油入ケーブル接続部の線形サポートカーネルマシンによる異常判定

抄録

Japanese electric power companies widely use the gas-in-oil based diagnostic criterion, which was developed in 1999, for determining the anomaly ranking of intermediate joint boxes of single-core cables during the maintenance of oil-filled cable joint boxes. However, it can determine neither the fault location nor aid in the diagnosis of terminal joint boxes. In addition, several joint boxes that are determined as normal using this criterion have recently been found to be anomalous in the overhaul. In this paper, we propose a new relatively accurate diagnostic criterion that covers both the intermediate and terminal joint boxes and aids in determining the anomaly ranking and fault location using the multiclass ν-linear support kernel machine (SKM), which we propose as an extension of the linear support vector machine (SVM). The proposed multiclass ν-linear SKM automatically scales data to maximize the performance of the linear SVM and obtains simpler linear evaluation functions. Furthermore, it is formulated as a linear programming problem, whereas general SKMs are formulated as semi-definite programming problems that are difficult to solve. The accuracy of our proposed linear criterion, which was estimated using 5 fold cross-validation, was approximately 75% which was almost comparable to 76% by the one-against-one non-linear RBF-support vector machine.