Charging policies for PREYs used for service delivery: a reinforcement learning approach

Abstract

This work examines a cost optimization problem for plug-in hybrid electric vehicles (PHEVs) used for service delivery, in the presence of energy consumption uncertainty. For the cost optimization problem, an optimal policy is found that dynamically decides, as the vehicle moves, at which charging station the vehicle should be charged, in order to minimize the service fuel cost. The problem is formulated as a Partially Observable Markov Decision Process (POMDP) and is solved by applying reinforcement learning (RL). The RL charging policy (RLCP), found after solving the POMDP, is compared to two benchmark policies and it is shown that RLCP outperforms both. Most importantly, RLCP can be automatically adjusted to significant variations on the vehicle's energy consumption behavior by continuously training the RLCP model according to the most recent information obtained from the vehicle's environment.