Abstract: When Kalman filter algorithm is applied in engineering, if there are variously ?inaccurate reasons in the model, the estimated signals can generate large state estimation error, even filtering divergence, which seriously influences the denoising effect of the filter in the signal process. In order to suppress filtering divergence and improve the filtering accuracy, the paper expounds the phenomenon of divergence, analyzes the main causes of Kalman filtering divergence, namely the model error and calculation error, and summarizes the four types of commonly used method of suppressing filtering divergence, respectively to adjusting gain method, prediction error covariance weighting method, limited memory method and adaptive Kalman algorithm. Simulating and analyzing these methods confirm that they all have their own advantages and diaadvantages. Adjust gain method requires the least amount of calculation, adaptive Kalman algorithm has the highest calculation accuracy and weighting method achieves a good balance between the less amount of calculation and high precision requirment. Limited memory method’s filtering effect is equal to weighting method’s, but calculation of limited memory method is somewhat complex. It has practical guiding significance on how to restrain filtering divergence in the engineering applications.

Key words: Kalman Filter（KF）, model error, filtering divergence, suppress divergence

摘要： 在Kalman滤波算法的工程应用中，常常由于各种原因导致较大的状态估计误差，甚至造成滤波发散。滤波发散严重影响着滤波器在信号处理过程中的去噪作用。为了抑制滤波发散并提高滤波精度，对此现象进行了简单阐述，分析了Kalman滤波产生发散的主要原因，即模型误差和计算误差，总结出四类常用的抑制滤波发散的方法，分别为调节增益法、预测误差协方差加权法、限定记忆法、自适应Kalman算法，并对各个抑制发散的方法进行了仿真对比分析，证实其各有优劣。其中，调节增益法所需的计算量最少，自适应Kalman算法的计算精度最高，而预测误差协方差加权法则使得计算量少和计算精度高的要求达到了较好的平衡；限定记忆法的滤波效果与加权法相当，但计算略复杂。这对在工程实践中如何抑制滤波发散有实际指导意义。

关键词: 卡尔曼滤波（KF）, 模型误差, 滤波发散, 抑制发散