基于WGAN-GP和高效卷积块注意力机制IPOA-ICNN的变压器故障诊断

Transformer Fault Diagnosis Based on WGAN-GP and Efficient Convolutional Block Attention Mechanism IPOA-ICNN

针对目前变压器故障诊断采集到的故障样本存在数据不平衡、特征信息提取不足的问题,提出了一种基于数据增强型和高效卷积块注意力机制(ECBAM)优化一维改进卷积神经网络(1D-ICNN)的变压器故障诊断方法。首先,建立一个基于Wasserstein梯度惩罚生成对抗网络(WGAN-GP),对不平衡的变压器数据样本进行训练以生成合成样本,用于数据增强,并采用方差分析法选取关联性强的气体特征参量;其次,使用残差和高效卷积块注意力机制模块对重构的平衡样本进行更为细节的特征提取,以实现故障诊断网络的分类;最后,利用改进的鹈鹕优化算法(IPOA)对ICNN参数进行寻优。算例对比分析表明,所提算法的故障诊断性能具备更高的精确度和稳定性,验证了所提模型故障诊断分类性能的有效性。

Aiming at the problems of unbalanced data and insufficient feature information extraction in fault samples collected by transformer fault diagnosis,a data-enhanced and efficient convolutional block attention module(ECBAM)is proposed to optimize one-dimensional improved convolutional neural network(1D-ICNN)for the transformer fault diagnosis.Firstly,a Wasserstein generative adversarial network with gradient penalty(WGAN-GP)is established to train the unbalanced transformer data samples,and synthetic samples are generated for data enhancement.The gas characteristic parameters with strong correlation are selected by variance analysis.Secondly,the residual and efficient convolutional block attention mechanism modules are used to extract more detailed features from the reconstructed balanced samples to realize the classification of fault diagnosis networks.The improved pelican optimization algorithm(IPOA)is used to optimize the ICNN parameters.The comparison and analysis of examples show that the proposed algorithm has higher accuracy and stability in fault diagnosis performance,and the effectiveness of the fault diagnosis classification performance of the proposed model is verified.