基于多尺度广义S变换和深度残差网络的雷击跳闸故障类型识别方法

Lightning Tripping Fault Type Identification Method Based on Multi-Scale Generalized S-Transform and Deep Residual Network

为提高雷击跳闸类型判断的效率和智能化水平,提出一种基于多尺度广义S变换和深度残差网络的雷击跳闸故障识别方法。首先采用广义S变换对雷击跳闸暂态信号进行时频变换,通过改变广义S变换中的调节因子λ获得多尺度时频分布图像,然后将变换得到的多尺度时频分布特征图作为特征图像输入到深度残差网络中,进行特征提取和识别模型建立,最终识别出雷击跳闸故障类型。以124个输电线路实测雷击跳闸暂态电压信号样本作为训练集和测试集输入到识别模型中,并与其他方法进行对比,验证提出方法的有效性。结果表明,采用多尺度广义S变换和深度残差网络构成的雷击跳闸故障识别模型识别反击跳闸和绕击跳闸的准确率分别为96.67%和95.94%。

In order to improve the efficiency and intelligence level of lightning tripping type judgment, a lightning tripping fault identification method based on multi-scale generalized S-transform and deep residual network is proposed. Firstly, the generalized S transform is used to time-frequency transform the lightning trip-out transient signal, through changing the regulating factor lambda in the generalized S transform multi-scale time-frequency distribution image. Then transform the multi-scale time-frequency distribution graph as the input characteristics of the image to the depth of the residual network for extracting feature and establishing recognition model. Finally, the lightning trip-out fault type can be identified. 124 measured lightning tripping transient voltage signal samples of transmission lines were put into the identification model as training set and test set, and compared with other methods to verify the effectiveness of the proposed method. The results show that the recognition accuracy of the lightning tripping fault model based on the multi-scale generalized S-trans form and deep residual network is 96.67% and 95.94%, respectively.