基于BP神经网络的多谐波源同次谐波叠加模型  

New Model of Same-Order Harmonic Superposition in Multiple Harmonic Sources Based on BP Neural Network

作　　者：曹栋[1] 扈罗全[1,2] CAO Dong;HU Luoquan(School of Rail Transportation,Suzhou University,Suzhou 215021,China;Suzhou Entry Exit Inspection and Quarantine Bureau,Suzhou 215104,China)

机构地区：[1]苏州大学城市轨道交通学院,江苏苏州215021 [2]苏州出入境检验检疫局,江苏苏州215104

出　　处：《测试技术学报》2018年第5期381-385,共5页Journal of Test and Measurement Technology

基　　金：国家质检总局科研资助项目(2014IK192);苏州出入境检验检疫局科研资助项目(SISC201604)

摘　　要：针对多谐波源系统中同次谐波叠加问题,IEC标准TR 61000-3-6:2008推荐的模型为非线性计算公式,工程实践时计算结果与实际值误差较大.基于BP神经网络强大的非线性映射能力,提出构造一个包含一层隐含层的BP神经网络.利用多个同型号吸尘器样品,模拟工频单相网络中的多谐波源系统,获得实测数据作为训练样本,经过训练后得到神经网络模型.将部分实测数据分别输入到该BP神经网络模型与IEC标准公式,计算两个模型数据与实际数据之间的误差,并进行误差分析.结果显示:神经网络算法的精度更高,表明本文提出的运用神经网络算法求解多谐波叠加问题的有效性和可行性.In order to solve the problem of the same-order harmonic superposition in the multiple harmonic sources,a non linear formula was proposed in the IEC standard TR 61000-3-6：200,which presented significant errors in engineering practice.Considering the powerful nonlinear mapping of the BP neural network,a BP neural network containing a hidden layer was constructed.Use multiple vacuum cleaner samples of the same model could simulate the multi-harmonic source system in a power frequency single phase network.The measured data was selected as a training sample,and a BP neural network model was obtained after training procedure.The errors of BP neural network model and IEC standard formula were calculated to compare with practical data when some of the measured data are input to them,and perform error analysis.The results showed that BP neural network is more accurate,it proves the proposed neural network algorithm is effective and feasible for solving multiple harmonic source systems.

关 键 词：BP神经网络 多谐波源 谐波叠加 

分 类 号：TM711[电气工程—电力系统及自动化]