Summary
The increasing penetration of low-carbon technologies (LCTs) has significantly altered fault characteristics in low-voltage alternating current (LVAC) distribution networks, challenging conventional protection schemes and motivating the need for robust fault detection and classification (FDC) with limited monitoring facilities. This paper proposes an AI driven and hybrid convolutional neural networks (CNN)- long short-term memory (LSTM)-Transformer framework for LVAC FDC. The architecture integrates convolutional layers for channel-wise feature enhancement, LSTM networks for modelling short-term fault transients, and
Read more Read lessTransformer encoders for capturing long-range temporal dependencies introduced by inverterinterfaced devices. The framework is evaluated on a detailed LVAC network model incorporating mixed single- and three-phase loads, and a 25kW photovoltaic-battery energy storage system. Under conventional load-only operation, the proposed method achieves 100.0% fault detection accuracy and fault classification accuracies above 99.7%, with an average detection latency of 0.14ms. With LCT integration, classification accuracies remain above 99.5%, with an average detection latency of 0.71ms. Sensitivity studies under noisy measurements and high-impedance fault conditions demonstrate strong robustness, with overall high-impedance fault recognition accuracy of approximately 99%.
Additional informations
| Publication type | Session Materials |
|---|---|
| Reference | B5_12071_2026 |
| Publication year | |
| Publisher | CIGRE |
| Country | United Kingdom |
| Study committees | |
| File size | 919 KB |
| Price for non member | 30 € |
| Price for member | 30 € |
Authors
YU James - CIGRE UK; FILGUEIRA Francisco - CIGRE SPAIN; DINIZ Mikaelle - Scottish Power Energy Networks UK; QI Yue - Glasgow University UK; LI Jiwei - Glasgow University UK; MA Miaorui - Glasgow University UK; YANG Jin - Glasgow University UK
Keywords
Low-voltage AC networks; fault detection and classification; deep learning; data-driven distribution network protection, LSTM
