In the process of ensemble data assimilation， due to the false correlation between the remote observation and the assimilation state， which affects the performance of DA， more attention has been paid to the localization methods. In addition， because of the limited ensemble size， it is easy to cause phenomena such as under-sampling and underestimation of covariance， which makes the filtering effect divergent. The fuzzy control algorithm is proposed， which is mainly used to judge the distance between the observation point and the state update point to assign the corresponding observation weight to the observation point， and then adjust the localization coefficient to update the background error covariance and observation error covariance， respectively. Thus， an effective state estimation is obtained. Based on BL and RL method， coupled with fuzzy control， the Background Covariance Fuzzy （BCF） and Fuzzy Observation Covariance （FOC） were proposed. We conducted an experiment on the Lorenz-96 model， and the BCF and the FOC method exhibited better assimilation effect with the small ensemble size and localization radius. By analyzing the Taylor diagram， it was found that the new algorithms had a high correlation with the observation point and small spatial variability. Finally， the robustness of BCF and FOC algorithms was further verified under the different dimensional fuzzy controller. It will provide a good research platform for data assimilation error processing in the future.