Learning-driven multi-timescale operation simulation and hierarchical boundary optimization for renewable-dominated energy systems under temporal and scenario uncertainties

This paper addresses the challenge of coordinating power system operation across multiple temporal horizons under high renewable penetration and scenario uncertainty, where conventional decoupled scheduling frameworks fail to maintain operational consistency and adaptability. To overcome this limitation, we propose a learning-driven multi-timescale operation simulation and hierarchical boundary optimization framework that integrates deep sequence forecasting, reinforcement learning, and rolling model predictive control within a unified optimization structure. The framework dynamically constructs and updates operational boundaries across long-term, short-term, and rolling horizons, enabling continuous coordination between planning and real-time control. Case studies on a renewable-dominated IEEE 118-bus system demonstrate that the proposed approach significantly improves system performance. Compared with conventional decoupled scheduling, the framework reduces total operating cost by approximately 9.6%, increases system reliability by 15.8%, and lowers renewable curtailment from 9.5% to 5.7%. In addition, boundary deviations during rolling operation are constrained within a narrow band below 2%, confirming the effectiveness of the adaptive boundary-learning mechanism. These results indicate that the proposed framework not only enhances economic efficiency and operational robustness, but also establishes a new paradigm for integrating learning and optimization in multi-timescale energy system management.