基于栈式稀疏自编码器的矿用变压器故障诊断

Fault diagnosis of mind-used transformer based on stacked sparse auto-encoder

鉴于将深度学习应用于变压器故障诊断具有良好的故障诊断效果,提出了一种基于栈式稀疏自编码器的矿用变压器故障诊断方法。通过在自编码器隐含层引入稀疏项限制构成稀疏自编码器,再将多个稀疏自编码器进行堆叠形成栈式稀疏自编码器,并以Softmax分类器作为输出层,建立矿用变压器故障诊断模型;利用大量无标签样本对模型进行无监督预训练,并通过有监督微调优化模型参数。算例分析结果表明,与栈式自编码器相比,栈式稀疏自编码器应用于矿用变压器故障诊断具有更高的准确率。

In view of application of deep learning to transformer fault diagnosis had a good fault diagnosis effect,a fault diagnosis method of mind-used transformer based on stacked sparse auto-encoder was proposed.Sparse auto-encoder is constructed by introducing sparse item constraint in hidden layer of auto-encoder,then the multiple sparse auto-encoders are stacked to form stacked sparse auto-encoder,and Softmax classifier is used as output layer to establish mine-used transformer fault diagnosis model.A large number of unlabeled samples are used to carry out unsupervised pre-training for the model,and the model parameters are optimized through supervised fine-tuning.The example analysis results show that stacked sparse auto-encoder is more accurate than stack auto-encoder in application of fault diagnosis of mind-used transformer.