摘要
讨论了用连续Ｈｏｐｆｉｅｌｄ神经网络实现最小二乘（ＬＳ）法同时求损失函数及辨识参数，进而对该网络的一次输入的矩阵参数进行多阶次辨识的计算。通过增加一维输入矩阵使其构成计算所需要的结构，及根据最小二乘法结构上的特点对该网络的反馈信息加以控制的手段，使该网络在稳定时，能够同时输出辨识参数及损失函数。依此顺序对反馈信号进行控制。
This objective focuses on a realization way by which lost functions and parameters of the least square(LS) method can be computed and the multi order identification by using continuous Hopfield network can be processed. Based on the LS construction characters, through controlling the network feedback signals and enlarging one dimension of the network input matrix, this method can make the network output LS parameters and lost function simultaneously. In the same way, the parameters and lost functions for each order which is lower than the input one can also be obtained.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
1997年第6期1-4,共4页
Journal of Tsinghua University(Science and Technology)
基金
清华大学基金
