Fault identifiability and pseudo-data-driven fault localization in a DC microgrid

Post-fault maintenance and power restoration in low voltage direct current (LVDC) microgrids are highly dependent on the fault localization criteria. This paper investigates the localization of an LVDC fault without communiations. In this paper, state-space modelling is firstly employed to investigate the identifiability of a DC fault in a linerized DC (LVDC) network. We show that a DC fault in is not identifiable in and unknown multi-bus DC network with local measurements, i.e. when they are outnumbered by the total states. In line with such theory, the localization of DC fault is proposed to be embedded in reclosing process to reduce the number of states during identification. And then, a pseudo-data-driven method is proposed to localize an LVDC fault. Combining an enhanced analytical approach and model-based artificial neural network, the proposed method can broadly localize the position of both underdamped and over- damped DC faults without communications. The robustness against higher fault level, low sampling rate, full-range fault position, sampling noises and source variations have been validated using time-domain simulations with Matlab/Simulink.