Research on Power-Computing Coordinated Scheduling Based on Dual Demand Coupling and Multi-Agent Learning

With the large-scale integration of renewable energy and rapid development of intelligent computing centers, traditional power system scheduling methods struggle to address the challenges in coordinated operation of wind-solar-storage-transmission systems and intelligent computing centers. To tackle this issue, this paper proposes an intelligent cooperative scheduling theoretical framework based on dual demand coupling and multi-agent reinforcement learning, realizing resource coordination optimization between power systems and intelligent computing centers through a bidirectional coupling model of power demand and computing power demand. First, a power-computing dual demand coupling mechanism is designed, establishing a multi-time scale coordinated scheduling model considering spatiotemporal uncertainties of wind and solar power. Second, a deep reinforcement learning framework based on multi-agent architecture is proposed, achieving distributed learning and collaborative decision-making. Third, a preference-based multi-objective Pareto optimization method is developed, solving the challenge of adaptive adjustment of multi-objective weights. Finally, a temporal optimization decision model considering energy storage lifecycle is constructed, balancing the system's short-term economic benefits and long-term returns. Experimental results demonstrate that, compared with traditional methods, the proposed algorithm shows significant advantages in comprehensive operation costs, renewable energy consumption rates, and system stability, especially in complex scenarios with both wind-solar resource fluctuations and computing power demand changes, effectively maintaining stable system operation and significantly enhancing system adaptability and robustness.