摘要
针对现有神经网络谐波检测方法计算繁琐、精度不高的缺点,提出了带有预处理分类判别的神经网络谐波检测方法。首先利用谐波分类算法对谐波信号进行预处理,按照是否含有间谐波将要处理的谐波信号分为两类。对于不含有间谐波的信号,采用简化神经网络算法分析整次谐波的幅值和相位;对于含有间谐波的信号,采用自适应步长神经元分组式神经网络进行检测。针对这两类谐波信号各自的特点,充分发挥预处理分类算法的作用以及两种神经网络各自的优势,形成完整的谐波检测算法。仿真结果表明:该方法能够提升谐波检测的速度和精确度,整次谐波检测算法仅用原算法3%的迭代次数即可达到相同的误差目标,间谐波检测算法仅用原算法42%的迭代次数即可使训练精度提升10倍。
Aiming at the shortcomings of the existing neural network harmonic detection methods,such as complex calculation and low precision,a method of harmonic signal detection is presented using neutral network with preprocess and classification.By using signal classifying algorithm,the harmonic signal is first divided into two categories according to the existence of inter-harmonic signal.For signals without inter-harmonics,the simplified neural network is used to analysis the amplitude and phase of the integer harmonics.For signals containing inter-harmonic,the networks with coupled neurons and adaptive step-size is used to detect.This method aims at the feathers of the two kinds of harmonic signals and fully utilizing the advantages of preprocess and classification algorithm and two neutral networks.Therefore,the complete algorithm of harmonic detection is developed.The simulation results confirm that this method can improve the detection velocity and accuracy.The integer harmonic detection algorithm can achieve the same error with only 3%iteration times of the original algorithm.The inter-harmonic detection algorithm can improve the training accuracy by 10 times with only 42%iteration times of the original algorithm.
作者
高圣伟
许煜
亚志政
GAO Sheng-wei;XU Yu;YA Zhi-zheng(Tianjin Key Laboratory of Advanced Technology of Electrical Engineering and Energy,Tiangong University,Tianjin300387,China)
出处
《天津工业大学学报》
CAS
北大核心
2020年第5期75-80,88,共7页
Journal of Tiangong University
基金
天津市优秀科技特派员计划资助项目(16JCTPJC46600)。
关键词
谐波检测
神经网络
预处理
分类
傅里叶变换
间谐波
整次谐波
harmonic detection
neural network
preprocess
classification
Fourier transform
inter-harmonic
integer harmonic
