Performance analysis of a learning structured fault locator for distribution systems in the case of polluted inputs

Abstract

Waveform distortion and the service continuity are the two main aspects which define the power quality; in the last, the fault location plays a fundamental role. Considering the requirement of reliable tools to improve the distribution system operation, this document analyses the confidence of a fault locator approach. This paper initially presents the adjusting and the validation methodology used to develop a fault locator, and also its performance is analysed by considering different distorted measurements as inputs. The approach here presented is a learning based fault locator (LBFL), specially customized for distribution systems, whose core are the Support Vector Machines (SVM). The LBFL is tested in an unbalanced 34.5kV distribution feeder considering diverse operating conditions and fault scenarios, which include several fault resistances, locations, and common disturbances on the voltage and current input signals. From the tests, the lowest LBFL performance is 96.5% in such cases of not distorted measurements of voltage and current. In the case of considering polluted inputs, the worse performance is obtained in the case of distortions caused by low sampling frequency, current transformer saturation and high harmonic distortion.