The GPS Height Fitting Based On BP Neural Network And Its Application In Hangzhou Gulf Bridge

In order to make full use of the superiority of GPS that can provide the high precision three-dimensional coordinates, particularly the application of GPS height in engineering survey , we have to convert GPS height to normal one. At present , it always is an exploratory problem in the application fields of GPS survey that how to raise the actual precision of converting GPS geodetic height to normal one. Aiming at this problem, the following parts of GPS height fitting is discussed.Firstly, the advantage of GPS in the applied realm of surveying and mapping and its applied status in some large project items of our country are introduced, and basing on which the necessity of GPS height conversion is pointed out. Secondly, the thesis introduces some related items of GPS height conversion , such as the geodetic height system, the normal height system, the measuring principle of GPS leveling etc. Then the present condition of GPS height conversion is analyzed. Theoretically the best way of converting GPS geodetic height to normal one is to make full use of the GPS observation data , the gravity measurement data and the earth gravity field models. But to general engineering departments, they don' t have the ability of acquiring the necessary gravity data. Therefore, the method of fitting is still their first choice to convert GPS height currently. According to the difference of mathematical models, there are many methods of GPS height conversion, such as the weighted average method, the curve fitting method, the surface fitting method, the moving surface(or curve) fitting method, the "remove-restore" method and the neural network method etc.However, these normal fitting methods, especially the quadratic polynomial surface fitting method which is often used in many projects and by some GPS' processing softwares, make a certain assumption to the quasi-geoid(or geoid) and imply model error. On the contrary the method of GPS height conversion based on neural network is a method of self-adaptive mapping. It makes no assumption, can avert from the influence made by unknown factors, can reduce model error, theoretically is more reasonable and should raise the accuracy of GPS height conversion. For this reason, the thesis is detailed to learn the BP algorithm of neural network and makes many items of experiment and research to BP network model, such as the design to input and output layers, nodes selection in the single and implicit layer, the influence made by different mean square errors of learning etc. Finally the thesis summarizes a structure of BP network modelfor GPS height fitting, and carries on the instance calculation taking Hangzhou Gulf Bridge as an example. Compared with the quadratic polynomial surface fitting method, conversion of GPS height by BP network method attains an anticipated result.