High impedance fault detection based on linear prediction

In many situations, conductor rupture in distribution systems or even its contact with structures external to the systems (such as trees) does not sensitize protection systems. This type of occurrence and its variations are typically called high impedance faults. The present work proposes a method for detection of high impedance faults based on linear prediction. Concerning validation of the method, a database was created through simulations performed in the Alternative Transients Program, based on a real energy distribution system. In addition to high impedance faults, load energization and capacitor bank switching situations were also simulated in order to test the robustness of the method against probable false positives. From obtained results, a decision criterion was developed aiming at the detection of high impedance faults based on the linear predictor of current signals through time, which obtained success rates above 80 % for real data.