations. However, an infrastructure upgrade would require an expensive monetary investment. The alternative option is to adopt an intelligent charging scheme utilizing the available chargers. Therefore, the latter would be the focus of this research. Smart Grid integrates advanced sensing technologies, control methods, and bidirectional communications into the current electricity network. These features support the growing penetration of EVs by allowing charging control utilizing intelligent meters. Moreover, these technologies introduced the electric vehicle charging scheduling problem (EVCSP), which aims to generate a schedule of charging activity for a fleet of EVs. The charging decision can be binary, i.e., to charge or not, or a continuous control variable in the optimization problem. In addition, the problem can be designed to benefit one of three stakeholders, namely the grid, the service provider, or the EV user. The service provider can decide the charging decision in a centralized control manner. EVCS can be addressed as offline optimization, i.e., day-ahead planning where the future load is known. However, as the EV load is highly unpredictable, making the problem time sensitive, online optimization is preferable. The literature studies the utilization of many solving techniques such as commercial solvers, heuristics, approximation algorithms and deep learning. The main objective of this project is to model and solve an EVCSP. The model aims to simulate a realistic setting by utilizing a real-world dataset of charging activity in an Adaptive Charging Network at Caltech lot and the average load profile in southern California to represent the baseload. Furthermore, the optimization problem's is to maximize users' social welfare while considering the network requirements once the proposed algorithm is employed. Moreover, the constraints do not allow exceeding the maximum apparent power. Lastly, the problem is solved in two stages. The first stage computes the reservation scheduling using an a provably efficient algorithm (with certifiable and tunable degree of optimality) and deep learning in an offline setting and inputs them to the second stage—the second stage provides a real time admission control by employing a heuristic algorithm in a receding horizon manner.

Date of Award	Jul 2022
Original language	American English