基于振动信号与深度学习的电力变压器故障诊断方法

Power transformer fault diagnosis method based on vibration signal and deep learning

针对当前电力变压器机械故障实时诊断准确率较低的问题,本文提出了一种基于振动信号与深度学习的电力变压器故障诊断方法。首先针对电力变压器箱体表面振动信号采用改进自适应噪声完备经验模态分解(ICEEMDAN)对其进行分解以获取重构信号,并引入模糊熵值构建振动特征向量。然后以卷积神经网络-双向门控循环单元(CNN-BiGRU)组成基础分类网络以实现特征分类,并引入高效通道注意力机制(ECAM)提升CNN学习性能。最后设计一种基于ICMIC混沌映射、自适应动态扰动和精英反向学习混合改进得到多策略协同优化秃鹰搜索(MSCOBES)算法,并将改进后的算法应用于实现CNN-BiGRU的超参数寻优,从而得到基于MSCOBES-CNN-BiGRU-ECAM的电力变压器故障诊断优化模型。在实验中对于试验变压器的机械故障进行诊断,实验结果表明本文所提出的方法对于电力变压器不同类型的机械故障的诊断准确率可达99.4%。

Aiming at the problem of low accuracy of real-time diagnosis of power transformer mechanical faults,this paper proposes a power transformer fault diagnosis method based on vibration signal and deep learning.Firstly,the vibration signal on the surface of the power transformer case is decomposed by the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(ICEEMDAN)to obtain the reconstructed signal,and the fuzzy entropy value is introduced to construct the vibration eigenvectors.Then,a Convolutional Neural Network-Bidirectional Gated Recurrent Unit(CNN-BiGRU)is used to form a basic classification network to achieve feature classification,and an Efficient Channel Attention Mechanism(ECAM)is introduced to improve the CNN learning performance.Finally,a Multi-strategy Co-optimization Bald Eagle Search(MSCOBES)algorithm is designed based on the hybrid improvement of ICMIC chaotic mapping,adaptive dynamic perturbation and elite inverse learning,and the improved algorithm is applied to realize hyper-parameter optimization of CNN-BiGRU to obtain the optimization of power transformer fault diagnosis based on MSCOBES-CNN-BiGRU-ECAM model.In the experiment for the test transformer,the experimental results show that the proposed method can reach an accuracy up to 99.4%for the power transformer with different types of mechanical fault.