摘要
半监督学习可借助有标签和部分无标签样本数据来构建电网暂态稳定评估模型,有效利用输入样本数据,可提高电网暂态稳定评估准确率,为此提出基于半监督近似流形支持向量机(manifold proximal support vector machine,MPSVM)的暂态稳定评估方法。首先,在MPSVM的正则项中引入判别变量,可最大限度捕捉样本数据内部的几何信息,并通过最大距离理论表征电力系统稳定类和不稳定类之间的差异,进而转化为求解特征值问题;然后,采用贝叶斯非线性分层模型确定最优参数,可进一步提高评估准确率;最后,采用IEEE 39标准系统和鞍山电网的仿真分析验证所提评估模型的有效性和准确性。
Semi-supervised learning constructs a transient stability assessment model with tabbed and partially unlabeled sample data.It is able to improve evaluation accuracy for power grid transient stability by effectively using the input sample data.Thus,this paper proposes a transient stability evaluation method using semi-supervised approximate manifold support vector machine(MPSVM).Firstly,it introduces the discriminant variable into the manifold regularization of the MPSVM,which can capture the geometric information inside the sample data to the maximum extent,and characterizes the difference between the stability class and the unstable class of the power system through the maximum distance,and then converts into a pair of eigenvalue problems.Afterwards,it uses the Bayesian nonlinear layered model to determine the optimal parameters so as to further improve the evaluation accuracy.Finally,the validity and accuracy of the evaluation model are verified by IEEE 39 node system and Anshan actual system.
作者
曲锐
王世荣
辛文龙
QU Rui;WANG Shirong;XIN Wenlong(Foshan Power Supply Bureau of Guangdong Power Grid Co.,Ltd.,Foshan,Guangdong 528000,China;School of Electrical and Electronic Engineering,Changchun University of Technology,Changchun,Jilin 130000,China)
出处
《广东电力》
2020年第4期74-81,共8页
Guangdong Electric Power
关键词
近似流形支持向量机
半监督分类
暂态稳定评估
贝叶斯非线性分层模型
机器学习
manifold proximal support vector machine(MPSVM)
semi-supervised classification
transient stability assessment
Bayesian nonlinear layered model
machine learning
