基于卷积神经网络和 LSTM 网络的矿用变压器
故障诊断

金属矿山 ›› 2025, Vol. 54 ›› Issue (6): 168-173.

基于卷积神经网络和 LSTM 网络的矿用变压器故障诊断

孙　朋¹　刘超然² 　马建民¹

1. 许昌职业技术学院机电与汽车工程学院,河南许昌 461000;2. 杭州电子科技大学电子信息学院,浙江杭州 310000

出版日期:2025-06-15 发布日期:2025-07-09
通讯作者: 刘超然(1987—),男,副教授,博士,博士研究生导师。
作者简介:孙　朋(1990—),男,讲师,工程师,硕士。
基金资助:
河南省高等学校重点科研项目(编号:24B450002,24B430018);国家自然科学基金项目(编号: 62111530298)。

Fault Diagnosis of Mining Transformers Based on Convolutional Neural Networks and LSTM Networks

SUN Peng ¹ 　 LIU Chaoran² 　 MA Jianmin ¹

1. School of Mechatronics and Automotive Engineering,Xuchang Vocational Technical College,Xuchang 461000,China; 2. School of Electronics and Information Engineering,Hangzhou Dianzi University,Hangzhou 310000,China

Online:2025-06-15 Published:2025-07-09

摘要/Abstract

摘要： 矿用变压器作为矿山电力系统的核心设备,其运行状态直接影响矿山生产的安全性与效率。然而,由于矿山环境的复杂性和设备长期运行的特殊性,变压器故障诊断面临着高噪声、数据不平衡以及故障类型多样等挑战。为此,提出了一种基于卷积神经网络( Convolutional Neural Network,CNN) 和长短期记忆网络( Long Short-Term Memory,LSTM)的混合模型(CNN-LSTM),用于矿用变压器的故障诊断。首先利用 CNN 对变压器运行数据进行特征提取,有效捕捉数据中的空间特征;然后采用 LSTM 对提取的特征进行时序建模,识别数据中的动态变化模式。试验结果表明:CNN-LSTM 模型对于多个故障类型的平均诊断准确率达到了 92. 82%以上,显著优于传统诊断方法和单一神经网络模型,反映出该模型在提高诊断精度和鲁棒性方面具有显著优势,具有一定的应用前景。

关键词: 矿用变压器　故障诊断　卷积神经网络　长短期记忆网络

Abstract: As a core device in the mine power system,the operation status of the mine transformer directly affects the safety and efficiency of mine production. However,due to the complexity of the mine environment and the special nature of long-term operation of the equipment,the transformer fault diagnosis faces challenges such as high noise,data imbalance,and diverse fault types. To address these issues,a hybrid model based on Convolutional Neural Network (CNN) and Long ShortTerm Memory (LSTM),namely CNN-LSTM is proposed for the fault diagnosis of mine transformers. Firstly,CNN is utilized to extract features from the transformer operation data,effectively capturing the spatial features in the data. Subsequently,LSTM is employed to model the time series of the extracted features,identifying the dynamic change patterns in the data. The experimental results show that the proposed CNN-LSTM model achieves an average accuracy rate of over 92. 82% for diagnosing multiple fault types,significantly outperforming traditional diagnostic methods and single neural network models. This indicates that the model has significant advantages in improving diagnostic accuracy and robustness,and holds certain application prospects.

Key words: mining transformer,fault diagnosis,convolutional neural network,long short-term memory network

中图分类号:

TD67

孙　朋 , 刘超然, 马建民. 基于卷积神经网络和 LSTM 网络的矿用变压器故障诊断[J]. 金属矿山, 2025, 54(6): 168-173.

SUN Peng, LIU Chaoran, MA Jianmin . Fault Diagnosis of Mining Transformers Based on Convolutional Neural Networks and LSTM Networks [J]. Metal Mine, 2025, 54(6): 168-173.

扫码分享