• Jiang, Xu
• Mcarthur, Stephen
• Stephen, Bruce

Abstract

Increased monitoring of distribution networks and power system assets present utilities with new opportunities to predict and forestall system failures. Although automated pattern recognition methodologies have given other industries significant advantage, power system operators face additional challenges before these can be realized. The effort of apportioning ground truth to fault data creates a knowledge bottleneck that can make utilizing automatic classification techniques impossible. Surrogate approaches using operational process outputs such as maintenance tickets as labels can be challenging owing to the causal ambiguity of these written records. To approach a solution, this paper demonstrates utilizing natural language processing techniques to disambiguate the free text in maintenance tickets for onward use in supervised learning of fault prediction and classification techniques. A demonstration of this approach on an established power quality fault data set is provided for illustration.

Topics

• impedance spectroscopy