@[TOC](递归最小二乘估计(Recursive Least Square Estimation))
递归最小二乘估计(Recursive Least Square Estimation)
随着测量次数增加,最小二乘计算量计算量会快速增加,递归最小二乘给出了在线实时计算的递推方程。
矩阵的迹
可参考链接
导数性质
∂ T r ( A X T ) ∂ X = A \begin{aligned} \frac{\partial{Tr(AX^T)}}{\partial{X}} = A \end{aligned} ∂X∂Tr(AXT)=A
推导
x ^ k − 1 , y k \hat x_{k-1},y_k x^k−1,yk分别为k-1时的估计值及k时测量值,线性递归估计可以表示为:
y k = H k x + v k y_k=H_kx+v_k yk=Hkx+vk
x ^ k = x ^ k − 1 + K k ( y k − H k x ^ k − 1 ) \hat x_k = \hat x_{k-1}+K_k(y_k-H_k\hat x_{k-1}) x^k=x^k−1+Kk(yk−Hkx^k−1)
在已知估计量 x ^ k − 1 \hat x_{k-1} x^k−1和当前测量值 y k y_k yk基础上计算 x ^ k \hat x_{k} x^k。 K k K_k Kk为增益矩阵, ( y k − H k x ^ k − 1 ) (y_k-H_k\hat x_{k-1}) (yk−Hkx^k−1)为矫正项。如果 ( y k − H k x ^ k − 1 ) = 0 (y_k-H_k\hat x_{k-1})=0 (yk−Hkx^k−1)=0,则 x ^ k = x ^ k − 1 \hat x_k = \hat x_{k-1} x^k=x^k−1。
估计量误差均值可以计算如下:
E ( ϵ x , k ) = E ( x − x ^ ) = E ( x − x ^ k − 1 − K k ( y k − H k x ^ k − 1 ) ) = E ( ϵ x , k − 1 − K k ( H k x + v k − H k x ^ k − 1 ) ) = E ( ϵ x , k − 1 − K k H k ϵ x , k − 1 − K k v k ) = ( I − K k H k ) E ( ϵ x , k − 1 ) − K k E ( v k ) \begin{aligned} E(\epsilon_{x,k})&= E(x-\hat x) \\ &=E(x-\hat x_{k-1}-K_k(y_k-H_k\hat x_{k-1}))\\ &=E(\epsilon_{x,k-1}-K_k(H_kx+v_k -H_k\hat x_{k-1}))\\ &=E(\epsilon_{x,k-1}-K_kH_k\epsilon_{x,k-1}-K_kv_k)\\ &=(I-K_kH_k)E(\epsilon_{x,k-1})-K_k E(v_k) \end{aligned} E(ϵx,k)=E(x−x^)=E(x−x^k−1−Kk(y