Fast Fault Location Method for a Distribution System with High Penetration of PV

Abstract

Distribution systems with high levels of solar PV may experience notable changes due to external conditions, such as temperature or solar irradiation. Fault detection methods must be developed in order to support these changes of conditions. This paper develops a method for fast detection, location, and classification of faults in a system with a high level of solar PV. The method uses the Continuous Wavelet Transform (CWT) technique to detect the traveling waves produced by fault events. The CWT coefficients of the current waveform at the traveling wave arrival time provide a fingerprint that is characteristic of each fault type and location. Two Convolutional Neural Networks are trained to classify any new fault event. The method relays of several protection devices and doesn't require communication between them. The results show that for multiple fault scenarios and solar PV conditions, high accuracy for both location and type classification can be obtained.

What Makes a Climate Change Denier Popular? Exploring Networked Social Influence in a Disinformation Spreader Group

“…where q : y q = y p denotes those data samples q have the same labels with y p , and r : y r ∈ [1, c] represents any data sample r. Hence, the correct prediction of the unlabeled data sample p has high probability if (1) The paths from the data samples with the same labels y p are more than those with other labels; (2) the number of labeled data samples with y p is significant.…”

“…Another line of work relies on the physical property of data, such as the spatial relations of line impedance and the sparsity of fault currents, but these methods either require the full network observability [14] or high sampling rates (e.g., 10M Hz [10]). The last research line is based on supervised machine learning to locate faults on the bus or line-level [1,3,17]. These approaches show superior performance in efficiency and accuracy, especially in large-scale networks with low observability.…”

Traveling Wave Energy Analysis of Faults on Power Distribution Systems

“…In reference [4], the Dynamic Mode Decomposition (DMD) is the selected approach to find anomalies in the measured signal, which could indicate the presence of a fault. The amount of energy contained in the high-frequency components of the signal can also indicate the arrival of a TW [17]. Finally, Machine-Learning (ML) methods can be trained to detect incipient faults [18].…”

“…A widespread option is to decompose the signals by frequency bands applying WTs and to study them independently. This method has been successfully used for event characterization, fault classification, and fault location [2,17,18,[20][21][22][23]. This method is especially suitable for studying TWs as the fault signatures, and its frequency components are heavily influenced by the fault location and the propagation path.…”

Embedded, Real-Time, and Distributed Traveling Wave Fault Location Method Using Graph Convolutional Neural Networks

“…For this reason, many approaches use a time-frequency decomposition to analyze the TW frequency components. The works in [22,23] used the continuous wavelet transform (CWT) for this purpose, while [24,25] opted for discrete WTs. The reference [26] used the S-transform instead.…”