东北大学学报:自然科学版 ›› 2017, Vol. 38 ›› Issue (5): 630-633.DOI: 10.12068/j.issn.1005-3026.2017.05.005

• 信息与控制 • 上一篇    下一篇

基于ELM神经网络的FAST节点位移预测研究

沙毅1, 陈曦1, 张立立1, 朱丽春2   

  1. (1. 东北大学 计算机科学与工程学院, 辽宁 沈阳110169; 2. 中国科学院 国家天文台, 北京100012)
  • 收稿日期:2015-12-14 修回日期:2015-12-14 出版日期:2017-05-15 发布日期:2017-05-11
  • 通讯作者: 沙毅
  • 作者简介:沙毅(1959-),男,江苏无锡人,东北大学副教授,博士; 朱丽春(1964-),女,北京人,中国科学院国家天文台研究员.
  • 基金资助:

    国家自然科学基金资助项目(11273001).

Research on FAST Node Displacement Prediction Based on ELM Neural Network

SHA Yi1, CHEN Xi1, ZHANG Li-li1, ZHU Li-chun2   

  1. 1.School of Computer Science & Engineering, Northeastern University, Shenyang 110169, China; 2. National Astronomical Observatories, Chinese Academy of Sciences, Beijing 100012, China.
  • Received:2015-12-14 Revised:2015-12-14 Online:2017-05-15 Published:2017-05-11
  • Contact: SHA Yi
  • About author:-
  • Supported by:

    -

摘要:

针对ELM神经网络隐含层节点数目需要人工设定,容易出现过拟合现象从而导致网络的泛化能力降低的问题,引出了基于误差最小化的ELM神经网络的改进方法EM_ELM算法,并在理论上论证了EM_ELM算法对于提高ELM神经网络预测精度和泛化能力的可行性. 随后将EM_ELM算法应用到FAST节点位移的预测模型中,并且进行了仿真验证. 仿真结果表明虽然EM_ELM神经网络在训练时间上有了一定的损失,但是仍能满足实时性的要求,而且它的预测精度和泛化能力都得到提升,证明了改进算法的有效性与可行性,进一步说明了EM_ELM神经网络更适合应用于FAST节点位移预测.

关键词: FAST节点, ELM, 神经网络, 位移预测, 可行性

Abstract:

Due to the problems that the numbers of nodes in hidden layers of ELM neural network are in need of manual setting, and the over-fitting phenomenon is easy to appear, resulting in a reduction in the network generalization, an EM_ELM algorithm was proposed to improve ELM neural network based on error minimization. The feasibility was proved in theory which could improve the prediction accuracy and generalization of ELM neural network. Meanwhile, the algorithm was also applied into the model of FAST node displacement prediction and conducted simulation finally. The results show that although EM_ELM neural network is not sufficient in training time to a certain degree, it is still proper in real-time requirement. Besides, its prediction accuracy and generalization capabilities are enhanced, which is just a proof in the effectiveness and feasibility of the improved algorithm, thereby further illustrating that the EM_ELM neural network is more suitable for FAST node displacement prediction.

Key words: FAST node, ELM, neural network, displacement prediction, feasibility

中图分类号: