Abstract: Based on the in-situ monitoring data of the dam, back analysis of the material parameters is an effective way to obtain the real parameters. Aiming at the problem of multi-material and multi-parameter back analysis in the NHRI model parameter back analysis of concrete faced rockfill dam, a parameter back analysis method based on XGBoost was proposed. We first conducted parameter sensitivity analysis through orthogonal test, simplified the parameters to be inverted, and further established a machine learning sample set through orthogonal test. On the basis of comparing the training results of different machine learning models, XGBoost was proposed to use dam deformation monitoring data to inverse the NHRI model parameters of dam rockfill. The deformation results calculated by inversion parameters were compared with the actual monitoring results. The results show that XGBoost algorithm has more advantages than other algorithms such as decision tree algorithm. The combination of orthogonal test can effectively reduce the number of finite element calculations and improve the calculation efficiency and accuracy. The calculated results of inversion parameters are in good agreement with the actual monitoring results.