摘要
针对BP网络在处理单一时序变形数据时,网络输入节点数目难以确定的不足,提出利用G-P算法计算单时序变形数据的关联维数,把饱和关联维数Dh所对应的嵌入维数m作为最佳建模维数,并借助重构理论进行相空间重构,得到等价的相空间,为BP网络提供合理的输入。理论上该模型既能考虑到影响变形时间序列的动力因子,又能解决网络的输入节点数难以确定的问题。实验证实,该模型应用于变形预测中能取得精度较高的预测结果。
It is difficult for BP network to determine the number of input layer＇ s nodes as dealing with the deformation data of single time series. In order to solve this problem,the G-P algorithm is used to calculate the correlation dimension. Taking the embedding dimension m as the best model dimension and reconstructing an equivalent phase-space according to the theory of reconstruction so as to provide a reasonable input for BP network. Considering the influence of dynamical factor of deformation as well as the difficulty of calculating number of input cell,the model can be applied to simulate and forecast the deformation. Two actual examples show that this model is reasonable and precise.
出处
《大地测量与地球动力学》
CSCD
北大核心
2009年第6期113-117,共5页
Journal of Geodesy and Geodynamics
基金
教育部博士点基金(新教师)(200802901516)
教育部博士点基金(200802900501)
中澳科技合作特别基金
关键词
BP网络
最佳维数
G—P算法
相空间重构
形变预测
BP neural networks
optimal dimension
G-P algorithm
chaos phase-space
deformation monitoring
