Optimizing EV Routing and Charging/Discharging under Time-Variant Electricity Prices

Abstract

The integration of electric vehicles (EVs) and the power system has been becoming an increasingly important field of research, due to the rapid EV penetration and the evolvement in vehicle-to-grid (V2G) techniques in the past decade. Under appropriate management of EV charging and discharging, the current grid capacity can satisfy the energy requirements of a considerable number of EVs and EVs could help enhance grid reliability and stability through ancillary service provision.

In this thesis, we investigate the operational strategies of commercial EV fleets under the V2G context where energy price signals are utilized to incentivize EV owners to time-shift charging schedule and discharging EVs during peak hours. We propose and formulate a new EV routing problem with time windows under time-variant electricity prices (EVRPTW-TP), considering practical constraints of commercial EV fleets providing logistic services and optimizing over its overall electricity cost. In order to solve the EVRPTW-TP, we then formulate a Lagrangian relaxation as well as a variable neighborhood search and tabu search hybrid (VNS/TS) heuristic to approximate the optimal solution from below and above respectively. Our numerical experiments on small instances suggest that both algorithms are able to provide high quality bounds to the EVRPTW-TP. The VNS/TS heuristic outperforms CPLEX in terms of solution quality and efficiency on instances of $10$ or more customers. In addition, we utilize the VNS/TS heuristic to study a use case of an EV fleet providing grocery delivery service in the Kitchener and Waterloo (KW) region in Ontario, Canada. Insights about the impacts of energy pricing scheme, service time slot design as well as fleet size are presented. Finally, as the first step towards implementing advanced machine learning techniques to solve the EVRPTW-TP, we develop a reinforcement learning (RL) model for the electric vehicle routing problem with time windows (EVRPTW) and provide computational results to assess the performance of the RL model.