Intelligent Energy Management Using Multi - Agent Dynamic Learning for Scheduling Commercial Electric Vehicle Charging Stations

For commercial electric vehicles (CEVs), an underexplored challenge is the complexity of demand and supply management, which is vital for the efficient operation and broader adoption of CEVs. By leveraging advanced smart grid technologies and intelligent energy management systems, the research endeavors to create a cost-effective software solution for optimizing the charging process. This study deploys proximal policy optimization (PPO) multi-agent deep reinforcement learning (MARL) within an actor-critic network architecture. Agents are responsible for managing the supply and demand of energy from two grids welcoming ten charging stations each pumping energy from the integrated uninterruptible power supply (UPS). Performance metrics are compared against a dynamic program-ming (DP) approach, serving as a benchmark. The DP model excels when prior information is readily available. In contrast, PPO agents exhibit remarkable robustness and adaptability in environments lacking such information obtaining 95 % accuracy. These insights not only enrich the existing academic discourse but also establish new performance benchmarks for practical implementations.