基于Bagging-异质k近邻的输电电缆故障诊断方法

Fault Diagnosis Method for Transmission Cables Based on Bagging-hexogeneous k-nearest Neighbor

针对110 kV交叉互联电缆输电线路故障分类不全、分类准确率低等问题。提出了一种基于Bagging-异质k近邻提升学习的交叉互联电缆故障分类方法,首先通过对各类故障得到的主芯及护层电流、电压等相关电气参数归一化处理,构建电气参数特征矩阵;然后基于k近邻(k-NN)算法采用不同k值及不同距离度量作为个体学习器并构建高差子学习器,通过引入Bagging算法提高异质学习器的整体学习效率,可以实现针对不同类型、不同区域交叉互联输电电缆故障的有效区分。该方法相比于传统SVM、k-近邻及逻辑回归等分类方法,降低了多分类误差率及空间复杂度并提高了模型泛化能力,具有较大的工程运用潜力。

In view of such problem as incomplete fault classification and low classification rate of 110 kV cross-connected cable transmission line,a kind of fault classification method of cross-connected cable based on Bagging-heterogeneous k-nearest neighbor(k-NN)lifting learning is proposed.Firstly,such related electrical parmeers as current and voltage of the main core and sheath,which are obtained by various faults,are normallized so to construct the characteristic matrix of electrical parameters.Then,based on the k-nearest neighbor algorithm,different k values and different distance measures are used as individual learners and the height difference sub-learners are constructed.Introduction of Bagging algorithm to improve the overall learning efficiency of the heterogeneous learners can achieve effective distinguish of the faults of cross-connected transmission cables in different types and regions.Compared with traditional SVM,the classification method of k-nearest neighbor and logical regression reduce the multiple classification error rate and space complexity as well as improve model generalization ability,which has great potential in engineering application.