基于K均值与WPA-RBF神经网络模拟电路故障诊断

针对模拟电路故障诊断进行了研究，提出了一种新的方法。该方法包括haar的小波分解，对数据的归一化处理，以及用K均值优化RBF的中心向量和宽度，用狼群算法优化RBF的权值。首先用haar小波对所得的电路原始故障数据集进行变换，然后对变换后的数据进行归一化处理，最终得出RBF神经网络训练所需的输入数据。针对RBF神经网络中隐层节点中心、基函数宽度及权值选取困难问题，这里用K均值优化RBF的中心向量和宽度，用狼群算法优化RBF的权值，以提高网络训练稳定性与诊断成功率。最终通过两个电路的诊断实例，来论述该方法的具体实现过程，验证用该方法进行模拟电路故障诊断的可行性。

Fault Diagnosis of Analog Circuit Based on K - means and WPA - RBF Neural Network

Aiming at the fault diagnosis of analog circuit, a new method is proposed. Including haar wavelet decomposition, the method of normalization of data processing, and using k-means optimized vector and the width of RBF center, with wolves in the weights of RBF algorithm optimization. First with haar wavelet transformed the original circuit failure data set on income, and then to transform the data after normalization processing, finally it is concluded that the input data needed for the RBF neural network training. For RBF neural network hidden layer nodes in the center and width of basis function and weight selection difficult problem, here with k-means optimized RBF center vector and the width, with wolves algorithm optimizing the weights of RBF, in order to improve the stability of the network training and diagnosis rate. Finally through two circuit diagnosis examples, to discuss the methods of concrete implementation process, verify the feasibility of using this method for analog circuit fault diagnosis.