摘要
为解决电力数据特征挖掘不充分导致识别精度不高的问题,提出一种基于多域特征分析与选择的电力数据识别方法。首先针对现有电力数据特征提取方法存在的不足,提出一种基于经验模态分解(EMD)与Hilbert变换(EMD-Hilbert)的特征提取方法,并对电力数据的功率特征和V-I轨迹特征进行量化表征;然后基于随机森林与广义序列后向选择搜索策略相结合的特征选择算法(RF-GSBS)得到最优特征子集,并采用RF算法构建电力数据的识别模型;最后通过仿真算例验证所提方法的有效性和准确性。结果表明,该算法可利用不同特征互补性解决单一特征识别精度不高的问题,并通过特征选择进一步提高学习算法的性能。
To solve the problem of low recognition accuracy caused by insufficient power data feature mining,this paper proposed a novel power data identification method based on multi-domain feature analysis and feature selection.Firstly,aiming at the shortcomings of existing power data feature extraction methods,a feature extraction method based on empirical mode decomposition(EMD)and Hilbert transform(EMD-Hilbert)was proposed,and the power features and V-I trajectory features of power data were quantified.Secondly,based on random forest and generalized sequence backward selection search strategy,the optimal feature subset was obtained.The random forest was employed to build a recognition model for the power data.Finally,the experimental results verified the effectiveness and identification accuracy of the proposed method.The results show that the proposed method can utilize the complementarity of different features to overcome the problem of low accuracy by single feature,and further improve the model recognition performance through feature selection.
作者
洪德华
刘翠玲
赵林燕
雷沁怡
王海鑫
HONG De-hua;LIU Cui-ling;ZHAO Lin-yan;LEI Qin-yi;WANG Hai-xin(Information and Communication Branch of State Grid Anhui Electric Power Co.,Ltd.,Hefei 230061,China;School of Electrical Engineering,Shenyang University of Technology,Shenyang 110870,China)
出处
《水电能源科学》
北大核心
2023年第9期211-215,共5页
Water Resources and Power
基金
国网安徽省电力有限公司科技项目(521207220002)。
关键词
电力数据识别
多域特征提取
特征选择
随机森林
序列后向选择
power data identification
multi-domain feature extraction
feature selection
random forest
generalized sequencebackward selection
