基于AdaBoost与GABP的模拟电路软故障诊断方法

针对容差模拟电路软故障诊断精度较低的问题,提出了一种基于AdaBoost与GABP的组合分类器诊断方法;首先,在Pspice中对故障模式进行Monte-Carlo分析,并利用波形有效点提取法提取故障特征,在此基础上,做归一化处理构建神经网络的原始样本；其次,利用GA算法与L-M算法组合优化BP网络构建GABP分类器；最后,利用AdaBoost算法对GABP单分类器进行迭代提升,构建AdaBoost-GABP组合分类器;诊断实例的结果表明,该方法比传统的单分类器诊断方法具有更高的诊断精度、更低的绝对误差,能够克服单分类器容易陷入局部最优,诊断结论不可信的缺陷。

Analog Circuit Soft Fault Diagnosis Method Based on AdaBoost and GABP

The accuracy of soft fault diagnosis for analog circuit with tolerance is relative low, therefore, a new method based on AdaBoost and GABP is proposed. Firstly, fault modes are simulated by Monte-Carlo method, furthermore, the effective point extraction method is used to extract the characteristic of the fault-pattern, on this basis, original samples of neural network is constructed using the normalized fault data. Secondly, GA algorithm and the L-M algorithm are used to optimize BP neural network to construct GABP classifier. Finally, the GABP network was boost by the AdaBoost algorithm to construct the AdaBoost-GABP combination classifier. The example shows that, the method proposed has higher accuracy and lower error than the traditional single classifier, beyond that, the method overcomes the defect that it is easy to fall into local optimum for the single classifier.