>> 自然科学版期刊 >> 2024 >> 2024年05期 >> 正文
基于改进LSTM-SVM的双向DC-DC电力变换器故障诊断
供稿: 王福忠, 任淯琳, 张丽, 王丹 时间: 2024-07-31 次数:

王福忠, 任淯琳, 张丽,等.基于改进LSTM-SVM的双向DC-DC电力变换器故障诊断[J].河南理工大学学报(自然科学版),2024,43(5):118-126.

WANG F Z, REN Y L, ZHANG L,et al.Fault diagnosis of bidirectional DC-DC power converter based on improved LSTM-SVM[J].Journal of Henan Polytechnic University(Natural Science) ,2024,43(5):118-126.

基于改进LSTM-SVM的双向DC-DC电力变换器故障诊断

王福忠1, 任淯琳1, 张丽1, 王丹2

1.河南理工大学 电气工程与自动化学院,河南 焦作  454000;2.黄河交通学院 智能工程学院,河南 焦作  454950

摘要: 目的 为了解决双向DC-DC电力变换器的软故障诊断精度不高的问题,  方法 提出基于改进LSTM-SVM的双向DC-DC电力变换器故障诊断模型。首先,分析双向DC-DC电力变换器中电容、电感和MOSFET管的故障机理,通过仿真实验模拟各元件失效后变换器的输出电气参数变化,从而确定变换器不同元件故障时对应的故障特征参数;其次,构建改进的LSTM-SVM双向DC-DC电力变换器故障诊断组合模型,在LSTM中添加Mogrifier门机制,提高LSTM提取时间序列原始数据中微弱特征的能力;最后,由于传统LSTM的末端分类器为Softmax,其主要解决单一元件诊断问题,变换器故障类型较多,维数较高,所以采用麻雀搜索算法优化的SVM代替原有的Softmax函数,对LSTM输出的数据进行故障分类,提高故障诊断的准确率。设置双向DC-DC电力变换器充放电两种状态下,包含电解电容、电感和MOSFET单双管故障在内的24组故障,分别采用本文构建的改进的LSTM-SVM和原始的LSTM-SVM双向DC-DC变换器故障诊断模型进行诊断。  结果 结果表明,改进的LSTM-SVM故障诊断模型诊断准确率平均值为99.71%,原始的LSTM-SVM故障诊断模型诊断准确率平均值为88.48%,改进的LSTM-SVM故障诊断模型对各元件的故障诊断正确率均高于原始的LSTM-SVM故障诊断模型的。  结论 基于改进LSTM-SVM的双向DC-DC电力变换器故障诊断模型实现了对双向DC-DC电力变换器中的电解电容、电感和MOSFET单双管故障的准确诊断。

关键词:双向DC-DC变换器;软故障;改进长短期记忆网络;麻雀搜索;支持向量机;故障诊断

doi:10.16186/j.cnki.1673-9787.2022060019

基金项目:国家自然科学基金资助项目(U1804143);河南省科技攻关项目(232102241028;202102210295);河南省高校基本科研业务费专项项目(NSFRF210424);河南理工大学青年骨干教师资助项目(2019XQG-17)。

收稿日期:2022/06/09

修回日期:2023/04/20

出版日期:2024/07/31

Fault diagnosis of bidirectional DC-DC power converter based on improved LSTM-SVM

WANG Fuzhong1, REN Yulin1, ZHANG Li1, WANG Dan2

1.School of Electrical Engineering and Automation,Henan Polytechnic University,Jiaozuo  454000,Henan,China;2.School of Intelligent Engineering,Huanghe Jiaotong University,Jiaozuo  454950,Henan,China

Abstract: Objectives In order to solve the problem of low accuracy of soft fault diagnosis for bidirectional DC-DC power converter, Methods the fault diagnosis model of bidirectional DC-DC power converter based on improved LSTM-SVM was proposed.Firstly,the fault mechanisms of capacitors,inductors and MOSFET tubes in bidirectional DC-DC power converter were analyzed.The variations of the output electrical parameters of the converter after the failure of each component were simulated by simulation experiment,and the fault characteristic parameters corresponding to the failure of different components of the converter were determined.Then,an improved LSTM-SVM bidirectional DC-DC power converter fault diagnosis model was constructed.The Mogrifier gate mechanism was added to LSTM to improve the ability of LSTM to extract weak features from the original time series data.Finally,since the end classifier of traditional LSTM was Softmax,it mainly solved the problem of single component diagnosis,the converter had many fault types and high dimension,so SVM optimized by sparrow search algorithm was used instead of the original Softmax function to classify faults from LSTM output data and to improve the accuracy of fault diagnosis.24 groups of faults including electrolytic capacitor,inductor and MOSFET single and double tube faults were set up under two states of charge and discharge of bidirectional DC-DC power converter.The improved LSTM-SVM constructed in this paper and the original LSTM-SVM bidirectional DC-DC converter fault diagnosis model were respectively used for diagnosis.  Results The average accuracy of the improved LSTM-SVM fault diagnosis model was 99.71%,and the average accuracy of the original LSTM-SVM fault diagnosis model was 88.48%.The accuracy of the improved LSTM-SVM fault diagnosis model for each component was higher than that of the original LSTM-SVM fault diagnosis model. Conclusions The fault diagnosis model of bidirectional DC-DC power converter based on improved LSTM-SVM was realized to accurately diagnose the electrolytic capacitor,inductor and MOSFET single and double tube faults in bidirectional DC-DC power converter.

Key words:bidirectional DC-DC converter;soft fault;improved long and short term memory network;sparrow search;support vector machine;fault diagnosis

最近更新