Abstract：According to the non-stationary and nonlinear characteristics of GNSS time series, we analyze the applicability and characteristics of XGBoost and Prophet models, and construct a Prophet-XGBoost prediction model. Firstly, using the Prophet model, we decompose the GNSS original time series, then carry out partial prediction by XGBoost model. We obtain prediction results by equal weight addition. We select the daily coordinate time series data of U component of ALGO, ALRT and BRST IGS stations in the experiment, and use MAE and RMSE as evaluation indexes. The experimental results show that compared with the single XGBoost model and Prophet model, the MAE and RMSE values of Prophet-XGBoost model are optimized to a certain extent. The effectiveness of this method is verified and can be used for GNSS time series prediction.
