【摘要】 为提高广泛应用的电力电子电路故障的诊断准确率与速度，提出一种基于小波包变换和改进粒子群算法的电力电子电路故障诊断方法。首先，利用小波包对故障信号进行分解与重构得到小波包系数，应用Fisher准则降维并进行归一化处理，得到优化的故障特征向量；然后，采用具有扭曲粒子位置的措施和增加动态惯性权重系数的改进粒子群算法求取各类故障的故障特征中心，通过计算测试样本与故障特征中心的欧氏距离实现对故障的分类诊断；最后，通过典型电力电子电路仿真实验对所提出的方法进行实验验证。实验结果表明，文中选用改进后的粒子群算法进行故障诊断时，与小波包-BP神经网络和小波包-极限学习机比较，准确率分别提高了3.52%和6.3%，诊断所需时间分别减少2.4 s和3.5 s。更多还原

【Abstract】 In order to improved the accuracy and speed of fault diagnosis of power electronic circuits,a fault diagnostic approach based on wavelet packet transform and improved particle swarm optimization is proposed. The wavelet packet is utilized to obtain the wavelet packet coefficients by decomposition and reconstruction of fault signals,and then the optimized fault feature vectors are created by using the Fisher′ s criterion for dimensionality reduction and normalization processing. An improved particle swarm optimization method possessing the distorted particle position measures and enhancing the dynamic inertial weight coefficients is employed to calculate the fault feature center of each type of fault,and then Euclidean distance between test sample and fault feature center is calculated to realize the fault classification. Finally,a typical power electronic circuit simulation experiment is used to validate the proposed method. The experimental results show that,compared with the wavelet packet-BP neural network and the wavelet packet-extreme learning machine,the accuracy of the improved particle swarm optimization algorithm for fault diagnosis is increased by 3.52% and 6.3%,the time required for diagnosis is reduced by2.4 s and 3.5 s,respectively.更多还原