Multi-agent temporal communication framework for scheduling in hydrogen-integrated energy systems

EV charging scheduling in urban multi-energy systems presents significant challenges due to complex temporal dependencies, dynamic multi-agent interactions, and strict physical constraints. To address these issues, this paper proposes a communication-centric multi-agent reinforcement learning framework that enables intelligent coordination under diverse and uncertain conditions. At its core, the framework implements a graph-attention and mutual-information-based communication mechanism, allowing agents to dynamically identify and exchange critical information, thereby enhancing system-wide situational awareness and coordination robustness. Building upon this communication foundation, the framework integrates multiple time-series modeling approaches to capture hierarchical temporal patterns and adapt to varying operational scenarios. This unified design supports long-term decision-making in high-dimensional, multi-agent environments with nonlinear dynamics and coupling constraints. Extensive experiments across different system scales, temporal horizons, and operational contexts demonstrate that the proposed method consistently outperforms baseline approaches in scheduling performance, resource utilization, and generalization capability. Compared with traditional optimization algorithms, our method achieves a 45.1% improvement in overall scheduling efficiency. These results highlight the importance of efficient communication and flexible temporal modeling in advancing multi-agent optimization for low-carbon urban energy systems.