Collaborative Optimization Scheduling Strategy of Electric Vehicle Load Cluster and Data Center for Multi-Station Integration

To harness the collaborative advantages of integrated multi-station smart energy hubs and mitigate their carbon footprint, this study investigates dense urban travel patterns and addresses privacy concerns inherent in traditional centralized grid dispatch systems. We propose a distributed dispatch optimization framework for smart energy stations leveraging multi-agent deep reinforcement learning. For the first time, an optimal carbon emissions decision model is introduced, synthesizing human social behavior at edge data centers, electric vehicle charging dynamics, and active distribution network operations. The multi-station carbon optimization challenge is reformulated as a Decentralized Partially Observable Markov Decision Process (DEC-POMDP), resolved through an Actor-Critic-based decentralized execution framework. This approach enables real-time scheduling where stations autonomously make decisions while collaboratively processing raw user data locally to ensure privacy preservation. Furthermore, spatial migration of data center computing tasks and coordinated vehicle charging-discharging strategies effectively address the dual volatility arising from stochastic renewable energy generation and electric vehicle load fluctuations. Experimental results demonstrate that our method fully exploits the synergistic potential between electric vehicles and data centers, significantly enhancing clean energy utilization and reducing carbon emissions across smart energy stations.