【摘要】 提出一种基于小波包分解、主元分析、归一化处理、粒子群算法与神经网络相结合的模拟电路故障诊断新方法。该方法使用小波包分解来对信号进行消噪和小波多尺度分解,再进行正交主元分析和归一化处理来提取故障特征信息,作为神经网络的输入样本。在充分考虑传统BP算法中采用梯度下降法所固有的极易陷入局部极小等缺陷的基础上,提出了采用粒子群算法来优化传统BP网络的方法。文中研究了故障特征信息的提取、样本选择及诊断系统,并通过电路诊断实例,阐述了该方法的具体实现,验证了所提方法的有效性。更多还原

【Abstract】 Based on wavelet packet decomposition, principal component analysis (PCA), normalization, particle swarm optimization (PSO) and neural networks (NNs), a new analog circuit diagnosis method is proposed. The proposed method is based on wavelet packet transformation that using the wavelet packet decomposition as a de-noise tool, the feature information is extracted by wavelet de-noising, multi-resolution, orthogonalization and normalization. The input patterns are satisfied when the feature information applies to the neural networks. Under considering the characteristics of the traditional back-propagation (BP) algorithm, the PSO algorithm is used to substitute the gradient descent method in tradition BP neural network. Finally, the realization of the proposed strategy is expounded by using practical circuits. The simulation and experimental results demonstrate the effectiveness of the proposed method.更多还原