Large-scale Optimisation for Charging Scheduling of a Commercial Electric Vehicle fleet

Abstract

The rapid adoption of Commercial Electric Vehicles (CEVs), such as electric trucks and buses, is driven by advancements in battery technology, supportive government policies, and increasing environmental awareness. This growth imposes substantial demand on power grids, necessitating efficient charging strategies to ensure cost-effective fleet operations, grid stability, and environmental sustainability. This thesis addresses the challenge of scalable charging scheduling for a heterogeneous fleet of CEVs at depots with heterogeneous chargers, where the number of chargers is less than the number of vehicles. The optimisation objectives include minimising electricity cost, minimising battery degradation, and maximising robustness to variations in arrival and departure times. A Mixed Integer Non Linear Programming (MINLP)-based centralised framework was used as a baseline, and a Multi-Agent Systems (MAS)-based hybrid framework was proposed, in which vehicle, charge request, charger, and depot agents coordinate via direct communication and a shared blackboard. The hybrid method generates charging schedules through four sequential steps, combining centralised decision-making by the depot agent with distributed data processing by other agents. Case studies with fleets of 10, 25, 50, and 100 vehicles showed that the hybrid method produces feasible solutions with minimal deviations in objective values, reduces computation time by up to 99\%, and exhibits approximately linear scalability with fleet size across all objectives. Its flexibility also allows experimentation with agent rules and objective function definitions to evaluate their impact on performance. The results demonstrate that the proposed MAS-based hybrid framework enables scalable and adaptable charging scheduling under grid constraints.