基于SLLE的电缆附件局部放电模式识别

Pattern recognition for partial discharge of cable accessories based on SLLE

为更全面地提取局部放电信号的特征值信息,提高识别率,将局部放电统计特征参数和矩特征参数相结合,提取出高维的特征值。从不同的角度出发,结合两种不同的方法对局放特征进行提取。同时在流形学习非监督的基础上引入了监督信息,从而保证高维到低维的映射在保留流形某些结构,同时也可进一步分离不同类别的流形。利用基于监督的局部线性嵌入(Supervised Locally Linear Embedding,SLLE)对局部放电特征值进行降维优化处理,提取出具有较高分类能力的最优特征值,利用电力电缆附件的4种典型缺陷进行实验对比,结果表明文中方法较好地提取出最优特征值,且能得到更准确的识别结果。

In order to extract the feature value information of the partial discharge signal more comprehensively and improve the recognition rate,statistical feature parameters and moment feature parameters are combined to extract high-dimensional eigenvalues from different angles.The advantages of partial discharge feature extraction are combined with two different methods.Meanwhile,based on non-supervised manifold learning,the supervised information is induced to the algorithm to ensure the map from high dimension to low dimension to retain some manifold structures and also to separate different kinds of manifolds.The supervised locally linear embedding(SLLE)is used to reduce the dimension of the eigenvalues,and the optimal eigenvalues with higher classification ability are extracted.Then,the experimental comparison of four typical defects of power cable accessories shows that the proposed method can extract the optimal eigenvalues better and obtain more accurate recognition results.