Efficient Fault Detection and Categorization in Electrical Distribution Systems Using Hessian Locally Linear Embedding on Measurement Data

TL;DR

This research presents an effective methodology for robust fault detection and categorization in electrical systems, contributing to the advancement of fault management practices and the prevention of system failures.

Abstract

Faults on electrical power lines could severely compromise both the reliability and safety of power systems, leading to unstable power delivery and increased outage risks. They pose significant safety hazards, necessitating swift detection and mitigation to maintain electrical infrastructure integrity and ensure continuous power supply. Hence, accurate detection and categorization of electrical faults are pivotal for optimized power system maintenance and operation. In this work, we propose a novel approach for detecting and categorizing electrical faults using the Hessian locally linear embedding (HLLE) technique and subsequent clustering with t-SNE (t-distributed stochastic neighbor embedding) and Gaussian mixture model (GMM). First, we employ HLLE to transform high-dimensional (HD) electrical data into low-dimensional (LD) embedding coordinates. This technique effectively captures the inherent variations and patterns in the data, enabling robust feature extraction. Next, we perform the Mann-Whitney U test based on the feature space of the embedding coordinates for fault detection. This statistical approach allows us to detect electrical faults providing an efficient means of system monitoring and control. Furthermore, to enhance fault categorization, we employ t-SNE with GMM to cluster the detected faults into various categories. To evaluate the performance of the proposed method, we conduct extensive simulations on an electrical system integrated with solar farm. Our results demonstrate that the proposed approach exhibits effective fault detection and clustering across a range of fault types with different variations of the same fault. Overall, this research presents an effective methodology for robust fault detection and categorization in electrical systems, contributing to the advancement of fault management practices and the prevention of system failures.

Figures (7)

• Figure 1: Block diagram of the general process flow for fault detection using feature extraction.
• Figure 2: The complete process flow for fault detection and clustering using the proposed method; initially, the entire signal is divided into $N$ smaller segments. Then, HLLE transformation is applied to each segment, reducing the dimensionality from 6 to 1. Subsequently, the Mann-Whitney U test is used to determine the p-value for each transformed segment, using the HLLE of the first segment as the reference. These p-values are utilized to identify the starting point of the fault. It was observed that if the p-value is less than the selected threshold alpha value, a fault is detected. Once fault is detected, the following segments are considered as a fault event, and if the corresponding p-values during this event are for eight consecutive segments, then the fault event is considered to have ended. The average of all six signals consisting of three-phase voltages and currents, along with the average of HLLE values for the first fault segment, is extracted as a feature for performing clustering. After the data for clustering is extracted, it undergoes t-SNE dimensionality reduction and GMM clustering to distinctly identify the fault type.
• Figure 3: Schematic of the simulation model consisting of a 400 kW solar PV farm connected to the grid with three loads and two feeders.
• Figure 4: The measurement data consists of HD data with 14000 samples for 6 columns consisting of three-phase (a) voltages and (b) currents. For each 20 samples, we perform HLLE to transform the data into a lower one-dimension. Then, we compare the transformed 1D HLLE values of 20 samples with the 1D HLLE values of reference 20 samples. The (c) p-value provides the statistical information from which fault could be detected and key inferences are observed from the (d) zoomed p-values. Every set of 20 samples is considered as segments and converted into 1D HLLE values. The reference is the (e) 1st segment consisting of samples 1 to 20. This is compared with each segment of the data; there are a total of 700 segments for one fault case. At (f) 160th segment, the fault has started, which results in different 1D HLLE values compared to the reference. The variations in the 1D HLLE values could be observed all throughout the fault event as seen in the (g) 250th segment as well. Once the fault is cleared, the transformed values also return similar values of the no-fault segment as observed in (h) 400th segment.
• Figure 5: For fault cases - AG [(a), (b), (c), (d)], ABG [(e), (f), (g), (h)], TLG [(i), (j), (k), (l)], the measurements of three-phase voltages and currents are taken as input to the HLLE model where lower dimensional reduction is performed and statistical tests of each segment with the reference segment gives us p-values from which faults are detected and observations from the zoomed p-values provide the difference between in $t_f, t_E$ and $t_D$.