摘要
提出时频分析结合二维非负矩阵分解的混合特征提取算法识别不同局部放电类型。在实验室环境下采集了4种典型绝缘缺陷模型的局部放电超高频(UHF)波形,引入自适应最优径向高斯核时频分析挖掘局部放电UHF信号的时频信息,在对时频幅值矩阵进行二维非负矩阵分解提取降维特征后,采用模糊k-近邻分类器对4种不同类型的局部放电信号进行识别。对试验样本的识别结果表明:自适应最优径向高斯核时频分布能较好地表征局部放电单次波形的时频信息;二维非负矩阵分解降维后的特征矩阵能保存原始时频矩阵的大部分有用信息;模糊k-近邻分类器比k-近邻分类器和3层反向传播神经网络具有更高的识别率,并较反向传播神经网络具有容易拓展的优点。
A hybrid feature extraction algorithm based on TFA（Time-Frequency Analysis） combined with 2DNMF（Two-Dimensional Non-negative Matrix Factorization） is proposed to identify the defect types of PD （Partial Discharge）. The UHF（Ultra High Frequency） signals of various defect models are measured in laboratory and AORGK（Adaptive Optimal Radially Gaussian Kernel） time-frequency analysis is then introduced to represent the UHF signals. The obtained time-frequency amplitude matrices are further compressed by 2DNMF and FkNN（Fuzzy k-Nearest Neighbor） classifier is then applied to recognize the four typical PD defects. Experimental results show that,TFA describes the time-frequency information of PD UHF signals effectively;the extracted features reserves most useful information of original time-frequency matrices;FkNN classifier has a higher recognition rate than those of kNN（k-Nearest Neighbor） classifier and BPNN（Back Propagation Neural Network）,and is easier to expand than BPNN.
出处
《电力自动化设备》
EI
CSCD
北大核心
2013年第3期20-25,共6页
Electric Power Automation Equipment
基金
国家自然科学基金资助项目(51277187)
国家重点基础研究发展计划(973计划)项目(2009CB724505-1)~~
关键词
绝缘
局部放电
模式识别
时频分析
二维非负矩阵分解
模糊k-近邻分类器
insulation
partial discharge
pattern recognition
time-frequency analysis
two-dimensionalnon-negative matrix factorization
fuzzy k-nearest neighbor classifier
