基于特征优选模拟电路故障诊断方法

Method for Fault Diagnosis of Analog Circuits Based on Feature Selection

传统基于最小二乘支持向量机模拟电路故障诊断方法都是使用单一的特征向量组合训练支持向量机所有二分类器,然而实际上每个二分类器对不同的特征向量组合有不同的分类精度。因此,提出了基于马氏距离的粒子群优化算法,为最小二乘支持向量机所有二分类器优选出近最优的特征向量组合。然后,将近最优特征向量组合用于训练和测试该支持向量机。最后把该方法应用于模拟电路早期故障诊断,实验结果表明,基于近最优特征向量组合的诊断精度要高于单一特征向量组合的诊断精度。

Traditionally, multi-fault diagnosis of analog circuits based on least squares support vector machine （LSSVM） usually uses a single feature vector combination to train all binary LSSVM classifiers. However, in fact, each binary LSSVM classifier has different classification accuracy for different feature vector combinations. Therefore, the Mahalanobis distance （MD） based on particle swarm optimization （PSO） is proposed to select a near-optimal feature vector combination for each binary classifier. Then, the near-optimal feature vector combinations are used to train and test LSSVM for diagnostics of the incipient faults in analog circuits. The experimental results show that the accuracy using the near-optimal feature vector combinations is higher than the accuracy using a single vector combination.