摘要
特征的选择和排序是分析故障数据的重要环节,高效的特征提取和排序算法可以减少在数据分析中的特征数量,可以从数据中提取更有意义的设备状态信息,减少信息冗余。由于在特征空间中不同区域的特征相对重要性不同,该文提出一种基于上下文感知特征映射的局部线性特征排序框架。实验证明,该算法可以在外界条件变化的情况下识别设备的重要特征,甚至能使特征重要性排序和数据分类变得更简单。
The selection and sorting of features is an important part of analyzing failure data. Efficient feature extraction and sorting algorithm can reduce the number of features in data analysis. We can extract more meaningful device status information from the data and reduce the information redundancy. In this paper, we propose a local linear feature sorting framework based on context-aware feature mapping. The experimental results show that the proposed algorithm can identify important features of the device under varying external conditions and even make it easier to sort the feature importance and classify the data.
作者
黄佳林
茅大钧
倪新宇
HUANG Jia-lin;MAO Da-jun;NI Xin-yu(Automation Engineering Institute,Shanghai University of Electric Power,Shanghai 200090,China;China Huadian Corporation Jiangsu Wangting Power Generation Branch,Suzhou 215155,China)
出处
《电力科学与技术学报》
CAS
北大核心
2019年第4期48-53,共6页
Journal of Electric Power Science And Technology
基金
国家青年自然科学基金(61503237)
关键词
数据挖掘
上下文感知
特征排序
数据分类
data mining
context aware
feature sorting
data classification