基于TDFFCNN模型的电能质量扰动分类  

Power quality disturbance classification based on TDFFCNN model

作　　者：胡杰 程志友[1,2,3] 姜帅 李悦 汪德胜 HU Jie;CHENG Zhiyou;JIANG Shuai;LI Yue;WANG Desheng(School of Electronice and Information Engineering,Anhui University,Hefei 230601,China;Power Quality Engineering Research Center,Ministry of Education,Anhui University,Hefei 230601,China;School of Internet,Anhui University,Hefei 230039,China)

机构地区：[1]安徽大学电子信息工程学院,安徽合肥230601 [2]安徽大学教育部电能质量工程研究中心,安徽合肥230601 [3]安徽大学互联网学院,安徽合肥230039

出　　处：《安徽大学学报（自然科学版）》2023年第5期58-64,共7页Journal of Anhui University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61672032);安徽省科技重大专项(18030901018)。

摘　　要：为提高电能质量扰动(power quality disturbances,简称PQD)分类准确率,提出双模态特征融合的卷积神经网络(two-dimensional feature fusion convolutional neural network,简称TDFFCNN)模型,该模型包含2个轻量化子模型.子模型1运用全局最大池化,在大量降低数据量的同时学习电能质量扰动(power quality disturbances,简称PQD)时间序列的时序特征,以增加模型提取时序特征的能力;子模型2使用小卷积核与小步长为参数的堆叠单元提取图像振幅特征,对高中低层特征进行融合,以获得细节与结构性语义.实验结果表明:TDFFCNN模型具有较强的抗噪性能;相对于其他3种模型,TDFFCNN模型的平均准确率最高.因此,TDFFCNN模型具有更强分类性能.In order to improve classification accuracy of power quality disturbances(PQD),a two-dimensional feature fusion convolutional neural network(TDFFCNN)was proposed,which includes two lightweight submodels.Sub-model 1 used global maximum pooling to learn time sequence features of PQD time series while greatly reducing the amount of data,so as to increase the ability of the model to extract time sequence features.Sub-model 2 used small convolution kernel and stacked units with small step length as parameters to extract image amplitude features,and fused high and low level features to obtain details and structural semantics.The experimental results showed that TDFFCNN model had stronger anti-noise performance.Compared with the other four models,TDFFCNN model had the highest average accuracy.Therefore,TDFFCNN model had stronger classification performance.

关 键 词：电能质量 特征融合 全局最大池化 振幅特征 

分 类 号：TM712[电气工程—电力系统及自动化]