A LSTM-based approach for detection of high impedance faults in hybrid microgrid with immunity against weather intermittency and N-1 contingency

Abstract

The low magnitude of fault current coupled with the non-linear dynamics makes detecting high impedance faults (HIF) challenging in hybrid microgrids. In addition to reduced fault current, the stochastic variation in the operation of the PV-based distributed energy resources (DERs) due to weather intermittency and network reconfiguration arising due to N-1 contingency further hinders achieving high reliability and accuracy in HIF detection. For increased sensitivity against HIF, a long short-term memory (LSTM) based protection scheme is proposed with improved immunity to weather intermittency and adaptiveness to N-1 contingencies. Using metrological data, a statistical model of solar irradiance is derived to examine the impact of weather variation on the voltage and current dynamics post fault. At the same time, the inclusion of contingency data in the learning of the LSTM model ensures adaptability to changes in the network topology. A limited set of critical sensors provides the information necessary to perform the protection tasks thereby avoiding the complexity of the communication network, cost associated with sensor installation and computational cost. The validation of the proposed hybrid microgrid protection scheme under diverse operating scenarios reflects its effectiveness in providing reliable and accurate protection with robustness against N-1 single-line contingencies and weather variation.