Recognition of Fault Location and Type in a Medium Voltage System with Distributed Generation using Machine Learning Approach

This work describes a preliminary research investigation to access the feasibility of using advanced machine learning techniques for predicting and diagnosing fault type and fault location in a power distribution network consisting distributed generation. The proposed approach uses three phase voltage and current measurements data, assumed to be available at all the source bus. To understand the potential of the machine learning methodology, practical scenarios in a distribution grid such as all types of faults i.e. SLG, LLG, LL, and LLL with different fault locations are addressed in this work. Initially, the fault data is generated which is used to train a fault locator module. Further same data is used to design a fault type detector model in offline mode. The online real time data when fed to these models are able to give exact location and type of fault. The results are obtained from seven techniques of machine learning and their comparison is also done. The approach is proved to be a feasible tool for fault analysis.