This paper proposed the optical weighting combined mode of Least Square Support Vector Machine (LS-SVM) and BP Neural network. According to the measured data, this paper compared and analyzed the accuracy of LS-SVM model, BP Neural network model; quadratic polynomial curve surface fitting based on Total least-square algorithm and optimal weighting combined model, the data shows that the optimal weighting combined model has higher precision then others.

The common method to determine Quasi-Geoids is GPS leveling however the Quasi-Geoid of this method determined is a kind of trend surface which not take the physical property of geoid into consideration, and the fitting method is surface fitting which only consider the surveying error, lead to inaccurate fitting result. In allusion to these problems, Remove-restore method is used to remove the long wave information of earth gravity field model to get more smooth residual gravity height anomaly, then compared the influence of different covariance function to the fitting result of least square collocation which take surveying error and model error into account. The results show that Gaussian and resemble Gaussian function can achieve higher fitting precision to the large area with height anomaly value changes significance; the Remove-restore method can effectively improve the fitting precision to least square collocation method which depend on the covariance value of each points.

In this paper, a new zenith tropospheric delay (ZTD) correction model is established by using 4 years of ZTD over Southern California, which not require any meteorological data and only related to the site’s elevation and the day of year. The results show that the stability and precision of the new model are better than Saastamoinen model using ECMWF date, and the precision increases along with increasing site’s elevation. The overall accuracy of the new model is about 3.86cm when used to predict zenith troposhperic delays in 2012 over Southern California area.

In this paper, the relationship between zenith tropospheric delays and the altitude of stations is analyzed using the EGNOS tropospheric correction model. The new model (EHT model) is proposed for estimating zenith tropospheric delays from regional CORS data without meteorological data. The proposed model is compared with the direct interpolation method and the remove-restore method using data from Guangxi CORS. The results show that the new models significantly improve the calculated precision.

Using Chebyshev polynomial to fit precise ephemeris of GPS, the nodes selection has a certain influence on the precision. In this paper we use 3 kinds of precise ephemeris ( IGF, IGR, IGU ) to analyze the difference precision of randomly selected interpolation node and Chebyshev points fitting orbit and compare the difference and precision of fitting orbit by 3 kinds of ephemeris and orbit provided by IGS. The result shows that using Chebyshev points to fit precise ephemeris, the precision of IGF and IGR can achieve mm levels, the precision of IGU can achieve cm levels.