一种系统在线辨识算法的改进研究

An improvement of the system online identification algorithm

研究了线性单输入单输出系统在线辨识的递推算法.为确保复杂的智能控制有更充裕的时间,在保证辨识精度的情况下,提出了减少参数辨识运算量的变步长递推算法.传统的递推最小二乘法采用的方式是每获得一组新观测数据就修正一次参数估计值,而变步长递推算法增加了改变每次修正参数估计值前获得新观测数据的组数,合并了一些重复的运算.对该算法进行了推导,并给出了参数误差的差分方程,在理论上证明了算法的收敛性.仿真和实验表明,该算法的运算量有明显减少,而收敛速度和辨识精度几乎没降低.

This paper investigated the recursive algorithm of online identification for the linear SISO system. In order to save sufficient time for the complicated intelligent control, the authors presented a recursive algorithm to cut down the calculation of parameter estimation with guaranteeing identification accuracy. The traditional recursive weighted least squares （RWLS） estimation works in such a way that the parameters of system model are revised once a new group of data is obtained. Differently, the introduced variable step-size recursive algorithm broadens the groups of newly-observed data before revising the parameter estimates, which merges some repetitive operation in RWLS. The algorithm is deduced and difference equation of parameter error given by which the algorithm convergence is proved theoretically. At last, the computer simulation and the experiment application indicated the amount of calculation has decreased obviously in this algorithm while its convergence speed and identification precision hardly reduced.