Probabilistic transmission line fault diagnosis using autonomous neural models

Abstract

Fault diagnosis is a critical task when a severe disturbance is caused by insulation failure on a transmission line. Fault diagnosis should be concluded as soon as possible to avoid further financial and social costs because of load interruptions. Intelligent systems have been successful in dealing with fault diagnosis. This paper proposes the application of autonomous neural networks for mapping the relationship between electrical signals at one terminal and fault information on transmission lines. The proposed autonomous neural networks allow automatic adaptation of the neural models responsible for fault detection, classification, and location. The proposed autonomous neural models can provide the uncertainty associated with the inference process. If a fault is likely, the detection model provides the inference probability of a false positive. Probabilities for each class are given in fault classification, while fault location provides an error margin around the estimated short-circuit position. Another contribution of the paper is the extraction of useful information from oscillography to feed the neural models. An efficient voltage and current representation is proposed. The tests consider realistic fault conditions for main transmission lines in the Brazilian power network. The methodology has proven to be robust also for dealing with multi-terminal and series-compensated lines.