[Paper Review] A Feature Selection Method for High Impedance Fault Detection

Qiushi Cui, Khalil El‐Arroudi+1 more|arXiv (Cornell University)|Aug 13, 2018

Power Systems Fault Detection13 references4 citations

TL;DR

This paper proposes a systematic feature selection framework for high impedance fault (HIF) detection in distribution networks using a hybrid approach combining signal processing, expert knowledge, and Minimum Description Length (MDL)-based feature ranking. By extracting time, duration, and magnitude features via DFT and Kalman filtering, and integrating power system expert insights, the method achieves high detection reliability and security across diverse fault scenarios, outperforming traditional threshold-based techniques.

ABSTRACT

High impedance fault (HIF) has been a challenging task to detect in distribution networks. On one hand, although several types of HIF models are available for HIF study, they are still not exhibiting satisfactory fault waveforms. On the other hand, utilizing historical data has been a trend recently for using machine learning methods to improve HIF detection. Nonetheless, most proposed methodologies address the HIF issue starting with investigating a limited group of features and can hardly provide a practical and implementable solution. This paper, however, proposes a systematic design of feature extraction, based on an HIF detection and classification method. For example, features are extracted according to when, how long, and what magnitude the fault events create. Complementary power expert information is also integrated into the feature pools. Subsequently, we propose a ranking procedure in the feature pool for balancing the information gain and the complexity to avoid over-fitting. For implementing the framework, we create an HIF detection logic from a practical perspective. Numerical methods show the proposed HIF detector has very high dependability and security performance under multiple fault scenarios comparing with other traditional methods.

Motivation & Objective

• To address the challenge of detecting high impedance faults (HIFs) in distribution networks, which often evade conventional overcurrent relays due to low fault currents.
• To overcome the limitations of ad hoc feature selection in existing HIF detection methods by establishing a comprehensive, systematically designed feature pool.
• To develop a robust, implementable HIF detection logic using mature, low-cost techniques like DFT and Kalman filtering.
• To balance information gain and model complexity through MDL-based feature ranking, reducing overfitting and improving generalization.
• To evaluate the method’s performance across multiple fault scenarios, including varying X/R ratios and fault resistance levels.

Proposed method

• The feature pool is constructed by extracting temporal features (onset time, duration), magnitude features (harmonic content via DFT), and fault intensity indicators (using Kalman filter for harmonic coefficient estimation).
• Expert-derived features, such as the angle difference between zero and negative sequence voltages, are integrated to enhance physical interpretability and detection accuracy.
• A Minimum Description Length (MDL)-based ranking algorithm is applied to prioritize features that maximize information gain while minimizing model complexity.
• The detection logic is implemented using simple, practical components like logic gates and widely deployed techniques (DFT, KF), ensuring real-time feasibility.
• The framework is evaluated using a dynamic HIF model with randomly varying arc resistance and voltage sources, simulating realistic fault behavior.
• Performance is validated across multiple fault scenarios, including different fault resistance levels and X/R ratios, using a benchmark distribution feeder model.

Experimental results

Research questions

• RQ1 How can a comprehensive and systematic feature pool be designed to capture the diverse characteristics of high impedance faults?
• RQ2 What is the optimal balance between information gain and model complexity in feature selection for HIF detection?
• RQ3 Can MDL-based feature ranking effectively identify a minimal yet highly informative feature set for HIF detection?
• RQ4 How does the proposed method perform across diverse fault scenarios, including varying fault resistance and system X/R ratios?
• RQ5 Can the detection logic be implemented using mature, low-cost, and widely deployable techniques like DFT and Kalman filtering?

Key findings

• The proposed feature selection framework significantly improves HIF detection reliability and security compared to traditional threshold-based methods.
• The MDL-based ranking procedure effectively reduces overfitting by selecting features that maximize information gain while minimizing complexity.
• The method maintains consistent performance across different HIF models and fault scenarios, including varying fault resistance and X/R ratios.
• The use of DFT and Kalman filtering enables accurate, real-time harmonic estimation with low computational cost and high reliability.
• The integration of expert knowledge—such as zero- and negative-sequence voltage angle differences—enhances detection accuracy without increasing model complexity.
• Numerical results show that the method achieves high detection performance even under low fault current conditions (10–50 A), where conventional relays often fail.