首页|期刊导航|计算机与现代化|轻量化Swin Transformer的充电设备复杂故障诊断方法

轻量化Swin Transformer的充电设备复杂故障诊断方法OA

Complex Fault Diagnosis of Charging Equipment Based on Lightweight Swin Transformer Model

中文摘要英文摘要

针对现有深度学习方法在处理充电设备复杂多变的故障时,存在局部与全局信息捕捉能力不足的问题,本文提出一种轻量化的基于Swin Transformer模型的充电设备故障诊断方法.该方法首先在原始Swin Transformer模型的第4个Stage层中引入轻量化的稀疏双曲正切(SparseTanh)模块,以替换原有的Swin Transformer Block模块,利用稀疏令牌机制减少模型参数量、降低模型计算复杂度,提高全局信息捕捉能力.在此基础上,在第1个Stage层中嵌入金字塔压缩注意力模块,通过提前引入多尺度和通道注意力机制,来增强模型对充电设备故障曲线局部细节特征的感知能力,提升故障识别的准确性.本文以直流充电桩真实运行报文数据为研究对象,构建充电设备故障图像数据集,进行实验分析.实验结果表明,所提方法在多分类故障诊断任务中准确率达到86.05%,且精确率、召回率及F1值相较对比模型均有更优异的表现.同时,较原始Swin Transformer模型降低了模型参数量与计算开销,增强了模型的实时性.研究表明,本文方法能有效兼顾充电设备复杂故障诊断的准确率与计算效率,对智能运维和工程部署具有较高的应用价值.

Aiming at the problems such as insufficient ability to capture local and global information when existing deep learning methods deal with the complex and changeable faults of charging equipment,this paper proposes a lightweight fault diagnosis method for charging equipment based on the Swin Transformer model.This method first introduces the lightweight SparseTanh module in the fourth stage layer of the original Swin Transformer model to replace the original Swin Transformer Block module.The sparse token mechanism is utilized to reduce the number of model parameters,lower the computational complexity of the model,and enhance the global information capture ability.On this basis,a Pyramid Squeeze Attention module is embedded in the first stage layer.By introducing multi-scale and channel attention mechanisms in advance,the model's perception ability of the local detail features of the charging equipment fault curve is enhanced,and the accuracy of fault identification is improved.This paper takes the real operation message data of DC charging piles as the research object,constructs a fault image dataset of charging equipment,and conducts experimental analysis.The experimental results show that the proposed method achieves an accuracy rate of 86.05%in the multi-classification fault diagnosis task,and its precision rate,recall rate and F1 value all per-form outstandingly compared with the comparison models.Meanwhile,compared with the original Swin Transformer model,the number of model parameters and computational overhead have been reduced,and the real-time performance of the model has been enhanced.Research shows that the method proposed in this paper can effectively balance the accuracy and computational ef-ficiency of complex fault diagnosis of charging equipment,and has high application value for intelligent operation and mainte-nance and engineering deployment.

张建洲;张建伟;张建军;云秋晨;杨天林;汪映辉

国网电力科学研究院有限公司,江苏 南京 211100||国电南瑞科技股份有限公司,江苏 南京 211106国网电力科学研究院有限公司,江苏 南京 211100||国电南瑞科技股份有限公司,江苏 南京 211106国网山东省电力公司菏泽供电公司,山东 菏泽 274000国网智慧车联网技术有限公司,北京 100032国网智慧车联网技术有限公司,北京 100032国网电力科学研究院有限公司,江苏 南京 211100||国电南瑞科技股份有限公司,江苏 南京 211106

信息技术与安全科学

充电设备故障诊断Swin Transformer稀疏双曲正切模块金字塔压缩注意力

charging equipmentfault diagnosisSwin TransformerSparseTanh modulepyramid squeeze attention

《计算机与现代化》 2026 (9)

46-54,9

国家电网公司总部科学技术项目(5400-202371588A-3-2-ZN)

10.3969/j.issn.1006-2475.2026.09.006

评论