Evaluating a k-nearest neighbours-based classifier for locating faulty areas in power systems

This paper reports a strategy for identifying and locating faults in a power distribution system. The strategy was based on the K-nearest neighbours technique. This technique simply helps to estimate a distance from the features used for describing a particular fault being classified to the faults presented during the training stage. If new data is presented to the proposed fault locator, it is classified according to the nearest example recovered. A characterisation of the voltage and current measurements obtained at one single line end is also presented in this document for assigning the area in the case of a fault in a power system. The proposed strategy was tested in a real power distribution system, average 93% confidence indexes being obtained which gives a good indicator of the proposal's high performance. The results showed how a fault could be located by using features obtained from voltage and current, improving utility response and thereby improving system continuity indexes in power distribution systems.