由于电力系统拓扑结构复杂多变，基于数据驱动的静态电压稳定评估通常存在模型泛化能力不足的问题。针对该问题，文中提出了一种基于迁移学习的智能静态电压稳定评估方案。基于最大相关最小冗余（Maximal Relevance Minimal Redundancy，MRMR）准则和shapley值构建S-MRMR特征选择框架，对离线生成的数据集进行数据降维；基于梯度提升分段线性回归树（Gradient Boosting With Piecewise Linear Regression Trees，GBDT-PL）算法构建静态电压稳定评估模型，提取电力系统运行特征与静态电压稳定指标间的映射关系；利用迁移学习对GBDT-PL模型进行实时更新，提高模型的泛化能力。在由电力系统仿真软件PSS/E提供的23节点系统和1648节点系统上的仿真结果表明，文中所提方案对电力系统拓扑结构变化具有较强的鲁棒性，能够满足在线电压稳定评估的要求，为数据驱动方法实际应用于静态电压稳定评估提供了有益的参考。
英文摘要:
Due to the complex and changeable topology of power systems, static voltage stability assessment based on data-driven usually has insufficient model generalization ability. In response to this problem, this paper proposes an intelligent static voltage stability assessment scheme based on transfer learning. Firstly, a S-MRMR feature selection framework based on the maximal relevance minimal redundancy (MRMR) criterion and shapley value is constructed to reduce the dimensionality of the database generated offline. Secondly, the static voltage stability assessment model based on the gradient boosting with piecewise linear regression trees (GBDT-PL) is established to extract the mapping relationships between the system operating features and static voltage stability index. Finally, the trained GBDT-PL model is updated by transfer learning to improve the generalization ability. The simulation results on the 23-bus system and the 1648-bus system provided by power system simulation software PSS/E demonstrate that the proposed scheme is robust to the topological changes of the power systems, and can meet the requirements of online voltage stability assessment, which provides a useful reference for the practical application of the data-driven method in the static voltage stability assessment.