风险主动型鲁棒能量与备用调度框架：考虑快速启动资源的离线到在线数据驱动方法

The integration of high-penetration variable renewable energy (VRE) intensifies the demand for flexible reserves in power systems. Conventional robust scheduling methods with endogenous reserves, which rely on prescribed uncertainty sets (US), incur significant operational risks when actual VRE output exceeds prescribed boundaries. To address this, this paper proposes a risk-proactive energy and reserve scheduling framework. First, adopting a data-driven approach, a conservative sparse neural network (CSNN) is employed to learn the mapping relationship between US boundaries and operational risks, with the trained model linearized via Big-M formulation. Subsequently, a risk-driven adjustable US is constructed. Building on this, a risk-proactive robust scheduling framework integrating offline training and online mapping is developed, enabling rapid operational risk assessment through the CSNN model, thereby guiding decision-making for adjustable US boundaries and reserve allocation. Furthermore, an energy storage system (ESS) reserve model considering operating mode switching fully exploits quick-start resource flexibility. To address the binary variables introduced by the ESS operation, an improved C&CG algorithm is proposed to solve the two-stage robust optimization model with a mixed-integer linear programming problem. Case study demonstrates that: i) The data-driven CSNN model can effectively quantify operational risks, achieving a balanced trade-off between operational costs and risks through co-optimization of adjustable US; ii) The linearized reserve model exploiting the regulatory potential of ESS, enhancing the accommodation capability of VRE.