局部放电稀疏分解模式识别方法

A Partial Discharge Pattern Recognition Method Based on Sparse Decomposition

为实现电气设备局部放电(简称局放)模式的准确识别,提出了一种局放稀疏分解模式识别方法。首先由各放电模式局放训练样本信号统计特征向量构建局放统计特征过完备原子库,对此原子库进行非线性映射,可得非线性局放统计特征过完备原子库。对待识别局放信号统计特征向量进行非线性变换,得到非线性统计特征向量,此向量在非线性局放统计特征过完备原子库中进行稀疏分解时,仅可由相应放电模式子原子库中原子进行稀疏表示而难以由其它放电模式子原子库中原子进行表示,进而实现局部放电稀疏分解模式识别。同时,提出一种核函数优化匹配追踪算法,可在无需知道非线性映射具体形式基础上完成稀疏分解,并基于相似性度量系数确定最佳核函数及其参数。设计了两套放电模型,并在不同实验环境中进行了局放测试,所测信号分别作为训练样本信号及测试样本信号,采用所提方法进行了模式识别实验,同时与采用神经网络方法、K近邻法、支持向量机法的局放模式识别实验结果进行了对比。实验结果表明该方法识别效果较好,准确率较高。

To achieve an accurate partial discharge（PD） pattern recognition result of electrical equipment,a PD pattern recognition method based on sparse decomposition was presented.The PD statistical overcomplete dictionary can be built by statistical vectors extracted from each pattern PD training sample signals.Nonlinear mapping was conducted to this dictionary,and the nonlinear PD statistical overcomplete dictionary can be obtained.Decomposing the nonlinear PD statistical vector extracted from the PD signal to be recognized,this vector can only be sparse represented by the atoms extracted from the corresponding pattern sub-dictionary while can＇t be represented by atoms from other sub-dictionaries.Thus,the goal of PD pattern recognition can be achieved.Besides,a kernel improved matching pursuit algorithm was raised to obtain sparse decomposition result without accurate form of the nonlinear mapping.The kernel function and its parameters can be determined based on similarity measuring coefficient.PD signals were measured in two different experimental environments by different artificial discharge models,which were used as training samples and testing samples respectively.The recognition results are obtained by the presented method.The recognition results based on neural network,K nearest neighbor（KNN） and support vector machine（SVM） were obtained for comparison.The experimental results show that the presented method has a higher recognition rate.