摘要：为使伺服系统始终具备优良的动态响应，有必要对系统等效转动惯量进行辨识。含遗忘因子的递推最小二乘惯量辨识方法对存储字长要求高，多次递推计算会降低协方差矩阵的正定性，导致辨识失真甚至发散。仿照FFRLS，同样可以推导出含遗忘因子的递推平方根改进算法，该方法降低了FFRLS对字长的要求，更适合应用于字长有限的嵌入式系统。仿真和实验结果验证了FFRSR辨识方法稳定性较高，辨识效果明显优于FFRLS。

关键词：伺服；惯量辨识；递推最小二乘；递推平方根；遗忘因子
Abstract: It is necessary to estimate equivalent moment of inertia in order to obtain well performance of the servo packs. The inertia identification based on Forgetting Factor Recursive Least Square (FFRLS) demands storage of long word length. Multiple recursive calculation will reduce the positive definiteness of the covariance matrix, resulting in identification of distortion or even divergence. Like FFRLS, an improved online identification of inertia based on Forgetting Factor Recursive Square Root (FFRSR) was derived. The proposed method makes it quite feasible in embedded processors with finite word length. The better results in simulations and experiments validated the obvious advantage of stability over FFRLS.

Key words: servo；inertia identification；recursive least square；recursive square root；forgetting factor