On the application of a machine learning technique to fault diagnosis of power distribution lines

This paper presents one method for fault diagnosis of power distribution lines by using a decision tree. The conventional method, using a decision tree, applies only to discrete attribute values. To apply it to fault diagnosis of power distribution lines in practice, it must be revised in order to treat attributes whose values range over certain widths. This is because the sensor value or attribute value varies owing to the resistance of the fault point or is influenced by noise. The proposed method is useful when the attribute value has such a property, and it takes into consideration the cost of acquiring the information and the probability of the occurrence of a fault