Intelligent Optimization of Computing-Power Collaborative Systems with High-Penetration Renewable Energy

This paper proposes a multi-timescale intelligent optimization operation framework for the co-optimization of computing and power systems. First, a “second-to-minute” spatiotemporal dynamic coupling model based on task elastic mapping and renewable energy fluctuations is constructed, revealing the bidirectional feedback mechanism between computing and power systems. Second, a multi-timescale operational optimization strategy based on the Compressed Sensing-based Hierarchical Nearest-Neighbor Policy Optimization (CS-DLPPO) algorithm is designed to resolve the coordination challenge between power frequency regulation/peak shaving and computing resource allocation/pooling. Subsequently, a multi-objective federated learning optimization method integrating spatio-temporal attention mechanisms with Weighted Delayed Deep Deterministic Policy Gradient (WDDDPG) is proposed to achieve cross-regional resource coordination. Finally, the proposed method's effectiveness in enhancing system stability, economic efficiency, and lowcarbon performance was validated through a hardware/software co-simulation platform.