一种基于多层前向神经网络的谐波检测方法.pdfVIP

26 卷 18 期 中 国 电 机 工 程 学 报 Vol.26 No.18 Sep. 2006 2006 年9 月 Proceedings of the CSEE 2006 Chin.Soc.for Elec.Eng. 文章编号：0258-8013 (2006) 18-0090-05 中图分类号：TM93 文献标识码：A 学科分类号：47040 一种基于多层前向神经网络的谐波检测方法 汤胜清，程小华 (华南理工大学电力学院，广东省 广州市 510640) A Harmonic Measuring Approach Based on Multilayered Feed Forward Neural Network TANG Sheng-qing, CHENG Xiao-hua (South China University of Technology, Guangzhou 510640, Guangdong Province, China) ABSTRACT: Magnitude and phase angle of harmonics are two 0 引言 key measuring parameters in active power filter , both of them 自有源电力滤波技术出现以来，各国学者提出 can be non-linearly mapped by ANN. A method is proposed to 许多针对有源滤波器的谐波检测技术，其中，得到 simultaneously measure the two parameters. A three layers feed forward neural network whose hidden layer adopts tangent 广泛应用的是20 世纪80 年代初H.Akagi 提出的基 function and the output layer adopts linear function is 于瞬时无功功率理论的谐波检测技术，它将有源滤 constructed to simulate the method. The training method and 波器从实验研究推向了应用实践。然而随着对有源 training sample composition in the neural network are presented 滤波器的控制要求越来越高，基于瞬时无功功率理 using 3rd harmonic as an example. Training the neutral network 论的控制方法的不足逐渐被人们所认识，因此，很 with training samples first, then measuring untrained samples, 多学者正试图寻找更有效的检测手段。人工神经网 the simulation is performed by using Matlab’s toolbox and the 络(artificial neural network ，ANN)具有分布并行处 results show the validity of the proposed harmonic measuring 理、非线性映射、自适应学习和鲁棒容错等特性，