摘要
孤岛检测中整定判据及其阈值选取对于检测效果影响显著,然而孤岛检测整定值无固定原理公式。文章综合利用数据挖掘技术方法,从常规、多分辨率奇异熵判据中通过RELIEF算法筛选出最优特征判据,通过ROC评估,从C4.5,CART,SVM中选出最优分类算法;比较不同功率不平衡度条件下孤岛整定并加以试验验证。文章构建PSCAD仿真模型并对样本数据进行MATLAB预处理,最后3类算法的预测分类结果验证了结论。
Eigenvalue criterion and its threshold settings greatly affects accuracy of islanding passive detection, however there is no prescriptive formula or principle to achieve the threshold settings. Ac- cording to the characteristics of islanding detection, data mining technique is employed to select most appropriate features. An estimate approach ROC is applied to select the best classification algorithm from algorithms such as C4.5、ART、SVM; Further, islanding passive detection under different power imbalances are compared. PSCAD model is established and sample data is preprocessed in MATLAB. Classification results of three algorithms above verify the proposal.
出处
《可再生能源》
CAS
北大核心
2014年第1期39-43,共5页
Renewable Energy Resources
基金
国家高技术研究发展计划"863"课题(2012AA050803)
关键词
数据挖掘
阈值整定
孤岛检测
功率不平衡度
data mining
threshold settings
islanding detection
power imbalance
