针对单一的隐马尔科夫模型(HMM)或支持向量机(SVM)在模拟电路早期的软故障中识别率不高的特点, 将HMM-SVM混合模型应用到模拟电路早期的软故障识别中。首先通过主成分分析(PCA)将原始数据样本降维实现初步划分; 接着利用HMM计算测试样本与各故障状态的匹配程度形成特征向量; 最后由SVM做故障状态判别。实验结果表明, HMM-SVM混合模型的早期故障识别率优于单一的HMM或SVM模型, 将平均故障识别率提高到95%以上。

Since the incipient faults of analog circuit are hard to be identified well by using only Hidden Markov Model (HMM) or Support Vector Machine (SVM) a new fault diagnosis method based on HMM-SVM was proposed.Firstly the dimensions of the experimental samples were decreased and classified briefly by Principal Components Analysis (PCA).Then HMM was used to calculate the matching degree between the test samples and all the fault states which formed the feature vectors for SVM in final diagnosis.The result shows that HMM-SVM is better than single HMM or SVM model for the incipient fault diagnosis and the average fault recognition rate was increased by more than ninety-five percent.