基于T-DSCNN的故障选线方法研究  

T-DSCNN-based Faulty Line Determination

作　　者：鲁玉海 LU Yuhai(China Electricity Construction Group,Henan Electric Power Survey and Design Institute Co.,Ltd.,Zhengzhou 450000,China)

机构地区：[1]中国电建集团河南省电力勘测设计院有限公司,河南郑州450000

出　　处：《电工技术》2024年第22期193-196,共4页Electric Engineering

摘　　要：研究了一种基于迁移深度可分离卷积神经网络(T-DSCNN)的故障选线方法,旨在提高电力系统中故障选线的准确性和效率。通过引入迁移学习的概念,T-DSCNN能够利用预训练的模型参数作为初始权重,加速模型的训练过程并提高其泛化能力。深度可分离卷积技术的应用减少了模型的参数量,降低了计算复杂度,从而使得模型在保持高准确率的同时更适用于实时故障选线的应用场景。在基于标准数据集的故障选线测试中,T-DSCNN表现出了优异的性能,识别速度和准确率高于传统卷积神经网络和其他故障选线方法。In this paper,a faulty line determination method based on transfer-depthwise separable convolution neural networks(T-DSCNN)is studied.The method aims to improve the accuracy and efficiency of faulty line determination in power systems.By introducing the concept of migration learning,the T-DSCNN is able to accelerate the training process of the model and improve its generalisation ability by using the pre-trained model parameters as initial weights.Moreover,the use of deep separable convolution technique reduces the number of parameters of the model and lowers the computational complexity,thus making the model more suitable for the scenario of real-time determining faulty line while maintaining high accuracy.Through testing on standard datasets,T-DSCNN shows excellent performance on the fault routing task,which significantly improves the recognition speed and accuracy compared to conventional convolutional neural networks and other methods.

关 键 词：T-DSCNN 故障选线 迁移学习 融合图像 

分 类 号：TM862[电气工程—高电压与绝缘技术]