Abstract This paper addressed the challenges associated with the complexity, numerous parameters, computational resource demands, and slow processing speed of transformer fault identification models based on deep learning technologies. Sparse convolutional neural network (CNN) approach is proposed for identifying faults related to dissolved gases in oil. Leveraging an improved Gramian angular field, one‐dimensional fault samples are converted into two‐dimensional feature images and data augmentation is implemented to meet the input requirements of deep learning models. Building upon visual geometry group (VGG)19 and residual networks (ResNet)50 networks for fault diagnosis, sparsity techniques are introduced through pruning, the fusion of convolution layers and batch normalization layers, and parameter quantization. Numerical experiments and performance evaluations on dissolved gas in transformer oil fault data demonstrate that the proposed method effectively reduced model complexity, minimized parameter count, conserved computational resources, and improved processing speed while maintaining a considerable level of fault identification accuracy. This made it applicable to edge computing platforms characterized by small form factors and low power consumption in the power industry.

The motion blur problem of aerial insulator images seriously affects the accuracy of insulator defect detection and recognition. Traditional convolutional neural networks treat channel and spatial information equally when processing insulator images, resulting in the network being unable to fully consider the relationship between feature channels, and allocating weights unreasonably and losing details during the image processing. Therefore, an insulator image deblurring method based on channel and spatial attention mechanism is proposed in this paper. The channel attention mechanism can select clearer parts of the image based on attention weights for different channels, while the spatial attention mechanism can better capture the detailed information in insulator images based on attention weights for different regions. This paper combines L 2 loss and edge loss as the comprehensive loss function to improve the visual quality and edge sharpness of the restored image. Through experimental analysis, it can be seen from the results that the image defuzzification method proposed in this paper has significantly improved on the index of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM).

提出了一种基于格拉姆角场和深度压缩模型的变压器故障识别方法.针对故障样本稀缺和VGG(visual geometry group)网络一般不能直接读取一维数据的问题,首先提出了格拉姆角场变换方法将一维故障样本转换为三维特征图像,并使用数据增强方法进行样本扩充,使其满足故障识别方法的输入需求.进一步,针对VGG网络层数深、参数多以及结构复杂的缺点,提出了一种改进深度压缩模型.使用NiNNet(network in network)网络的全局平均池化层替换VGG网络的全连接层,减少VGG末端网络的层数和参数规模;提出了一种结构化剪枝方法对VGG网络的多层卷积核进行剪枝,进一步减少VGG前端网络的参数规模,实现网络的深度压缩.由变压器油色谱故障数据上开展的数值实验和性能评估结果表明,所提方法在不损失变压器故障识别结果精度的前提下实现了VGG网络的深度压缩和结构简化:此外,深度压缩模型能够有效降低模型存储所需的存储空间和运行所需的计算资源,使其能够应用于体积小、功耗低的边缘计算平台.

Missing load data is a common phenomenon, which prevents these measurements from being used properly in subsequent data analysis. In order to solve the problem of missing data, this paper proposes a load data missing completion method based on the generative adversarial imputation net. Firstly, according to the characteristics of load data and the spatio-temporal relationship, a data matrix was constructed considering the influence of meteorological factors on the variation of load data. Secondly, the mask matrix is used to represent the missing data, the missing data value under the mask matrix is predicted by the generator, and the performance of the generator is evaluated by the discriminator. Finally, in order to verify the effectiveness of the proposed method, load data are used to carry out experiments. Through a series of experiments, it is verified that the completion effect of this paper has obtained good indicators on both RMSE and MAE, and can effectively complete load data with missing rate less than 50%

为了提升电力设备红外巡检图像质量,最大程度还原图像内容的有效性和准确性,提出了一种融合残差密集与生成对抗网络的红外图像超分辨率重建方法.将残差密集网络引入到WGAN(Wasserstein Generative Adversarial Networks)网络生成器,使其在训练过程中形成连续记忆机制,提高网络对图像特征的融合能力;进一步,使用谱归一化方法优化WGAN的对抗器网络参数,提升对抗训练的稳定性和效率;构造由对抗、像素、感知和纹理损失构成的综合损失函数,完成生成图像高频细节信息的重建.超分辨率实验结果表明,重建后的电力设备红外巡检超分辨率图像在峰值信噪比上提升至32.048 2 dB,在结构相似性上提升至0.921 4,且视觉效果良好,验证了所提方法能够有效提升图像质量并具备较好的工程应用价值.

Deep learning-based methods have been popular in recent years for diagnosing power equipment faults. However, in practical application, the deep learning model can not be applied to one-dimensional condition monitoring data, and the scarcity of fault data will cause the overfitting of the deep learning model, which will seriously reduce the accuracy of fault diagnosis. In this paper, a data enhancement method based on the WGAN-GP network is proposed to address the aforementioned issues. Firstly, the raw data is preprocessed, and the sensor's one-dimensional vibration signal is transformed into a two-dimensional gray picture. A network based on WGAN-GP is established to generate sample images and these sample images are similar to the original images, which realize the expansion of image samples. A fault diagnosis approach based on Convolutional Neural Network (CNN) is created on this foundation. Numerical experiments are carried out and experiments data was obtained from the Case Western Reserve University (CWRU) Bearing Data Center. The results of the experiments demonstrate that this method is capable of achieving the reasonable transformation of data structure, the reasonable expansion of fault samples and the improvement of the accuracy of fault diagnosis results.