基于累积贡献率和可解释人工智能的静态电压稳定裕度估计特征量筛选方法

利用海量量测数据估计大规模互联电网静态电压稳定裕度时，合理地选择输入量测信号和裕度估计算法是实现高质量裕度估计的基础。提出了一种基于累积贡献率和可解释人工智能的关键特征量筛选方法。给出了基于沙普利值加性解释理论可解释模型的输入特征贡献值量化方法，并依据贡献值大小对特征降序排列；采用基于累积贡献率增量的循环优化过程剔除冗余特征，形成关键特征子集；在系统关键特征优选的基础上，采用轻量梯度提升机算法实现静态电压稳定裕度在线估计。所提方法在保证估计精度的同时，大幅降低初始样本维度，解决特征过拟合问题，有效提升静态电压稳定裕度估计在线性能。基于WECC 3机9节点系统、IEEE 10机39节点系统以及IEEE 300节点系统的仿真分析验证了所提关键特征量筛选方法在电力系统静态电压稳定裕度估计中的有效性。

Feature selection approach based on FCC-eAI in static voltage stability margin estimation

Reasonable selection of input measurement signal and margin estimation algorithm is the basis for realizing high-quality margin estimation, when estimating the static voltage stability margin for large-scale interconnected power grid using massive measurement data. Thus, a critical feature selection approach based on feature cumulative contribution rate-explainable artificial intelligence(FCC-eAI) is proposed. The quantification method of input feature contribution value based on the interpretable model of Shapley additive explanation theory is proposed, and the features are rearranged in descending order according to the contribution value. Then, the redundant features are eliminated through the circular optimization process based on the cumulative contribution rate increment, which constitutes the critical feature set. Along with the optimal selection of critical features for the system, the online estimation of static voltage stability margin is developed based on light gradient Boosting machine algorithm. The proposed method significantly reduces the initial sample dimension, solves the feature overfitting problem and effectively improves the online performance of static voltage stability margin estimation while ensuring the estimation accuracy. The effectiveness of the proposed critical feature selection approach used for the estimation of the static voltage stability margin in power system is verified through the simulation analysis based on WECC 3-machine 9-bus system, IEEE 10-machine 39-bus system and IEEE 300-bus system.