Locating short-circuit faults in HVDC systems using automatically selected frequency-domain features

Abstract

Summary In this paper, a novel fault‐location method is presented for high‐voltage direct current (HVDC) transmission lines based on the pattern recognition techniques and the machine learning strategies. In the proposed method, the voltage signal at one of the HVDC stations is stepped down via a resistive‐capacitive voltage divider (RCVD) and passed through an anti‐aliasing low‐pass active filter (LPAF). Then, the frequency spectrum is obtained by applying the discrete Fourier transform (DFT) to the postfault voltage signal. The input pattern for presenting to the fault‐location estimator is formed based on the most useful features selected from the extracted harmonic spectrum. In this paper, the regression relief (RReliefF) algorithm is utilized for automatic feature selection. Also, the random forest (RF) algorithm is used to build the fault‐location estimator founded on a group of regression decision trees. The method is applied for fault locating in a 700‐km‐long HVDC line considering various fault locations, short‐circuit resistances, prefault currents, and fault types. The obtained overall average of percentage errors in the fault‐location estimate for 1800 different unseen test cases is 0.188%, which confirms the accurate and good generalization performance of the presented method.

A novel approach to estimate fault location in current source converter–based HVDC transmission line by Gaussian process regression

“…The comparative study between the proposed and conventional methods for the different critical application such as traditional CSC-based HVSC and VSC-based HVDC system is summarized in Table 9. [13][14][15][16][17][18] This study reveals that the proposed method has less complexity in comparison with Tzelepis et al 13 as the scheme proposed by the author requires installation of numerous current sensors comprising of highly synchronized two-end current data along the total length of the transmission line. [13][14][15][16][17][18] This study reveals that the proposed method has less complexity in comparison with Tzelepis et al 13 as the scheme proposed by the author requires installation of numerous current sensors comprising of highly synchronized two-end current data along the total length of the transmission line.…”

“…The presented paper is more superior to the earlier proposed methods [13][14][15][16][17][18] in literature as it demands a lesser value of sampling frequency and helps to achieve higher accuracy than other reported techniques. [13][14][15][16][17][18] This study reveals that the proposed method has less complexity in comparison with Tzelepis et al 13 as the scheme proposed by the author requires installation of numerous current sensors comprising of highly synchronized two-end current data along the total length of the transmission line. In Hao et al, 14 the fault location is estimated accurately, but this method requires a high value of sampling frequency up to 1000 kHz, and with compromise in sampling frequency up to 250 kHz, the fault location error is raised up to 12 times that of normal.…”

“…The determination of fault location in a multi-terminal HVDC system using pattern matching-based method is studied in Tzelepis et al 13 by employing the continuous wavelet transform and Pearson correlation coefficient. The novel method, which exploits DFT as feature extraction technique and for feature selection, regression relief algorithm, is used over one-end postfault voltage signal to locate the fault in HVDC transmission line accompanying random forest algorithm for locating different types of fault is presented in Mohammad 15 Multi-layer perceptron neural network (MLPNN) has been used as the fault protection scheme for a multi-terminal HVDC network in Yang et al 16 It demands the frequency spectrum of the current signal along with MLPNN for the purpose of detection, classification, and estimation of fault location in the DC transmission line. In Hao et al, 14 an intellective approach, which utilizes different techniques such as ε-SVR for fault distance estimation, the Hilbert-Huang transform for input feature extraction, and bat algorithm in order to optimize the parameter of ε-SVR, is presented to determine unipolar ground faults in HVDC transmission line.…”

“…This technique provides the merit of using only one-end current data, while on the other hand to sustain the accuracy of preceding techniques it demands high frequency up to 1000 kHz. The novel method, which exploits DFT as feature extraction technique and for feature selection, regression relief algorithm, is used over one-end postfault voltage signal to locate the fault in HVDC transmission line accompanying random forest algorithm for locating different types of fault is presented in Mohammad 15 Multi-layer perceptron neural network (MLPNN) has been used as the fault protection scheme for a multi-terminal HVDC network in Yang et al 16 It demands the frequency spectrum of the current signal along with MLPNN for the purpose of detection, classification, and estimation of fault location in the DC transmission line. In this technique, even though the MLPNNs-based fault detector and classifier demand only one-end current signal in order to extract the input feature; nevertheless, the MLPNN-based fault locator 16 has an essential need of two-end current signals to extract the required input frequency samples.…”

Non‐characteristicharmonic resonance in line commutated converters‐high‐voltage direct current terminals: A practical case study

“…As high‐voltage direct‐current (HVDC) transmission has advantageous performances in long‐distance bulk‐power delivery, it has been widely employed in the interconnection of power grids 1‐5 . However, the inherent non‐linear feature of power converters makes it generate abundant harmonics.…”