Convolutional Neural Network and Low Current Fault Line Selection.

Directional earth fault detection (fault line selection) is still a challenging task because of the low current level during the single phase-to-ground fault in neutral point non-solidly grounded (low current grounded) systems, as a result to that the reliability of detection devices are yet to be improved. In the meantime, low current grounded systems are still widely used because of its good capability of sustaining the power supply during ground fault. Schneider-Electric has been working on this topic for years and has wide range of products its protective relay and feeder automation offers. In this paper we have presented a method to enhance the traditional detection logic with deep learning technique to improve the reliability of the detection. According to the existing tests the performance of the new method is significantly improved.