预先或在线获取系统噪声的统计特性，往往是有效设计一个估计器或控制器的先决条件．早期关于噪声辨识的工作主要针对平稳或统计特性缓变的噪声过程．本文提出一种混合系统噪声辨识新算法，该算法将新息滤波与交互式多模型算法结合起来，前者降低了权概率系数对量测噪声的敏感程度，而后者则是基于混合系统模态的马尔可夫链过程实现多模型的动态交互与动态切换．仿真结果证明了本文新算法的有效性． The statistical characterization of system noise in advance or online is often a prerequisite for effective design of an estimator or controller. Earlier work on noise identification focused on noise processes that are stationary or have slow statistical properties. In this paper, a new hybrid noise identification algorithm is proposed, which combines the new filter with the interactive multi-model algorithm. The former reduces the sensitivity of the weighted probability coefficient to measurement noise while the latter is based on the mixed system modal Markov chain process to achieve multi-model dynamic interaction and dynamic switching. Simulation results prove the effectiveness of the new algorithm in this paper.