Open Access

ARTICLE

Research on Low Voltage Series Arc Fault Prediction Method Based on Multidimensional Time-Frequency Domain Characteristics

Feiyan Zhou^1,*, Hui Yin¹, Chen Luo², Haixin Tong², Kun Yu², Zewen Li², Xiangjun Zeng²

1 China Electric Power Research Institute Distribution Technology Center, China Electric Power Research Institute, Beijing, 100192, China
2 National Key Laboratory of Disaster Prevention and Reduction for Power Grid, Changsha University of Science and Technology, Changsha, 410114, China

* Corresponding Author: Feiyan Zhou. Email:

Energy Engineering 2023, 120(9), 1979-1990. https://doi.org/10.32604/ee.2023.029480

Received 21 February 2023; Accepted 17 May 2023; Issue published 03 August 2023

Abstract

The load types in low-voltage distribution systems are diverse. Some loads have current signals that are similar to series fault arcs, making it difficult to effectively detect fault arcs during their occurrence and sustained combustion, which can easily lead to serious electrical fire accidents. To address this issue, this paper establishes a fault arc prototype experimental platform, selects multiple commonly used loads for fault arc experiments, and collects data in both normal and fault states. By analyzing waveform characteristics and selecting fault discrimination feature indicators, corresponding feature values are extracted for qualitative analysis to explore changes in time-frequency characteristics of current before and after faults. Multiple features are then selected to form a multi-dimensional feature vector space to effectively reduce arc misjudgments and construct a fault discrimination feature database. Based on this, a fault arc hazard prediction model is built using random forests. The model’s multiple hyperparameters are simultaneously optimized through grid search, aiming to minimize node information entropy and complete model training, thereby enhancing model robustness and generalization ability. Through experimental verification, the proposed method accurately predicts and classifies fault arcs of different load types, with an average accuracy at least 1% higher than that of the commonly used fault prediction methods compared in the paper.

---

Graphic Abstract

---

Keywords

---