Abstract: Aiming at the problem of low inversion accuracy and time resolution in single-system GNSS-MR tide level monitoring, a multi-system fusion tide level inversion method based on LS-SVM was proposed. Using GPS, BDS, Galileo and GLONASS satellite data from Hong Kong's HKQT station for 30 consecutive days, experiments were conducted to compare three multi-system fusion methods based on sliding window least square, SVR, and LS-SVM. The results show that compared with the BDS system with the smallest RMSE for single-system GNSS-MR tide level inversion, the RMSE value of multi-system fusion tide level inversion based on LS-SVM reduced by 55.8% and the correlation coefficient increased by 4.1%. Compared with the GLONASS system with the highest time resolution, the time resolution increased by 59.3%; Compared with the multi-system fusion tide level inversion based on the SVR model, the RMSE value reduced by 52.3%, and the correlation coefficient increased by 2.2%; Compared with the multi-system fusion tide level inversion of the sliding window least square method, the RMSE value reduced by 41.1%, and the correlation coefficient is increased by 1.2%. Multi-system fusion tide level inversion based on LS-SVM has better tide level inversion performance than that of sliding window least square method and SVR algorithm.

Key words: sea level, GNSS-MR, GNSS, multipath, machine learning