增广最小二乘的递推算法对应的噪声模型为滑动平均噪声，扩充了参数向量和数据向量H（k）的维数，把噪声模型的辨识同时考虑进去。最小二乘法只能获得过程模型的参数估计，而增广最小二乘法同时又能获得噪声模型的参数估计，若噪声模型为平均滑动模型，，则只能用RELS算法才能获得无偏估计。当数据长度较大时，辨识精度低于极大似然法。-Augmented least squares of recursion algorithm corresponding noise model for moving average noise, expanded parameter vector and data vector H (k) of the dimension, the noise model identification and at the same time into consideration. Least square method can only obtain process model parameter estimation, and augmented least squares method at the same time can obtain noise model parameter estimation, if noise model for the average sliding model,, it can only use RELS algorithm to obtain unbiased estimation. When the data length is bigger, the identification accuracy is lower than maximum likelihood method.