基于在线核学习的电网不良数据检测与辨识方法  

Power grid bad-data detection and identification based on online kernel learning method

作　　者：王兴志[1] 严正[1] 沈沉[2] 李乃湖 景雷 李慧杰 

机构地区：[1]电力传输与功率变换控制教育部重点实验室(上海交通大学电气工程系),上海200240 [2]电力系统国家重点实验室(清华大学电机系),北京100084 [3]阿尔斯通电网中国技术中心,上海201114

出　　处：《电力系统保护与控制》2012年第1期50-55,共6页Power System Protection and Control

基　　金：国家自然科学基金资助项目(90612018)~~

摘　　要：遥测数据和电网参数的准确性是智能电网技术支持系统各种电网分析计算的基础。考虑多时段RTU/PMU信息和电网复功率平衡方程,提出了基于在线核学习的不良数据检测与辨识方法。基于复功率和电压幅值信息来定义复功率平衡性评价指标;引入基于贝叶斯数据处理策略的扩展卡尔曼滤波算法进行不良数据动态检测;采用局部加权投影回归策略对电网参数进行在线辨识。不同规模电网的计算结果表明此算法可行,并具有较高的计算精度,为智能电网技术支持系统高级应用软件的研究提供了一种可行的方法。The accuracy of remote measurements and network parameters is the basis for a variety of power system analysis in the smart grid technology support system. Considering multi-period RTU/PMU information and complex power balance equations, an online kernel learning-based bad-data detection and parameter identification method is proposed. First, the information of complex power and node voltage amplitude is used to define the indices of complex power balance. Second, the Bayesian data analysis-based extended Kalman filter method is described to detect the bad-data dynamically. Third, locally weighted projection regression method is adopted to identify the network parameters. The calculation results of different scale power system show that this algorithm is feasible and has a high computing accuracy. It provides a viable approach for the advanced application research of smart grid technology support system.

关 键 词：电力系统 不良数据 状态辨识 卡尔曼滤波 在线核学习 广域测量系统 

分 类 号：TM76[电气工程—电力系统及自动化]