Fault Detection and Localization Techniques in Power Distribution Systems

Power distribution systems are vulnerable to a range of faults—including single-phase-to-ground, phase-to-phase and high-impedance faults—that can disrupt supply and damage equipment. Fault detection and localisation are essential for ensuring reliability and minimising outage duration. Classical approaches include impedance-based methods, which infer fault distance from voltage and current phasors, and travelling-wave techniques, which use high-frequency transients to pinpoint fault inception. Signal-processing methods such as wavelet and Stockwell transforms extract time-frequency features for classification. Recent advances in sensor technology, notably micro-phasor measurement units (micro-PMUs), provide high-resolution, time-synchronised measurements for real-time monitoring. Concurrently, machine learning has been deployed to automate feature extraction and enhance accuracy: support vector machines, k-nearest neighbours and, increasingly, deep neural networks (convolutional, recurrent and attention-based architectures) trained on simulated or field data. Hybrid schemes that integrate physics-based models with data-driven learning address data scarcity and improve generalisation. These techniques adapt to complex network topologies, distributed generation and smart-grid functionalities, delivering sub-cycle detection times and localisation errors of only a few per cent of line length.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Simulation-driven, physics-informed deep learning has achieved over 99 per cent accuracy in locating faulty insulators on overhead lines. By calibrating a line model to match measured leakage currents and generating extensive synthetic fault data, researchers trained long short-term memory and convolutional neural networks that automatically extract spatio-temporal features for both fault detection and precise localisation. An adaptive convolutional neural network has been proposed for two-terminal fault location in distribution feeders, improving classification accuracy by nearly 8 per cent and reducing training time by over 40 per cent compared with previous deep belief network approaches. This end-to-end model uses simulated fault currents to enable rapid convergence and high recognition rates for single-phase ground faults. In smart distribution networks equipped with micro-PMUs, a machine learning framework utilises voltage phasor recordings at substations and distributed generators to locate faults. Frequency-domain features extracted from micro-PMU data are dimensionally reduced by a neighbour-component feature selection algorithm, then classified by a support vector machine to achieve precise section-level fault identification regardless of fault characteristics or generation mix.

Fault Detection and Localization Techniques in Power Distribution Systems publication trend

The graph below shows the total number of articles in fault detection and localization techniques in power distribution systems across all publications each year (not limited to Nature Index journals).

Technical terms

Fault detection: The process of identifying the occurrence of an abnormal electrical event in a power system.

Fault localisation: Determining the physical location of a detected fault along a power line or network.

Impedance-based method: Technique that estimates fault distance by comparing measured voltage and current phasors.

Travelling-wave method: Approach using high-frequency transient waves generated by faults to calculate fault position.

Wavelet transform: Signal-processing tool that decomposes time-domain signals into time-frequency representations for feature extraction.

Phasor measurement unit (PMU): Device providing time-synchronised measurements of voltage and current phasors at high sampling rates.

Convolutional neural network: Deep learning architecture that automatically learns spatial hierarchies of features from input data.

References

1. Simulation-driven deep learning for locating faulty insulators in a power line. Reliability Engineering & System Safety (2023).
2. Two-Terminal Fault Location Method of Distribution Network Based on Adaptive Convolution Neural Network. IEEE Access (2020).
3. Machine Learning-Based Fault Location for Smart Distribution Networks Equipped with Micro-PMU. Sensors (2022).