基于PSO-PF-SVM的模拟电路故障诊断

Fault diagnosis of simulation circuit based on PSO-PF-SVM

针对模拟电路输出信号存在的非线性、高维数等特点所带来的诊断困难问题,提出一种支持向量机(SVM)分类器参数优化算法,进行模拟电路故障诊断.首先,运用S变换与灰度共生矩阵(GLCM)组合方法S-GLCM,对电路输出信号进行故障特征提取.其次,采用粒子群算法(PSO)与粒子滤波算法(PF)融合,通过重采样实时更新粒子的位置和速度,对SVM参数进行高效寻优,并将特征向量代入模型中进行训练和测试,完成对电路各故障模式的高精度故障诊断.最后,通过两个国际基准电路试验对该方法进行可靠性分析.试验结果表明:S-GLCM在处理非线性、非平稳信号时表现出很大优势,将电路输出信号每组1500个采样点降为8维特征向量,减少冗余信息;该SVM分类器参数优化算法的诊断准确率较未优化算法提升约11.2%.

To solve the difficult diagnosis problem caused by the nonlinearity and high dimensionality of the analog circuit output signal,the support vector machine(SVM)classifier parameter optimization algorithm was proposed for analog circuit fault diagnosis.The fault features were extracted from the circuit output signal by the combination S-GLCM of S-transform and gray-level co-generation matrix(GLCM)method.The particle swarm algorithm(PSO)fused with particle filtering algorithm(PF)was used to update the position and velocity of particles in real time by resampling to efficiently optimize the SVM parameters,and the feature vectors were brought into the model to conduct the training and testing and complete the high-precision fault diagnosis of each fault mode of the circuit.The reliability of the method was analyzed by two international benchmark circuit experiments.The results show that S-GLCM has great advantages in the processing of nonlinear and non-stationary signals,and each group of 1500 sampling points of the circuit output signal can be reduced to 8-dimensional feature vector with reduced redundant information.Compared with the unoptimized algorithm,the diagnostic accuracy of the proposed method is improved by about 11.2%.