PSO改进算法及其在模拟电路故障诊断中的应用

Optimized PSO Algorithm for Analog Circuit Fault Diagnosis

为了改善标准的粒子群算法在模拟电路故障诊断中存在的不足,采用了自适应变异粒子群算法来优化BP神经网络的故障诊断方法。首先对待测电路的可测点的响应信号提取故障特征,并进行小波包分解和归一化从而构建样本集;然后利用粒子群改进算法来优化BP神经网络的权值和阈值,从而实现对待测电路的训练和测试。在针对某电路的故障诊断中发现了该方法的故障诊断时间和诊断率比改进之前有了明显的改善,并且在中心偏差范围为0.3时诊断率达到了99%。

In order to improve the shortcomings of particle swarm optimization algorithm in fault diagnosis of analog circuits,a method using the adaptive particle swarm algorithm to optimize the BP neural network is presented in this paper. Firstly,extract the fault feature of response signals of the test point in the circuit,and then build sample set by using wavelet packet and normalization. Next,using the adaptive particle swarm algorithm to optimize weights and threshold of the BP neural network. Now,train and test the circuit by using the new neural network. Finally,the simulation results of a circuit show that the method is more effective in reducing the training time and improving the diagnostic rate of the model. And the diagnostic rate has reached 99% in the range of center deviation of 0. 3.