Fault identifiability and pseudo-data-driven fault localization in a DC microgrid. (June 2023)

• Record Type: Journal Article
• Authors: Javed, Waqas; Chen, Dong; Kucukdemiral, Ibrahim
• Is Part Of: International journal of electrical power & energy systems. Volume 148(2023)
• Publication Date: 2023-06
• Subjects: Low-voltage DC microgrid -- Fault-localization -- LVDC protection -- Data-driven -- Artificial neural networks; Electrical engineering -- Periodicals; Electric power systems -- Periodicals; Électrotechnique -- Périodiques; Réseaux électriques (Énergie) -- Périodiques; Electric power systems; Electrical engineering; Periodicals; 621.3
• DOI: 10.1016/j.ijepes.2023.108944
• Languages: English
• ISSNs: 0142-0615

Abstract

Highlights: Mathematical proof of fault identifiability in an unknown LVDC network. New philosophy of LVDC fault localization with local measurement. Hybrid method combing analytical and pseudo-data-driven approach in estimating the distance between a DC fault and a measurement unit (device with relay function) Online rejection of non-linear inputs to preserve system autonomy for localization and fault estimation. Improved accuracy with lower sampling rate. High robustness against sampling noises by eliminating the need for differential calculations. Abstract: 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.