基于优化广义S变换和混合输入神经网络的电能质量扰动识别  

Power Quality Disturbance Identification Based on Optimized Generalized S⁃transform and Hybrid Input Neural Network

作　　者：刘海涛 武祥 张淑清[1] 刘大鹏 刘勇 穆勇 LIU Haitao;WU Xiang;ZHANG Shuqing;LIU Dapeng;LIU Yong;MU Yong(Institute of Electrical Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;College of Electrical Engineering,North China University of Science and Technology,Tangshan,Hebei 063210,China;Tangshan Power Supply Company,North Hebei Electric Power Co.Ltd,Tangshan,Hebei 063000,China)

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004 [2]华北理工大学电气工程学院,河北唐山063210 [3]国网冀北电力有限公司唐山供电公司,河北唐山063000

出　　处：《计量学报》2025年第1期53-61,共9页Acta Metrologica Sinica

基　　金：国家自然科学基金(52275067);河北省自然科学基金重点项目(F2020203058);国网冀北电力有限公司唐山供电公司项目(SGJBTS00FZJS2311425)。

摘　　要：利用广义S变换时频矩阵中时间最大幅值曲线和频率最大幅值曲线与电能质量信号幅值和频谱包络线的相关性,提出优化广义S变换的方法对高斯窗函数参数进行自适应选取,充分保留了电能质量扰动的幅值和频率特征。然后提出一种混合输入神经网络框架,分别对原始时间序列和优化广义S变换得到的时频矩阵进行自动特征提取,最后将2种输入提取到的特征进行组合并利用全连接层来识别扰动类型。通过对包含26种电能质量扰动类型的仿真数据集进行训练和验证,结果表明所述方法对扰动识别准确率为99.77%;然后对2种实际电网扰动信号进行测试,对扰动识别准确率仍然能达到92.5%,高于传统单一输入神经网络。Based on the correlation between the time maximum value curve and frequency maximum value curve of generalized S-transform time-frequency matrix and the amplitude and spectrum envelope of power quality signal,a method of optimizing generalized S-transform is proposed to select the parameters of Gaussian window function adaptively,which fully preserves the amplitude and frequency characteristics of power quality disturbance.Then,a hybrid input neural network framework is proposed to automatically extract the features of the original time series and the time-frequency matrix obtained from the optimized generalized S-transform.Finally,the features extracted from the two inputs are combined and the disturbance types are identified by the fully connected layer.Through the training and verification of the simulation data set containing 26 types of power quality disturbance,the results show that the disturbance recognition accuracy of the proposed method is 99.77%,and then the two kinds of actual grid disturbance signals are tested,and the disturbance recognition accuracy can still reach 92.5%,which is higher than the traditional single input neural network.

关 键 词：电学计量 电能质量 扰动识别 S变换 卷积神经网络 混合输入 

分 类 号：TB971[一般工业技术—计量学]