基于多特征的APSO-SVR的模拟电路故障预测

Analog circuit fault prognostic based on multi-feature and SVR optimized by APSO

针对模拟电路故障特征提取困难和难以准确预测剩余寿命等问题,提出了多特征向量提取和自适应粒子群(Adaptive Particle Swarm Optimization,APSO)算法优化支持向量机回归(Support Vector Regression,SVR).首先,结合统计特征与小波包能量特征构造多特征融合的特征向量;然后,计算特征向量间的欧氏距离来量化模拟电路中元件的退化状态,并由此得到模拟电路的故障阈值;最后,利用APSO优化SVR构造故障预测模型进行预测,并用基准电路进行仿真实验验证方法的实用性.仿真实验结果表明新方法对模拟电路故障预测有更高的准确度.

A prognostic model based on multi-feature vector and Adaptive Particle Swarm Optimization(APSO)optimized Support Vector Regression(SVR)is proposed to solve the difficulties in extracting fault features of analog circuits and predicting the remaining life accurately.Firstly,a multi-feature vector is constructed by combining the statistical feature and wavelet packet energy feature.Then calculating the Euclidean distance of feature vector between the initial and the degradation process to quantify the degradation state of the components in the analog circuit,the fault threshold is obtained accordingly.Finally,the fault prognostic model of SVR optimized by APSO is used for analog circuit fault prognostic,and a reference circuit is used to perform simulation experiments to verify the practicability of the method.The simulation results show that this method has a higher accuracy.