基于非线性特征提取的模拟电路状态识别研究

Study of State Recognition for Analog Circuit Based on Nonlinear Feature Extraction

摘要:

针对常用的线性特征提取方法只能处理线性数据的局限性,提出通过核方法将线性特征提取方法扩展到非线性方面;首先提取模拟电路的幅频响应；然后针对信号特征存在冗余性及高维性,采用核Fisher判别分析法(KFD)对电压特征进行提取；选择SVM作为状态监测器,考虑到SVM的参数设置对识别率有较大影响,所以 采用PSO对SVM进行优化；实验结果表明,采用本方法的状态识别率达到70%,高于其他两种方法；说明运用KFD进行特征降维以及采用PSO对SVM进行优化能明显提高状态监测的识别率,体现了本方法良好的模拟电路早期状态监测能力。

Abstract:

Aiming at the common linear feature extraction methods can only deal with the limitation of linear data, a method of linear feature extraction extended to nonlinear by using kernel function is proposed. First extracting the amplitude frequency response of analog circuit; then due to redundancy and high domain of original signal features, the voltage features are obtained by using the Kernel Fisher Discriminator (KFD); Select SVM as the state monitor, considering the SVM parameters have great effect on the recognition rate, so using PSO to optimize SVM. The experimental results show that, the state recognition rate of this method can reach 70%, higher than the other two methods. It shows that using KFD for feature extraction and the PSO for SVM can significantly improve the recognition rate of state monitoring, reflects the new method has excellent capacity of monitoring incipient fault.