电力系统低频振荡模式的自动分类研究  

Research on power system low frequency oscillation modes classification

作　　者：陆超[1] 陆秋瑜[1] 

机构地区：[1]清华大学电机系电力系统国家重点实验室,北京100084

出　　处：《电力系统保护与控制》2010年第4期35-38,48,共5页Power System Protection and Control

基　　金：电力系统国家重点实验项目(SKLD08Z01);中国南方电网有限责任公司重大科技专项

摘　　要：大规模复杂电网在不同运行条件下其低频振荡模式变化明显,基于扰动信号和类噪声信号的振荡辨识海量结果需在线分析,以完成多个模式的自动识别与分类。针对上述问题,提出并设计了一套不依赖于经验的振荡模式自动分类系统。该系统包括特征选择和分类器两个部分,利用特征选择实现了大范围的降维,并对比分析了线性的Fisher分类器、非线性的二次型和k近邻分类器性能。进一步的,基于南方电网简化仿真数据进行了验证,其结果说明了所设计和实现方法的有效性,为进一步的预警、分群等提供了重要信息。In large scale power grids, the low frequency oscillations may change significantly in different operation conditions, and the mass results identified based on ringdown and ambient signals must be analyzed online to complete the automatic classification of multiple oscillation modes. In this paper, a computer-oriented oscillation modes recognition and classification system is designed, which includes feature selection and classifier. The first one is used to select the most sensitive features and reduce the dimensions greatly. About classifier, three linear and nonlinear methods, i.e. Fisher, k-neighborhood and quadratic, are compared. The classification performances are verified based on the China Southern Power Grid simplified data, and the results demonstrate the feasibility and advantages of the proposed approach.

关 键 词：低频振荡 振荡模式辨识 模式分类 特征选择 Fisher分类 K近邻分类 二次型分类 

分 类 号：TM712[电气工程—电力系统及自动化]