栈式降噪自编码网络在变压器故障诊断中的应用

为提高变压器故障诊断的准确率,提出了一种新型的变压器故障诊断网络,该网络以基于栈式降噪自编码网络为基础,把深度学习用在诊断变压器设备故障方面,建立深层网络模型,采取逐层贪婪编码的方式进行自适应的非监督式预训练,实现高维深层故障特征的自适应提取和挖掘,进而使用反向传播算法对模型进行监督式微调。最后利用Softmax分类器,对故障进行分类输出。最后通过实例验证表明,提出的栈式降噪自编码网络能准确、有效地对变压器进行故障诊断,与传统方法相比,该方法提高了变压器故障诊断的准确率。

Application of stacked denoising auto-encoder network in fault diagnosis of power transformer

The three-ratio method and Rogers method, which are widely used in the research of transformer fault diagnosis, have the disadvantage of incomplete coding and too absolute judgment criteria.To solve this problem,applied the deep neural network to fault diagnosis of power transformer,and proposed a fault diagnosis method based on the stacked denoising auto-encoder.A deep network model is established, and a self-adaptive and unsupervised pre-training is implemented by greedy coding. To achieve adaptive extraction and mining of high-dimensional deep fault features, the back-propagation algorithm is used to fine tune the model.Finally, the Softmax classifier is used to classify the faults.The experimental results show that compared with the traditional Back propagation (BP) neural network, the proposed deep neural network can solve the problem of easy to fall into the local minimization and slow convergence of the traditional BP neural network, and diagnosis the transformer faultmore effectively.