基于扩展频谱时域反射原理的电容反射波特性及容值估算方法  

Capacitance Reflection Wave Characteristics and Capacitance Estimation Method Based on Extended Spectrum Time Domain Reflection Principle

作　　者：成庶[1] 张多 刘畅 向超群[1] 吕壮壮 Cheng Shu;Zhang Duo;Liu Chang;Xiang Chaoqun;LüZhuangzhuang(School of Transportation Engineering Central South University,Changsha 410083 China)

机构地区：[1]中南大学交通运输工程学院,长沙410083

出　　处：《电工技术学报》2025年第4期1156-1168,共13页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金资助项目(52072414)。

摘　　要：电容是电力电子领域的核心元件,在多种应用场合都发挥着重要作用。基于此,该文提出一种利用扩展频谱时域反射原理估算电容容值的方法,根据注入信号与反射信号所做出的互相关波形,发现信号种类、信号频率、互相关系数峰值大小、互相关系数峰值对应时延4个特征参数与容值的映射关系显著,并以上述4个特征参数作为输入,建立基于遗传算法(GA)-BP神经网络的容值估算模型。结果表明,随着信号码长的增加,信号互相关系数图像旁瓣值越小,相关特性越好;信号频率通过影响电容的特征阻抗,进而影响信号的互相关波形;互相关系数峰值所对应时延与容值呈严格正相关趋势。同时,相较于传统的BP神经网络模型,GA优化后的模型误差减少了28.32%,估算精度在99%以上,精度高于传统BP神经网络模型,适合更好地挖掘各特征参数与容值之间的映射关系。Capacitors are the core devices for decoupling,filtering,and bypassing in power electronics.Therefore,identifying and detecting its operating state is essential.This paper proposes a method to estimate the capacitance value using the principle of extended-spectrum time domain reflection.Firstly,considering that the ceramic capacitor has excellent high-frequency response and small parasitic parameters,multiple pseudo-random signals modulated by BPSK are injected into the selected 15 groups of pF-level ceramic capacitors.Set the injection signal code length N to 7,31,63,and the injection signal frequency to 5 groups of integer frequency bands with 6~10 MHz.Secondly,the collected signals are preprocessed by filtering,normalization,and denoising,and the principal components are extracted.According to the cross-correlation waveform image of the injected and reflected signals,the mapping relationship between the four characteristic parameters(signal type,signal frequency,peak values of the cross-correlation coefficient with or without time delay)and capacitance is significant.Finally,using the above four characteristic parameters as inputs,a capacity estimation model is established based on the BP neural network.The initialization parameters are optimized by the GA genetic algorithm.The results show that the ceramic capacitor has good high-frequency response characteristics.Its parasitic parameters can be ignored in the high-frequency band,and the capacitance value almost determines the impedance.With the increase in signal length,the sidelobe value of the signal cross-correlation coefficient image is decreased,and the correlation characteristics are good.The signal frequency affects the cross-correlation waveform of the signal by affecting the characteristic impedance in the capacitor.Due to the capacitance’s time delay effect,the corresponding peak value of the cross-correlation coefficient is strictly positively correlated with the capacitance value.At the same time,compared with the traditional BP neural network

关 键 词：多层陶瓷电容(MLCC) 容值估算 扩展频谱时域反射法 遗传算法(GA)-BP 神经网络 

分 类 号：TM534[电气工程—电器]