基于可拓云和集成学习的电力现货市场运营风险评价与预警

[Objective] The construction of a unified national electricity market system has been gaining momentum. However，the complexity of market mechanisms and lack of real-time capabilities of existing methods make dynamic risk warnings in spot market operations challenging. [Methods] To address this issue，an operational risk assessment model based on pre-clearing feature variables is proposed，utilizing a stacking ensemble learning model for prediction. First，key feature variables with significant market impacts are selected，and an improved Latin hypercube sampling method is used to generate numerous risk scenarios. Second，an operational risk evaluation index system for the spot market is established，and the risk indicators are calculated after the spot market is cleared. Subsequently，a cloud-entropy-optimized matter-element extension cloud model is employed to balance the fuzziness and uncertainty in the transformation between the quantitative and qualitative evaluations，enabling reasonable risk grading. Finally，random forest，deep neural network，and XGBoost models are used as base learners for a five-fold cross-validation to extract the market features. Logistic regression is then employed as the meta-learner to provide early warnings of overall market risks and specific indicator risks. [Results] Simulation results based on the IEEE 24-bus system indicated that the proposed model effectively alleviated the underfitting issue for minority classes. It achieved an accuracy of 79. 36% in overall risk prediction and a Top-2 accuracy of 97. 01%，representing a 4%~ 12% improvement over traditional models. The average accuracy of single-indicator risk warnings was 78. 95%. [Conclusions] The proposed model facilitates reasonable evaluation and accurate early warning of operational risks in electricity spot markets. It effectively identifies risk sources and provides reliable support for real-time risk management and decision making.