基于交叉熵方法和支持向量机的模拟电路故障诊断

Analog circuit fault diagnosis based on cross-entropy method and support vector machine

针对故障诊断系统中存在的大量无关或冗余的特征会严重影响故障诊断性能的缺陷,提出了基于交叉熵和支持向量机方法进行特征选择和参数优化的故障诊断方法.首先以某种概率分布产生若干随机样本,并依据交叉熵最小原理建立分布参数的更新规则进行特征搜索和SVM参数优化;然后利用优化后的特征向量和参数训练支持向量机获得故障诊断模型.故障诊断实验结果表明,该故障诊断方法能有效地优化故障特征和模型参数,提高故障诊断性能.

Considering that many irrelevant or redundant features in fault diagnosis system seriously spoil the fault diagnosis performance, a fault diagnosis method based on the cross entropy method and support vector machine is proposed. Firstly, a population of random variable samples is generated by some kinds of probability distribution, and the object value of the samples is evaluated by using SVM classifiers. Parameter-updating rule of distribution parameters is established based on min-cross-entropy theory. After several iterations, the best object feature subset and optimized parameters are selected out. Then the CEM-SVM model of the circuit fault diagnosis system is built by training the SVM with optimized features and parameters. Finally, analog circuit fault diagnosis experiment on leapfrog filter shows the effectiveness of feature selection and parameters optimization of the proposed method which improve the fault classification rate and the speed fault diagnosis time.