摘要
为提高多尺度散布熵对信号演化敏感度,提升变压器故障声纹诊断准确率,将分数阶精细复合多尺度散布熵(Fractional Refined Composite Multiscale Dispersion Entropy,FRCMDE)运用于变压器声纹特征提取。首先,确定FRCMDE参数,提取不同状态下变压器声音信号的FRCMDE熵特征;其次,采用改进蝴蝶算法(Improved Butterfly Optimization Algorithm,IBOA)对最小二乘支持向量机(Least Squares Support Vector Machine,LSSVM)进行参数优化并构建IBOA-LSSVM模型,利用IBOA-LSSVM模型对特征数据进行分类,实现变压器故障声纹诊断;最后,为验证该方法的有效性,将其与其他经典方法比较,研究结果表明:所建FRCMDE-IBOA-LSSVM模型可有效区分8种状态下的变压器声音信号,诊断准确率达到99.69%,均高于其他方法。该方法可为变压器不停电监测与故障声纹诊断提供参考。
In order to improve the sensitivity of multiscale dispersion entropy to signal evolution and raise the accuracy of transformer fault voiceprint diagnosis,the fractional-refined composite multiscale dispersion entropy(FRCMDE)was applied to transformer voiceprint feature extraction.Firstly,the parameters of FRCMDE are determined and the FRCMDE entropy features of transformer sound signals in different states were extracted.Secondly,the Improved Butterfly Optimization Algorithm(IBOA)was introduced to optimize the parameters of the Least Square Support Vector Machine(LSSVM)to construct an IBOA-LSSVM model,which was used to classify data features for transformer fault voiceprint diagnosis.Finally,to verify the effectiveness of this method,its result was compared with that of other classical methods.The research results show that the proposed FRCMDE-IBOA-LSSVM model can effectively distinguish transformer′s sound signals in 8 states,and the diagnosis accuracy reaches 99.69%,which is higher than other methods.This method may provide a reference for transformer non-stop monitoring and fault voiceprint diagnosis.
作者
高家通
康兵
许志浩
王宗耀
丁贵立
袁小翠
GAO Jiatong;KANG Bing;XU Zhihao;WANG Zhongyao;DING Guili;YUAN Xiaocui(School of Electrical Engineering,Nanchang Institute of Technology,Nanchang 330099,China;Jiangxi Engineering Research Center of High Power Electronics and Grid Smart Metering,Nanchang Institute of Technology,Nanchang 330099,China;Jiangxi Paiyuan Power Technology Co.,Ltd.,Nanchang 330099,China)
出处
《噪声与振动控制》
北大核心
2025年第5期123-130,共8页
Noise and Vibration Control
基金
国家自然科学基金资助项目(62001202)。
