Abstract: In order to obtain unbiased estimates in the presence of colored noise, a two-stage recursive Bayesian identification algorithm is proposed based on auxiliary model principle and decomposing technique. In this algorithm, the original model is decomposed into two fictional sub-models firstly, and then identified respectively; the estimated noise variance and a new covariance resetting method are also integrated into the algorithm to obtain improved estimates. Compared with recursive least squares algorithm, the proposed algorithm can reduce the computational burden. According to the simulation, the estimation error of the proposed algorithms is smaller than that of the recursive least squares. An industrial application validates the proposed algorithm.

Key words: two-stage recursive algorithm, recursive Bayesian algorithm, the least squares algorithm, covariance resetting

摘要： 为了在有色噪声干扰情况下获得无偏估计，基于辅助模型思想和分解技术，提出了一种带协方差重置的两阶段递推贝叶斯辨识算法。该算法首先把待辨识模型分解成两个虚拟子模型，然后分别辨识；同时，把估计到的噪声方差引入算法，并加入了一种新的协方差重置方法。计算量分析表明，与带协方差重置的最小二乘算法相比，所提算法可以减少计算量。仿真结果显示，所提算法的估计误差比传统最小二乘算法要小。实例建模证明了算法的有效性。

关键词: 两阶段递推算法, 递推贝叶斯算法, 最小二乘算法, 协方差重置