基于改进残差网络的输电线路雷击过电压识别研究

Research on Identification of Lightning Overvoltage in Transmission Line by Improved Residual Network

为进一步提升高压输电线路在运行与维护过程中的过电压类型识别准确率,本文提出一种改进残差网络(residual network,ResNet)的输电线路雷击过电压识别方法。对传统ResNet-50网络的改进包含以下3个方面:首先,将7×7卷积核替换为3个3×3卷积核,以提高网络的特征提取能力;然后,将传统ResNet-50网络的ReLU激活函数替换为ReLU-Softplus激活函数,提高网络的收敛速度;接着,调整ResNet-50网络结构,以强化批标准化的作用。最后,结合迁移学习和改进残差网络构建输电线路雷击过电压识别模型。实验结果表明,改进残差网络在第70步收敛,识别精度为97.25%,识别准确率和收敛速度均优于其他模型。

To further improve the accuracy of over-voltage type identification during the operation and maintenance of high-voltage transmission lines,this paper proposes an improved Residual Network(ResNet)method for lightning over-voltage identification of transmission lines.The improvement of the traditional ResNet-50 network includes the following three aspects.Firstly,the 7×7 convolution kernel is replaced by three 3×3 convolution kernels to improve the feature extraction ability of the network.Secondly,the ReLU activation function of the traditional ResNet-50 network is replaced by the ReLU-Softplus activation function to improve the convergence speed of the network.Then,adjust the ResNet-50 network structure to enhance the role of batch standardization.Finally,a lightning overvoltage identification model of transmission lines is constructed by combining transfer learning and improved residual network.The experimental results show that the improved residual network converges in the 70th step and the recognition accuracy is 97.25%.The recognition accuracy and convergence speed are better than other models.