计算机工程与应用 ›› 2011, Vol. 47 ›› Issue (6): 228-230.

• 工程与应用 • 上一篇    下一篇

用于电磁兼容预测的自适应泛化回归神经网络

张煜东1,吴乐南1,陈书文1,2   

  1. 1.东南大学 信息科学与工程学院,南京 210096
    2.江苏省辐射环境保护咨询中心,南京 210096
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2011-02-21 发布日期:2011-02-21

Adaptive generalized regression neural network used for prediction of EMC

ZHANG Yudong1,WU Lenan1,CHEN Shuwen1,2   

  1. 1.School of Information Science & Engineering,Southeast University,Nanjing 210096,China
    2.Radiation Environmental Protection Consultation Center of Jiangsu Province,Nanjing 210096,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-02-21 Published:2011-02-21

摘要: 为了更好地对电磁兼容进行预测,提出一种自适应泛化回归神经网络(AGRNN),与传统泛化回归神经网络(GRNN)区别在于:将光滑因子设为最小数据距离的1/2,将偏置设为光滑因子的倒数。对简单一维数据的测试表明,无论数据如何分布,AGRNN的拟合曲线均较GRNN更加接近样本点、且更平滑。以平行线间电磁耦合干扰为具体算例,证明AGRNN对训练数据与测试数据的预测优于改进BP算法,且网络不需要训练。

关键词: 电磁兼容预测, 泛化回归, 神经网络

Abstract: To predict the Electromagnetic Compatibility(EMC) more effectively,an Adaptive Generalized Regression Neural Network(AGRNN) is proposed.It sets the smooth factor as half of the minimum data distance,and sets the bias as the reciprocal of the smooth factor.Test on simple one dimensional data shows the fitting curve of AGRNN is smoother and more adjacent to sample points no matter how data distribute.The specific example on electromagnetic coupling interference between two parallel wires demonstrates the median square error of the prediction by AGRNN is less than that of the improved BP,and the network needs no training.Thus,the proposed algorithm is valid and effective.

Key words: electromagnetic compatibility prediction, generalized regression, neural network