多任务并行卷积神经网络的电压暂降分类方法

Voltage sag classification based on multi-task parallel convolutional neural network

针对电力系统中暂降事件引起的单相或多相电压在幅值和相位上的变化,提出了一种基于多任务学习的并行卷积神经网络的电压暂降分类方法.网络模型包括特征提取、特征融合和分类识别三个部分.首先针对三相电压之间的关联性,通过一维和二维卷积子网络分别捕获暂降信号在单相和三相电压上同一时刻的畸变特征.然后利用两个子网络的特征互补性将提取的特征进行融合,得到更具辨识力的特征信息.最后,采用多任务学习网络对电压暂降进行分类,同时辨别出其发生的相序、幅值和相位的变化.使用结果表明:所提方法能有效增强三相电压暂降信号的变化特征,实现电压暂降信号的精细化分类,显著提升现有暂降分类方法的准确率.

A novel voltage sag classification method based on parallel convolutional neural network with multi-task learning for the amplitude and phase changes of single-phase or multiphase voltage caused by sag events in power systems is proposed.The network includes feature extraction,feature fusion and classification recognition.Firstly,one-dimensional and two-dimensional convolution sub-network are adopted to capture the distortion characteristics of the sag samples at the same time on the single-phase and three-phase voltages respectively.Then,the extracted features are fused by using the feature complementarity of the two subnetworks to obtain more discriminative feature information.Finally,multi-task learning is introduced to classify the voltage sag and identify its phase sequence,amplitude and phase changes.The results indicate that the proposed method can effectively enhance the change features of three-phase voltage sag signals,realize the fine classification of voltage sag signals,and improve the accuracy of the existing sag classification methods.