孤岛检测的关键特征识别及元学习方法

发布时间：2018-01-10 16:03

本文关键词：孤岛检测的关键特征识别及元学习方法　出处：《电力系统自动化》2014年18期 　论文类型：期刊论文

【摘要】：数据挖掘技术能有效解决孤岛检测中检测阈值的整定问题,已成为重要的孤岛检测方法。文中提出由关键特征识别、基学习器和元学习器等3个环节构成的孤岛检测数据挖掘系统。首先,分析了孤岛检测样本中的弱相关特征对分类的不利影响,提出利用RELIEF(recursive elimination of features)算法首先识别孤岛检测的关键特征。然后,分析了单一分类器的归纳偏置现象,提出利用多个分类器的互补性提高孤岛检测的精度;最后,提出了基于元学习的新的孤岛检测方法。为验证上述方法的有效性,仿真算例中充分考虑了功率不平衡度、电压扰动等因素。仿真结果表明,上述3个环节对提高孤岛检测的精度和泛化能力具有重要作用。
[Abstract]:Data mining technology can effectively solve the problem of setting the detection threshold of islanding detection, has become an important method for islanding detection. This paper proposed by key feature recognition, system islanding detection data constitute the base learner and meta learning for the 3 links of mining. First, analyzed the adverse effects of weak correlation characteristics of islanding detection in samples on the classification, proposed by RELIEF (recursive elimination of features) algorithm first identifies the key features of islanding detection. Then, analyzes the phenomenon of inductive bias is proposed by using single classifier, multiple classifiers complementary improve islanding detection accuracy; finally, put forward a new islanding detection method based on meta learning. The methods are validated by simulation example considering power unbalance, voltage disturbance and other factors. The simulation results show that the above 3 aspects to improve the detection of solitary island Precision and generalization ability play an important role.

【作者单位】：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子信息与电气工程学院;
【基金】：国家高技术研究发展计划(863计划)资助项目(2012AA050803) 上海市科委项目(11dz1210402)~~
【分类号】：TM73
【正文快照】： 0引言将大量分布式电源(distributed generator,DG)以友好的方式接入系统是智能电网的重要特征。无论从系统运行、人员设备安全还是电能质量角度,都要求DG具备孤岛检测功能。现有孤岛检测方法主要包括被动检测法[1]、主动检测法[2-4]、开关状态检测法等3类。开关状态检测法依

【相似文献】