Real-Time Implementation and Evaluation of a Support Vector Machine Based Fault Detector and Classifier for Distribution Grids

Abstract

Bridging the gap between theoretical modeling and practical implementation is essential in fault detection, classification, and location methods for modern distribution grids. Currently distribution grids are characterized by dispersed infeed and distributed power generation that render conventional protection settings more challenging, and, hence, new methods must be investigated to resolve these issues. In this paper, a novel framework capable of detecting and classifying faults in power distribution grids is presented. The proposed algorithm formulates a unique fault classification technique based on measurement samples of three-phase voltage and currents after the occurrence of a fault event in power distribution grid. Thereby, negative sequence components of three-phase voltage and current quantities are used for online fault detection that triggers a fault classification method based on a support vector machine. In order to simulate fault scenarios, a model of a reference distribution grid incorporating distributed generation is developed, simulated, and analyzed under fault conditions. For the final evaluation, the designed protection scheme is implemented on a Programmable Logic Controller connected to the model running on a real-time simulator hardware that provides voltage and current measurements. Thus experimental results are provided to demonstrate and prove the contribution of the algorithm in its ability to correctly identify and classify faults in modern distribution grids.