Quantized SNNs for Smart Grid Fault Detection
International Journal of Scientific Research in Computer Science, Engineering and Information Technology · 25.09.2025 · 11 (5) · 539-565
DOI: 10.32628/cseit26121466 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Scientific Research in Computer Science, Engineering and Information Technology, online erschienen am 25.09.2025. Verfasst von Benjamin Agyekum und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This paper proposes a novel Quantized Event-Driven Spiking Neural Network, named Q-GridSNN, for real-time smart grid fault detection under noisy, dynamic, and resource-constrained operating conditions. The study addresses the computational burden of conventional deep learning models deployed for transmission-line and distribution-network fault diagnosis by combining spike-based temporal encoding, quantization-aware training, adaptive membrane-potential scaling, and lightweight edge inference. Voltage, current, phasor measurement unit signals, and derived symmetrical components are encoded into sparse spike trains using latency-rate hybrid coding, allowing the model to capture transient fault signatures with reduced switching activity. The proposed Q-GridSNN is benchmarked against Support Vector Machine, Random Forest, CNN, LSTM, CNN-LSTM, Graph Neural Network, full-precision SNN, and baseline Quantized SNN models. Evaluation is performed using multiclass fault conditions, including line-to-ground, line-to-line, double-line-to-ground, three-phase faults, high-impedance faults, and normal operating states. Comparative results show that Q-GridSNN achieves superior performance, with a projected accuracy of 99.31%, precision of 99.18%, recall of 98.96%, F1-score of 99.07%, and fault localization error below 0.42 km. The model also reduces inference latency to 2.6 ms and lowers estimated energy consumption by 58.4% compared with full-precision SNN and by 71.2% compared with CNN-LSTM. Graphical analysis using confusion matrices, ROC curves, precision-recall curves, latency-energy trade-off plots, and quantization sensitivity graphs confirms that the proposed model preserves diagnostic accuracy while substantially improving computational efficiency. The findings demonstrate that quantized SNNs can provide a practical pathway for intelligent, low-power, and fast fault detection in smart grids, especially for edge-based protection relays, distributed substations, and cyber-physical power-system monitoring platforms.
Zitieren
APA
Agyekum, B., Seffah-Duodu, D., & Gloria, A. (2025). Quantized SNNs for Smart Grid Fault Detection. International Journal of Scientific Research in Computer Science, Engineering and Information Technology, 11(5), 539-565. https://doi.org/10.32628/cseit26121466
Vancouver
Agyekum B, Seffah-Duodu D, Gloria A. Quantized SNNs for Smart Grid Fault Detection. International Journal of Scientific Research in Computer Science, Engineering and Information Technology. 2025;11(5):539-565. doi:10.32628/cseit26121466
BibTeX
@article{Agyekum2025,
author = {Agyekum, Benjamin and Seffah-Duodu, Daniel and Gloria, Agyapong},
title = {Quantized SNNs for Smart Grid Fault Detection},
journal = {International Journal of Scientific Research in Computer Science, Engineering and Information Technology},
year = {2025},
volume = {11},
number = {5},
pages = {539--565},
doi = {10.32628/cseit26121466}
}
EUVIMED