Optimal Autonomous Charging of Electric Vehicles with Stochastic Driver Behavior

This paper proposes the application of the Markov decision problem (MDP) framework for optimizing the autonomous charging of individual plug-in electric vehicles (EVs). Two infinite horizon average cost MDP formulations are described, one for plug-in hybrid electric vehicles (PHEVs) and one for battery only electric vehicles (BEVs). In both formulations, we assume no direct input from the driver to the smart charger about the driver's travel schedule. Instead, we use stochastic models of plug-in and unplug behaviors as well as energy required for transportation to represent a driver's charging requirements. We also assume that electric energy prices follow a Markov random process. These stochastic models can be built from historical data on vehicle usage. The objective of the MDPs is to minimize the sum of electric energy charging costs, driving costs, and the cost of any driver inconvenience. We demonstrate the solution of the MDPs with assumed parameter values and analyze the results. This work presents a new approach to minimizing EV charging costs while reducing the need for trip planning by a driver.