为了更精确地预测压气机叶栅流场，发展了一套基于两步集合卡尔曼滤波数据同化方法的试验数据驱动的流场预测框架。首先使用测试函数校验了集合卡尔曼滤波算法的准确性，并探讨了各超参数的选取准则，分别使用S-A和SST湍流模型对MAN GHH叶栅在设计马赫数不同攻角工况下流场进行了试验数据驱动的流场预测，预测流场与试验测量结果高度相符。结果表明：相比原始参数的预测结果，数据同化校正后的流场与试验测量结果的偏差减小了将近70%；对于多数工况校正后流场叶片吸力面尾缘分离泡尺寸明显减小，分离起始点延迟，2种湍流模型校正流场的来流边界条件、流场物理量分布具有较高的一致性，表明试验数据驱动的流场对湍流模型的依赖性较低。

To predict the cascade flow field more accurately， an experimental data driven prediction framework based on data assimilation with a two-step ensemble Kalman filter method was developed. The ensemble Kalman filtering algorithm was first verified by a test function and the selection criteria of hyperparameters were then discussed. The S-A and SST turbulence models were applied on the MAN GHH cascade under the design Mach number and different angles of attack to perform the experimental data driven flow field prediction， which shows that the predicted flow fields are highly consistent with the experimental measurement. The results indicate that compared with the predicted results under the original parameters， the errors between the flow field corrected by data assimilation and the experimental measurement results are reduced by nearly 70%； for most working conditions， the sizes of separation bubbles at the suction surface of the blade tail shrink obviously and the separation starting points move to downstream. The corrected coming boundary conditions and the physical quantities in the corrected flow field predicted by the two turbulence models are almost the same， indicating that the flow fields driven by the experimental data are nearly independent of turbulence models.