Applied Research Of GPS Leveling In Refinement Of Belt District Quasi-geoid

GPS measurements provide the three-dimensional coordinates, in other words, it can accurately determine plane location and ellipsoidal height of ground observation stations, but we use normal height in practical. How we directly use the GPS ellipsoidal height to obtain normal height in order to save manpower, resources and time, has been a focus of concern and research in surveying industry.This paper describes the basic theory of height fit and the refined principles and methods of regional quasi-geoid; Combining actual work needs, it discusses the principles and methods of GPS leveling; Combining specific examples of belt district, it place emphasis on comparison fitting precision among RBF, two surface function and three surface function, and through horizontal and vertical contrast of fitting result among different data, fitting model and fitting way, it get some useful conclusions.Common methods, such as two surface function and three surface function, inevitably exist model error. This paper introduced the RBF neural network model for height fitting to overcome this drawback. The RBFNN has excellent performance, such as adaptive structure identification and the output unrelated to the initial value and by experiment comparing, its fitting precision is better tan two, three surface function. In the height fitting of band region, if the terrain is more gentle, fitting selection point interval is less than 5 km, parameters are selected properly, the fitting result of RBFNN can meets IV grade leveling requirement, or even III grade leveling requirement.