基于PSO-LSSVM的模拟电路故障诊断

Analog Circuit Fault Diagnosis Based on PSO-LSSVM

针对传统神经网络技术在模拟电路故障诊断应用中存在的问题,提出了一种基于粒子群(Particle Swarm Optimization,PSO)和最小二乘支持向量机(Least Squares Support Vector Machine,LSSVM)的模拟电路故障诊断方法。该方法先从一个滤波器系统的频率响应数据中提取由小波系数的均值、标准差和熵构成的频率小波特征向量来训练最小二乘支持向量机,之后再采用粒子群算法来优化支持向量机的结构参数,避免了参数选择的盲目性,进而提高了模型的诊断精度。在对Elliptical Filter电路进行的故障检测中,验证了该方法的可行性。

In order to solve the problem of fault diagnosis method for analog IC diagnosis based on neural network,the method based on Particle Swarm Optimization （ PSO） and Least Squares Support Vector Machine （ LSSVM） is proposed. Use wavelet feature vectors from frequency response data of a filter system which are composed of the mean,standard deviation and entropy of wavelet coefficients to train LSSVM. The parameters of LSSVM are optimized with PSO algorithm to improve the accuracy of fault diagnosis,avoiding the blindness of parameters selection. The Elliptical Filter is used for the faults simulation experiment,the results demonstrated feasibility of the pro-posed method.