摘要
研究电路的故障问题,应提高快速性和准确性。为提高仿真电路故障诊断效率,给出了一种基于改进支持向量机的仿真电路故障诊断方法。首先通过小波包变换实现了信号的能量特征提取,根据主元分析完成了特征压缩;其次针对支持向量机多分类一对一方法存在的不可分类区,将其与最近邻分类法相结合,实现了电路的故障诊断,并提出了一种混合遗传算法实现了小波函数和支持向量机参数的同步选择;最后通过一仿真电路的仿真实验,与BP,RBF和PNN等神经网络对比,结果显示基于支持向量机的方法诊断精度最高,达到98%,为设计提供参考依据。
To improve the efficiency of analog circuit fault diagnosis,a method based on improved SVM is proposed.At first,energy feature extraction is realized through wavelet packet transform and feature compression is completed based on PCA.Secondly,multi-class one vs.one method of SVM combined with nearest neighbors method carries out the faults classification.And,a synthesis GA is introduced to choose wavelet function and parameters of SVM at the same time.At last,improved SVM is compared with BP,RBF,and PNN through simulation experiment of an analog circuit.The result shows that the method of SVM is much better than others,whose correct rate is 98%.
出处
《计算机仿真》
CSCD
北大核心
2010年第1期346-350,共5页
Computer Simulation
