Efficient adjust of a learning based fault locator for power distribution systems

The fault location method proposed in this paper uses a classification technique as the support vector machines (SVM), and an intelligent search based on variable neighborhood techniques to select the configuration parameters of the SVM. As result, a strategy is proposed to relate a set of descriptor obtained from single end measurements of voltage and current (input) to the faulted zone (output), in a classical classification task. The proposed approach is tested in selection of the best calibration parameters of a SVM based fault locator and the best error in classification of 3.7% is then obtained considering all of the fault types. These results show the adequate performance of the proposed methodology applied in real power systems.