New Control Framework for EV Fleets Cuts Charging Costs and Battery Wear

A new research paper from Mainak Dan and Arvind Easwaran introduces a hierarchical stochastic model predictive control (MPC) framework that integrates charging and service allocation for mobility-on-demand electric vehicles (MoD-EVs) under a flexible A-to-B rental model. The approach tackles the dual challenge of optimizing EV charging against fluctuating electricity prices while meeting stochastic customer demand — all while preserving battery health. By embedding a piece-wise linear battery model into a multi-objective, chance-constrained mixed-integer linear programming (MILP) formulation, the framework captures nonlinear battery dynamics without overwhelming computational resources. A mixed logical switching mechanism determines optimal charging sequences, and the system is designed to scale via a distributed architecture rather than centralized optimization.

Testing against a state-of-the-art commercial solver with stochastic rental requests and time-varying electricity prices shows clear advantages over two baselines: a business-as-usual (BAU) approach and the Laxity-based charging (LC) method. The integrated control mechanism reduces both charging costs and battery capacity degradation, with the benefits becoming more pronounced under higher uncertainty in demand and pricing. This work directly addresses the operational inefficiencies facing electric ride-hailing fleets, where uncoordinated charging can lead to missed rental opportunities or accelerated battery wear. The distributed approach also makes the solution practical for real-time deployment across large fleets, offering a path to more sustainable and cost-effective EV fleet management.