主管：中华人民共和国应急管理部
主办：应急管理部天津消防研究所
ISSN 1009-0029  CN 12-1311/TU
消防科学与技术

消防科学与技术 ›› 2026, Vol. 45 ›› Issue (9): 46-52.

• • 上一篇    下一篇

基于SMOTE和ANN的故障电弧引燃电缆预测研究

陈斌1, 刘宜金2, 马俊铭2, 张孝春2   

  1. （1.深圳市消防救援支队，广东 深圳 518000； 2.广东工业大学 环境科学与工程学院，广东 广州 510006）
  • 收稿日期:2025-06-30 修回日期:2025-09-27 出版日期:2026-09-15 发布日期:2026-09-15
  • 作者简介:陈斌，深圳市消防救援支队，高级工程师，主要从事建筑防火、电气火灾安全研究工作，广东省深圳市红荔路2009号，518000。
  • 基金资助:
    国家重点研发计划项目（2024YFC3014702）；国家自然科学基金项目（52276108）

SMOTE and ANN based prediction study of fault arc igniting cable

Chen Bin1, Liu Yijin2, Ma Junming2, Zhang Xiaochun2   

  1. (1. Shenzhen Fire and Rescue Division, Shenzhen Guangdong 518000, China; 2. School of Environmental Science and Engineering, Guangdong University of Technology, Guangzhou Guangdong 510006, China)
  • Received:2025-06-30 Revised:2025-09-27 Online:2026-09-15 Published:2026-09-15

PDF (PC)

12

摘要： 针对电缆火灾难以预测这一难题，提出一种故障电弧引燃电缆的预测技术。搭建试验平台，模拟电弧故障引燃环境。选取5种常见的电缆护套材料，通过试验，模拟电弧故障条件下电缆的引燃情况。为了构建更为精准的预测模型，引入查准率、准确率等多项科学评估指标，并采用合成少数过采样技术对数据进行优化处理。在此基础上，建立了人工神经网络（ANN）模型，并加入了5折交叉验证环节，以验证模型的稳定性和可靠性，从而实现对电缆引燃情况的准确预测。此外，与支持向量机（SVM）模型进行对比分析。最终结果显示，ANN模型的准确率高达85.7%，明显优于SVM模型的72.73%。结果表明，所提出的模型能够有效预测电缆引燃，为电气火灾防治及电缆系统安全设计提供科学依据。

关键词: 电缆火灾, 合成少数过采样技术, 故障电弧, 人工神经网络

Abstract: To address the challenge of predicting cable fires, this paper proposes a prediction technology for cable ignition caused by fault arcs. In the research, an experimental platform was established to simulate the environment of arc fault ignition. Five common types of cable sheath materials were selected, and through experiments, the ignition conditions of cables under arc fault conditions were simulated. To build a more precise prediction model, this paper introduces multiple scientific evaluation indicators, such as precision and accuracy. The synthetic minority over-sampling technique (SMOTE) was used to optimize the data. Based on this, an artificial neural network (ANN) model was established with a 5-fold cross-validation step to verify the stability and reliability of the model, thereby achieving accurate prediction of cable ignition conditions. In addition, a comparative analysis was conducted with the Support Vector Machine (SVM) model. The final results show that the accuracy of the ANN model is as high as 85.7%, significantly better than the 72.73% accuracy of the SVM model. The results indicate that the proposed model can effectively predict cable ignition, providing a scientific basis for electrical fire prevention and the safety design of cable systems.

Key words: cable fire, synthetic minority over-sampling technique, fault arc, artificial neural network