论文部分内容阅读
本化研究了递推辨识算法的一种代数等价实现——U—D 分解算法。其基本思想就是将方差矩阵分解成三个矩阵乘的积形式:P(y)=U(y)D(t)U~T(t)其中 U(t)为单位上三角矩阵,U(t)为非负对角矩阵。并在每步递推计算时,用其因子矩阵U(t)与 D(t)的更新来代替方差矩阵 P(t)的直接递推。这样,就可以在计算量基本不变的情况下,有效地保证方差矩阵的对称性,正定性,从而获得较好的数值特性。大量的数值仿真研究表明:这种算法的数值稳定性好,计算机的舍入误对估计精度的影响较小,因此该算法的计算精度较差,以利用在有限字长的微型机上实现自适应控制时,提高系统参数实时辨识的精度。
This study has studied a recursive identification algorithm algebraically equivalent realization --U-D decomposition algorithm. The basic idea is to decompose the variance matrix into three product matrix multiplications: P (y) = U (y) D (t) U ~ T (t) where U ) Is a non-negative diagonal matrix. And replaces the direct recursion of the variance matrix P (t) with the update of its factor matrices U (t) and D (t) at each step recursion calculation. In this way, the symmetry and positivity of the variance matrix can be effectively guaranteed under the condition that the computational cost is basically the same, so as to obtain better numerical characteristics. A large number of numerical simulations show that this algorithm has good numerical stability and rounding error of the computer has little effect on the estimation accuracy. Therefore, the algorithm has poor calculation accuracy, so as to utilize the adaptive algorithm in a microcomputer with finite word length Control, improve system parameters real-time identification accuracy.