Electric vehicle fleet charging management: An approximate dynamic programming policy

The growing prevalence of electric vehicles (EVs) requires efficient charging management strategies to tackle the challenges associated with their integration into the power grid. This requirement is particularly true for Charging-as-a-Service (CaaS) providers, who manage charging services for fleet operators in exchange for a fixed service fee. Incorporating uncertainty into optimization models for this dynamic environment further complicates the associated optimization problem, which falls into the NP-hard class. This research introduces an innovative approximate dynamic programming (ADP) policy for managing the charging of EV fleets at a charging depot equipped with diverse multi-connector chargers. A feature mapping analysis identifies critical system features that shape the future costs of a decision. A comparative analysis illustrates the effectiveness of the proposed policy in terms of cost reduction and service level. Moreover, we observe significant reductions in computation time when updating charging decisions compared to a two-stage rule-based model developed as a benchmark. In addition to benefits for EV fleet operators and CaaS providers, the proposed policy contributes to power grid sustainability by reducing charge load during peak hours, thereby enhancing overall grid stability and efficiency. • We address EV fleet recharge scheduling for a Charging-as-a-Service (CaaS) provider. • We consider a charging depot composed of heterogeneous multi-connector chargers. • We develop an ADP-based approach leveraging statistical learning. • We show the proposed policy significantly reduces charging costs. • We show the approach improves grid sustainability by reducing peak load charging.