Abstract: Extreme learning machine has quick learning speed and high accuracy. To improve the generalization performance, differential evolution algorithm, which has the features of global convergence and easy computation, is introduced in the parameter optimization of extreme learning machine. A parameter optimization model based differential evolution algorithm is established in order to combine the advantages of the algorithms, applied in analog circuit fault diagnosis. Firstly, the output voltage signals from the test nodes of all analog circuit are obtained and the fault feature vectors are extracted from principal component analysis. Then, using differential evolution algorithm global optimization ability encodes the connection weights and thresholds to get the optimal structure, achieving good generalization performance and robustness. This approach applied in diagnosis is more effective and can get satisfied results in a short time.

Key words: differential evolution, extreme learning machine, analog circuit, fault diagnosis

摘要： 极限学习机具有学习速度快、精度高的优点。为了进一步提高泛化能力，将差分进化算法的全局寻优和算法简单的特点引入到极限学习机的参数优化中，建立了基于差分进化算法优化极限学习机的模型，使两种算法的优点有机结合，应用于模拟电路故障诊断中。首先利用主元分析对电路采样信号进行处理，提取故障特征；其次利用差分进化算法的全局寻优能力优化极限学习机网络的权值和阈值，具有很好的泛化能力。此方法应用于电路仿真实例中，能在较短的时间内获得满意的结果。

关键词: 差分进化, 极限学习机, 模拟电路, 故障诊断